The maturity-wall thesis

Verdict map

Round-2 text (superseded): "…a firm-level distributional statistic — the share of outstanding debt at risk…"

Replacement: The May 2024 Federal Reserve FEDS Note "Stress Testing the Corporate Debt Servicing Capacity: A Scenario Analysis" publishes no issuer-level median rollover coupon differential. Its exhibits are six figures — historical aggregate interest-coverage ratio, projected aggregate interest-coverage ratio, aggregate interest expense and earnings, the same two for the non-investment-grade subset, and "Percent of Debt at Risk of U.S. Public Nonfinancial Firms" — plus one appendix table of scenario inputs. The debt-at-risk series is a single threshold aggregate: the total volume of debt held by firms with an interest-coverage ratio below the threshold, divided by projected total outstanding debt in each quarter. It is built from firm-level interest-coverage forecasts but reports no percentile, no moment and no dispersion of any firm-level quantity, and the note publishes no distribution of a firm-level quantity anywhere. Aggregate interest expense and aggregate earnings are projected 2023Q4-2026Q3. The note does compute a rollover yield differential as a methodological input, comparing within each rating category the yield projected to prevail at the time of rollover with the yield at the time the maturing bonds were issued. Only the negative fact survives, and it is a fact about this one document: it establishes nothing about the reported form of corporate debt-servicing analysis generally.

Basis: One narrowing lands, on retrieved evidence and conceded in the record's own defence: the round-2 word "distributional" over-reads a threshold count, since Figure 6 reports one aggregate share obtained by summing debt across firms below an interest-coverage ratio of 2 and no distributional exhibit exists in the note. That word is struck and the construction of the series is written out in its place. Everything else was confirmed again this round by two further independent passes: the figure inventory is exhaustive as enumerated, the rollover-methodology sentence is verbatim, and the 2023Q4-2026Q3 projection window prints exactly in the figure notes — the apparent one-quarter gap against the methodology's "from 2000:Q1 until 2023:Q3 … through 2026:Q3" is the historical/projection boundary, not an error. The falsifier is not met: no issuer-level median rollover coupon differential is reported as a headline statistic, and the within-rating-category rollover comparison is a methodology step. This record is now a checkable negative about one document that the document backs.

Attack · round 1 · SELECTION

The note's own headline conclusion is absent from the ledger, which mines the document only for its reporting format: "The debt-servicing capacity of the U.S. public corporate sector as a whole is robust to sustained elevated interest rates, both in the soft landing (baseline) scenario as well as in a stagflation scenario with a moderate economic downturn"

Attack · round 1 · INFERENCE

Weaker claim supported — "this single May 2024 FEDS Note projects aggregate interest-coverage ratios and does not publish an issuer-level median rollover coupon differential." One note's choice of statistic does not establish that consolidated aggregates are the reported form for corporate debt-servicing analysis, which is the load this LOAD-BEARING record carries

Attack · round 2 · PROVENANCE

the note is misread — it does NOT report only consolidated aggregates "across firm groups". Its headline risk output is a firm-level distributional statistic, the share of debt held by firms with ICR below 2 ("debt-at-risk"), built from firm-level ICR projections; and it does perform a rollover coupon differential, stating that "we compare, within each rating category, the yield projected to prevail at the time of rollover with the yield at the time the maturing bonds were issued". The claim's characterisation of the note's reported form is false at both ends; only the narrow "no ISSUER-LEVEL MEDIAN rollover differential" survives.

Attack · round 3 · INFERENCE

The note's "share of debt at risk" is a single threshold-count aggregate — Figure 6, "Percent of Debt at Risk of U.S. Public Nonfinancial Firms" — reporting no percentile, no moment and no dispersion, so calling it "a firm-level distributional statistic" over-reads it. Weaker supportable claim: the note reports one aggregate share obtained by summing debt across firms whose ICR falls below 2, and publishes no distribution of any firm-level quantity anywhere.

Null challenge · round 1 · NOT CONTESTED

Verified: the FEDS note projects aggregate ICR and "Aggregate Interest Expense and Earnings" across firm groups; it reports no issuer-level median rollover coupon differential. The negative claim holds.

Null challenge · round 2 · NOT CONTESTED

the negative fact is confirmed at the source. The May 2024 FEDS Note reports aggregate interest coverage ratios for all public nonfinancial firms and the non-IG subset, aggregate interest expense and aggregate earnings, and share of debt at risk, projected 2023Q4-2026Q3; it projects interest payments from aggregate maturity-wall data by rating and instrument type and reports no** issuer-level median rollover coupon differential. The round-1 scope hedge ("a fact about that one document") is exactly the right size for what the document supports. Nothing to contest.

Null challenge · round 3 · NOT CONTESTED

the claim is a negative about one document plus a description of that document's outputs, and both check at source.* No issuer- or firm-level median rollover coupon differential is reported. The reported outputs are aggregate ICRs for all public nonfinancial firms and for the non-IG subset, aggregate interest expense and aggregate earnings, and the share-of-debt-at-risk at ICR below 2 as the only firm-level distributional statistic. The methodological sentence is verbatim as quoted: "we compare, within each rating category, the yield projected to prevail at the time of rollover with the yield at the time the maturing bonds were issued."* I specifically probed the one detail that looked like unsupported narrowing — the 2023Q4–2026Q3 window against the methodology's "from 2000:Q1 until 2023:Q3 ... through 2026:Q3" — and the figure notes print "2023 Q4 – 2026 Q3" exactly, so the record has the projection window right and the apparent one-quarter discrepancy is the historical/projection boundary, not an error. The final sentence already refuses the generalisation that would have been contestable.

Advocate · round 1 · REBUILT

The May 2024 FEDS Note "Stress Testing the Corporate Debt Servicing Capacity" reports aggregate interest-coverage ratios and aggregate interest expense and earnings across firm groups, and publishes no issuer-level median rollover coupon differential — verified. Two narrowings the record must accept: its own headline finding, absent from the ledger, is that "aggregate corporate ICRs are likely to remain resilient to sustained elevated borrowing costs as long as the economy avoids a severe downturn with a sharp decline in corporate earnings"; and one note's choice of statistic cannot establish that consolidated aggregates are the reported form for corporate debt-servicing analysis generally. As a LOAD-BEARING record it carries only the negative fact about this document.

Advocate · round 2 · REBUILT

The May 2024 FEDS Note "Stress Testing the Corporate Debt Servicing Capacity" reports both aggregate quantities (aggregate interest coverage ratios for all public nonfinancial firms and the non-IG subset, aggregate interest expense, aggregate earnings, projected 2023Q4-2026Q3) AND a firm-level distributional output — "the percent of outstanding debt-at risk (ICR<2) would expand in all three Moody's scenarios" — and it does compute a rollover yield differential, stating: "To compute the change in interest expense from bonds that are rolled over, we compare, within each rating category, the yield projected to prevail at the time of rollover with the yield at the time the maturing bonds were issued." What the note does not publish is any ISSUER-LEVEL MEDIAN rollover coupon differential. Only that negative fact survives; it is a fact about this one document and establishes nothing about the reported form of corporate debt-servicing analysis generally. [I retrieved the note: the Inquisitor is right at both ends, and the round-1 characterisation of the note's reported form was false. The negative fact the record actually needs is intact.]

Advocate · round 3 · REBUILT

The May 2024 FEDS Note "Stress Testing the Corporate Debt Servicing Capacity: A Scenario Analysis" publishes no issuer-level median rollover coupon differential. Its exhibits are six figures — historical aggregate ICR (Fig 1), aggregate ICR (Fig 2), aggregate interest expense and earnings (Fig 3), the same two for the non-investment-grade subset (Figs 4-5), and "Percent of Debt at Risk of U.S. Public Nonfinancial Firms" (Fig 6) — plus one appendix table of Moody's scenario inputs. The word "distributional" is withdrawn: the debt-at-risk series is a single threshold aggregate, "the total volume of debt held by firms with an ICR below the threshold divided by the projected volume of total outstanding debt in each quarter," built from firm-level ICR forecasts but reporting no percentile, moment or dispersion of any firm-level quantity, and the note publishes no distribution of a firm-level quantity anywhere. It does compute a rollover yield differential as a methodological input: "we compare, within each rating category, the yield projected to prevail at the time of rollover with the yield at the time the maturing bonds were issued." Only the negative fact survives and it is a fact about this one document; it establishes nothing about the reported form of corporate debt-servicing analysis generally. [Verified at source 2026-08-23: figure inventory, debt-at-risk construction, methodology sentence.]

Verdict map

Round-2 text (superseded): "…A floor on the non-credit component is measurable from the curve source cited on this record: the Government of Canada benchmark curve at July 9, 2026 — 2y 2.83%, 5y 3.16%, 10y 3.56%, 30y 3.95% — puts 73 basis points of pure risk-free slope between two and ten years… and more against retired paper of 0.2-year duration."

Replacement: **Comparing 2026 ten-year new-issue coupons against the coupons of retired 2026 maturities confounds three distinct things — original tenor at issuance, the vintage of the risk-free curve at each bond's own issue date, and credit — and no record in the run decomposes the differential into them. The July 2026 Government of Canada curve is steep across the relevant span, sourced primarily: the Bank of Canada benchmark series at July 9, 2026 prints 2y 2.80%, 5y 3.12%, 10y 3.52% and long 3.93%, and one-month and six-month Treasury bills printed 2.25% and 2.35% at July 8, 2026, so 72 basis points separate two and ten years and roughly 117 to 127 basis points separate the front end from ten years, with no credit component of any kind. That is a fact about one day's curve, and it is a floor on the tenor component of a comparison between two contemporaneous yields only. It is not a floor on the non-credit component of a new-issue-coupon-versus-legacy-coupon gap: the retired bond's coupon was fixed at its own issue date under a different curve, which is a vintage effect and not curve slope, and nothing on this record distinguishes "72 basis points of the gap is slope" from "none of it is slope and all of it is rate level."**

Basis: The record's core proposition survives and is strengthened — the comparison confounds tenor, vintage, rate level and credit, and nothing in the run decomposes it. Its quantitative contribution does not. Three landings, all on retrieved evidence and all conceded in the record's own defence. (i) The round-2 "floor on the non-credit component" equivocates on tenor: today's two-year point is the rate on paper with the retired bond's remaining life, not the rate the retired bond was priced at, so a contemporaneous curve slope cannot bound the non-credit part of a gap one side of which was priced years earlier. The floor is withdrawn to the contemporaneous-yield form, which is all the curve supports. The run states the correct three-way decomposition at C-35 and C-48 and had collapsed vintage into tenor here. (ii) The "and more against retired paper of 0.2-year duration" extension is not measurable from the curve source cited, whose shortest published point is the two-year; the front-end anchor the extension needs is the Treasury-bill series, which the record never cited and which is now carried. (iii) The curve values themselves are re-sourced. The previously quoted 2.83/3.16/3.56/3.95 come from a rolling, undated monthly PDF path that is overwritten each month — the same non-persistence defect the run uses to demote the aggregator at C-40 — and they sit three to four basis points above the Bank of Canada's own benchmark series for the stated date. The record's source line was also dated 2026-08-22 while the data it carries is as of 2026-07-09. Falsifier defect recorded: as printed, the falsifier quantifies over "CAD non-financial 2026 issuance" as a universe that no source on this record enumerates, and cannot be run. Restated over the issuance this run actually records, it is not met: Finning printed five-year paper against a note with four days to run, Capital Power 7.2-year paper against a note with under three months to run, and Bell, 407 International, Waste Connections and Enbridge printed ten- and thirty-year paper against sub-year maturities. Consequence recorded in §1: this record now asserts a confound and no quantity.

Attack · round 1 · INFERENCE

Weaker claim supported — "on 9 July 2026 the Government of Canada curve was upward sloping by 79bp between five and thirty years." The comparison this record indicts spans 0.2-year duration (RQO) to 10.5 years (Bell); the 5s30s segment does not measure that span. The ledger's own 31 July 2026 sources do measure it and the record never cites them: RQO weighted average YTM 2.4% against XCB weighted average YTM 4.25%, a 185bp gap

Attack · round 2 · PROVENANCE

the load-bearing "one issuer's same-day three-to-ten-year coupon slope of 125 basis points with credit held constant" is in NEITHER cited source. 407 International's April 1, 2026 book is C$500M 4.48% due Oct 7 2036 and C$500M 5.06% due Apr 7 2056 — a 10s/30s pair, 58 bp, no three-year tranche. Bell's May 27, 2026 book is C$900M 4.70% due Nov 15 2036, C$700M 5.30% due Jun 3 2056, US$650M 5.450% due Nov 15 2036 — again no three-year. The actual 125 bp 3s/10s pair is Canadian Natural Resources' C$550M 3.30% due Dec 8 2028 against C$550M 4.55% due Feb 8 2036 — a different issuer, and a December 4, 2025 print, outside the 2026 window the record's source line asserts.

Attack · round 2 · PROVENANCE

the 185 bp warrant rests on two documents absent from the record's one-item source list. The "2.4% at 0.2-year duration" is the July 31, 2026 RQO fund profile, which shared-sources.txt assigns to C-22 — so C-04's warrant is not independent of C-22, it IS C-22's document. The "broad corporate index yielding 4.25% at 8.03-year average maturity" has no source anywhere on the record. And the one document actually cited, the NBC Monthly Fixed Income Monitor, contains no corporate credit spreads and no corporate bond yields at all — it carries only the GoC curve (2y 2.83%, 5y 3.16%, 10y 3.56%, 30y 3.95% as of July 9, 2026) — so it cannot "hold credit constant" for anything.

Attack · round 3 · SELECTION

The claim says the non-credit floor is "measurable from the curve source cited on this record ... and more against retired paper of 0.2-year duration," but the cited National Bank monitor's shortest published point is the 2-year (2.83%); it publishes nothing inside two years, so the 0.2-year extension is not measurable from it. The missing evidence is the front end itself: Bank of Canada Treasury-bill yields printed 2.25% and 2.35% at July 8, 2026, some 48-58 bp below the 2-year — the anchor the claim needs and never cites.

Attack · round 3 · PROVENANCE

The curve data is sourced to a rolling, undated path — nbc.ca/.../mensuel/monthly-fixed-income-monitor.pdf — which is overwritten each month and today serves the July edition; it is a non-persistent URL for a dated snapshot, the precise defect this ledger uses in C-40 to demote the Trading Economics quote page to non-primary. The same record's source line is dated 2026-08-22 while the data it carries is as of 2026-07-09.

Null challenge · round 1 · NOT CONTESTED

I tested its falsifier and it fails decisively. CAD issuance in this window was not tenor-matched: Bell 10y, 407 10y and 30y, Capital Power 7y, Finning 5y, Canadian Natural 3y/5y/10y, Waste Connections 7y and 10y — against retired paper with weeks to months remaining. Median new-issue term is nowhere within two years of the retired paper. C-01's same-day 125 bp intra-issuer slope confirms the mechanism. This is an E2 inference, but unlike its siblings its warrant survived testing.

Null challenge · round 2 · CONTESTED

**C-04 — the warrant's two load-bearing numbers are absent from the record's own two sources,

and the decomposition it asserts is not one its evidence can perform.**

Ill-posedness plus live rival, both grounded in retrieval.

1. The 125 bp figure has no home. The claim's warrant is "one issuer's same-day

three-to-ten-year coupon slope of 125 basis points with credit held constant." The record

names exactly two issuance sources: 407 International (April 1, 2026) and Bell (May 27, 2026).

Neither printed a three-year tranche. 407's April 1 book is two tranches — C$500m 4.48% due

October 7, 2036 and C$500m 5.06% due April 7, 2056. Bell's May 27 CAD book is C$900m 4.70%

due November 15, 2036 and C$700m 5.30% due June 3, 2056. There is no 3y leg in either. Every

same-day CAD multi-tranche print I could locate in the window gives a far flatter slope than

125 bp: Enbridge (closed February 26, 2026) 3.57% due 2031 against 4.35% due 2036 = **78 bp

across 5y→10y**; Waste Connections (July 27, 2026) 4.200% due 2033 against 4.550% due 2036 =

35 bp; 407 10y→30y = 58 bp; Bell 10.5y→30y = 60 bp. A 3y→10y slope of 125 bp is not

merely unsourced — it is steeper than any adjacent segment the run's own window contains.

2. The 4.25% / 8.03-year "broad corporate index" is not in either cited source. The source

line is the National Bank fixed income monitor plus two issuance releases. The 185 bp arithmetic

(4.25 − 2.4) is correct; the minuend is unattributed. Only the 2.4% / 0.2-year leg verifies

(RQO profile, July 31, 2026).

3. Named rival: the term structure of credit spreads. "Credit held constant" holds the

issuer's credit quality constant. It does not hold the issuer's spread constant, because

investment-grade credit curves themselves slope upward with tenor. A same-issuer 3y-to-10y

coupon gap is the sum of the risk-free curve slope and the credit-curve slope. So the record's

warrant cannot deliver its conclusion — "measures term premium and the level of the risk-free

curve, not credit repricing" — because the very instrument it uses to isolate term premium

has credit-curve slope baked into it. Nothing in the ledger separates the two.

4. Internal contradiction. C-22, load-bearing in the same round, states outright that "the

differential is not decomposed into tenor, rate-level and credit components by any record in

the run." C-04 asserts precisely that decomposition. Both cannot stand.

Null challenge · round 3 · CONTESTED

CONTESTED — C-04 — the floor measures a different object than the one it bounds; "tenor" equivocates.

The record derives 73bp from the 2y–10y slope of the July 9, 2026 Government of Canada curve

(2.83 / 3.16 / 3.56 / 3.95 — I verified this table in the NBC Monthly Fixed Income Monitor) and

calls it a floor on the non-credit component of the gap between a 2026 ten-year new-issue coupon

and a retired 2026 maturity's coupon. But a retired bond's coupon was fixed at its issue date

at its original tenor. Today's two-year point is the rate on paper with the retired bond's

remaining life; it is not the rate the retired bond was priced at. "Tenor" is doing two jobs

in one sentence — original term at issuance, and remaining term to maturity — and the floor is

computed on the second while the differential is generated by the first.

Named rival: the entire gap is a vintage / rate-level effect with zero tenor content. Two

ten-year bonds, tenor-matched, one issued 2021 and one issued 2026, would show most of the same

differential because the ten-year GoC point moved from roughly 1.5% to 3.56%. Nothing on this

record distinguishes "73bp of the gap is curve slope" from "0bp is slope and all of it is rate level".

The record itself states the correct decomposition on C-35 and C-48 ("confounds tenor, vintage and

the level of the risk-free curve") and then collapses vintage into tenor here.

Second defect: the falsifier is stated over "CAD non-financial 2026 issuance" as a universe, and

the record's own cited source cannot enumerate it — I checked, and the NBC monitor contains no

table of Canadian corporate new-issue tenors or issuance by term. Two issuers (407, Bell) are not

the universe. The falsifier cannot be run against the evidence on the record.

Advocate · round 1 · REBUILT

Comparing 2026 ten-year new-issue coupons against short-dated retired 2026 maturities measures term premium and the level of the risk-free curve, not credit repricing. The record's warrant is mis-chosen and must be replaced: the 5s30s Government of Canada segment does not span 0.2 years to 10.5 years. The correct measurements are (i) inside the ledger's own July 31, 2026 sources — RQO yields 2.4% at 0.2-year duration against XCB's 4.25% at 8.03-year average maturity, a 185 bp tenor gap on one date — and (ii) Canadian Natural's 125 bp same-day 3-to-10-year intra-issuer slope, which holds credit exactly constant. Re-warranted this way the inference is stronger than as filed, and I tested its falsifier: 2026 CAD issuance was not tenor-matched (Bell 10y and 30y, 407 10y and 30y, Enbridge 5y/10y/30y, Waste Connections 7y and 10y, Capital Power 7y, Finning 5y, Canadian Natural 3y/5y/10y) against retired paper with weeks to months to run.

Advocate · round 2 · REBUILT

Comparing 2026 ten-year new-issue coupons against short-dated retired 2026 maturities confounds tenor, the level of the risk-free curve, and credit, and a substantial part of the gap is demonstrably non-credit: the Government of Canada curve carried in the record's own National Bank source (July 9, 2026 — 2y 2.83%, 5y 3.16%, 10y 3.56%, 30y 3.95%) delivers 73 basis points of pure risk-free slope from two to ten years with no credit component of any kind, and more against a 0.2-year point. That is a FLOOR on the non-credit component, not a decomposition. Three round-1 elements are withdrawn: (i) the "one issuer's same-day three-to-ten-year coupon slope of 125 basis points" is real but belongs to Canadian Natural Resources' December 4, 2025 book (C$550m 3.30% due Dec 8 2028 against C$550m 4.55% due Feb 8 2036 = exactly 125 bp) — a different issuer, outside the 2026 window, and cited on no source on this record; 407 International's April 1, 2026 book is 10s/30s only (4.48% 2036 / 5.06% 2056) and Bell's May 27 book has no three-year leg; (ii) even correctly attributed, a same-issuer 3s/10s coupon gap holds credit QUALITY constant but not credit SPREAD, because investment-grade credit curves slope upward with tenor, so 125 bp is risk-free slope plus credit-curve slope and isolates neither; (iii) the "broad corporate index yielding 4.25% at 8.03-year average maturity" appears in neither cited source, and the National Bank monitor contains no corporate yields or spreads at all, so the 185 bp warrant is withdrawn entirely. With the claim reduced to a floor rather than a decomposition, the Null's internal-contradiction point against C-22 dissolves: no record in the run performs the decomposition, and this one no longer asserts it.

Advocate · round 3 · REBUILT

Comparing 2026 ten-year new-issue coupons against the coupons of retired 2026 maturities confounds three distinct things — original tenor at issuance, the vintage of the risk-free curve at each bond's own issue date, and credit — and no record in the run decomposes the differential into them. The July 2026 Government of Canada curve is steep across the relevant span, and it is now sourced primarily: Bank of Canada benchmark series at 2026-07-09 print 2y 2.80, 5y 3.12, 10y 3.52, long 3.93 percent, and 1-month and 6-month Treasury bills printed 2.25 and 2.35 percent at 2026-07-08, so 72 basis points separate two and ten years and roughly 117-127 basis points separate the front end from ten years, with no credit component of any kind. That is a fact about one day's curve and a floor on the tenor component of a contemporaneous yield comparison only. WITHDRAWN: its status as a floor on the non-credit component of a new-issue-coupon-versus-legacy-coupon gap. The retired bond's coupon was fixed at its own issue date under a different curve; that is a vintage effect, not curve slope, and nothing on this record distinguishes "73bp of the gap is slope" from "none of it is slope and all of it is rate level" — the decomposition the run states correctly on C-35 and C-48 and collapsed here. Also withdrawn: the previously quoted curve values (2.83/3.16/3.56/3.95) come from a rolling, undated monthly PDF path — the same non-persistence defect the run uses to demote the aggregator in C-40 — and do not reproduce at the Bank of Canada's own series for the stated date. The printed falsifier quantifies over "CAD non-financial 2026 issuance" as a universe that no source on this record enumerates, and cannot be run; it must be restated over the named issuers or struck. [BoC Valet retrieved 2026-08-23.]

Verdict map

Round-2 text (superseded): "…That line is a hand-picked secondary-market illustration, and the outlook performs no issuer-level pricing or valuation exercise on it."

Replacement: In its January 13, 2026 North American Fixed Income outlook, BMO Global Asset Management names one BBB-rated Canadian energy-infrastructure bond — Inter Pipeline's 2032 — introduced verbatim as "Let's review one compelling example in Inter Pipeline Ltd.'s 2032 bond. It is a BBB-rated energy-infrastructure credit offering a 4.3% yield and 125 bps spread over Canadian government bonds," in a passage about oil-price tail risk. The outlook publishes no cash-flow model, fair-value estimate or valuation output for that bond: a quoted yield and a quoted spread are the whole of the issuer-level content, and no comparison of either to a computed fair value appears anywhere. Whether unpublished valuation work stands behind the selection is not knowable from the document and is not asserted; absence of shown work is not claimed here as evidence of absent work. The document is not free of quantitative work: it publishes market-level 2026 forecast ranges against current marks — Canada ten-year 2.80-3.60% (current 3.25%), US ten-year 3.90-4.80% (current 4.10%), investment-grade spreads 45-125 bp (current 82 bp), high-yield spreads 290-495 bp (current 345 bp), investment-grade CDX 46-80 bp (current 54 bp), and a 5-6% corporate bond return forecast — and the 125 bp quoted for Inter Pipeline sits at the upper bound of the outlook's own investment-grade spread range, a juxtaposition the document itself never draws.

Basis: Everything checkable on this record checks and checked again this round: all six forecast ranges and current marks verify verbatim, the Inter Pipeline sentence verifies verbatim, and both limbs of the falsifier are unmet — the bond is quoted at 4.3 percent, below 5.5, and a Canadian BBB yield below 5 percent is named. What failed was the load-bearing predicate. "Performs no issuer-level pricing or valuation exercise" stated no criterion for what counts as an exercise, and independent readings of the identical text returned opposite verdicts on it — one reading returning that the piece does include specific pricing details for the bond, which on an undefined term is not a contradiction but a symptom. Two objections converge and both are accepted: the quoted line does give a yield and a spread over a named benchmark, which is issuer-level pricing data; and a named rival — that the quoted line is the published output of an unpublished valuation view, as house outlooks routinely leave it — cannot be excluded from the document. The replacement substitutes a definite criterion for the undefined predicate and explicitly disclaims the inference the rival attacked, which removes the disagreement rather than deciding it.

Attack · round 1 · INFERENCE

Weaker claim supported — "in a 13 January 2026 outlook piece BMO cited one BBB Canadian midstream bond, Inter Pipeline 2032, as 'a compelling example' at a 4.3% yield and 125bp spread over Canadian government bonds." That is a hand-picked secondary-market illustration inside a promotional outlook; BMO performs no pricing exercise, and the record's "priced a BBB rated Canadian energy infrastructure bond" imputes a valuation the source does not carry out

Attack · round 2 · PROVENANCE

"no pricing exercise is performed" is a misread of the document. The same outlook publishes index-level 2026 forecast ranges with current marks: Canada 10-year yield 2.80–3.60% (current 3.25%), US 10-year 3.90–4.80% (current 4.10%), investment-grade spreads 45–125 bp (current 82 bp), high-yield spreads 290–495 bp (current 345 bp), IG CDX 46–80 bp (current 54 bp), and a 5–6% corporate bond return forecast. The Inter Pipeline line is an illustration attached to a quantitative market-level view, not the outlook's only content.

Attack · round 3 · INFERENCE

"The outlook performs no issuer-level pricing or valuation exercise on it" is stronger than the document supports. The quoted line gives the bond's yield (4.3%) and its spread (125 bp over Canadas), and the same outlook publishes a 2026 IG spread range of 45-125 bp against a current 82 bp — so the document does place the named credit at the extreme wide edge of its own forecast band, which is a relative-value comparison. Weaker supportable claim: the outlook publishes no cash-flow, model or fair-value output for the Inter Pipeline 2032, only a quoted yield and a quoted spread.

Null challenge · round 1 · NOT CONTESTED

Verified verbatim at BMO: "Inter Pipeline Ltd.'s 2032 bond… a BBB-rated energy-infrastructure credit offering a 4.3% yield and 125 bps spread over Canadian government bonds."

Null challenge · round 2 · NOT CONTESTED

the quote, the bond, and the hedge all check.** The BMO outlook names Inter Pipeline's 2032 as "a BBB-rated energy-infrastructure credit offering a 4.3% yield and 125 bps spread over Canadian government bonds," and it is a single illustrative case inside a promotional outlook with no pricing or valuation exercise performed. The record's characterization is the source's own posture.

Null challenge · round 3 · CONTESTED

CONTESTED — C-13 — the load-bearing predicate is undefined, and independent reading of the same text returns the opposite verdict.

Everything checkable on this record checks. All six forecast ranges and current marks verify

verbatim (Canada 10y 2.80–3.60 vs 3.25; US 10y 3.90–4.80 vs 4.10; IG 45–125 vs 82; HY 290–495 vs

345; IG CDX 46–80 vs 54; 5–6% return), the Inter Pipeline sentence verifies verbatim, and the

falsifier is unmet in both limbs (the 4.3% quote is below 5%, and it is the only Canadian BBB yield

named). What does not check is the one predicate the conclusion hangs on: **"performs no

issuer-level pricing or valuation exercise."** No criterion is stated for what counts as an

"exercise". The document names an issuer, a specific bond, a yield and a spread over a named

benchmark — that is issuer-level pricing data. My own read of the same page returned "Yes, the

piece includes specific pricing details for Inter Pipeline Ltd.'s 2032 bond," the opposite of the

record's verdict, on identical text. That divergence is what an undefined term produces.

Named rival: the 4.3% / 125bp line is the published output of a valuation view — a BBB

energy-infrastructure credit selected because it was judged to offer value — with the work

unpublished, as house outlooks routinely leave it. Absence of shown work is not evidence of absent

work, and nothing in the document discriminates between the two.

Advocate · round 1 · REBUILT

In its January 13, 2026 North American Fixed Income outlook BMO cited one BBB-rated Canadian energy-infrastructure bond — Inter Pipeline's 2032 — as a compelling example at a 4.3% yield and a 125 bp spread over Canadian government bonds. The record's "priced" imputes a valuation exercise the source does not perform: this is a hand-picked secondary-market illustration inside a promotional outlook, and it supports only that a BBB Canadian midstream 2032 bond was quoted at that level on that date.

Advocate · round 2 · REBUILT

In its January 13, 2026 North American Fixed Income outlook, BMO Global Asset Management cited one BBB-rated Canadian energy-infrastructure bond, Inter Pipeline's 2032, as "a BBB-rated energy-infrastructure credit offering a 4.3% yield and 125 bps spread over Canadian government bonds." That line is a hand-picked secondary-market illustration and the outlook performs no issuer-level pricing, valuation or fair-value exercise on it. The round-1 phrase "no pricing exercise is performed" is withdrawn as stated: the same outlook publishes market-level 2026 forecast ranges against current marks — Canada 10-year 2.80-3.60% (current 3.25%), US 10-year 3.90-4.80% (current 4.10%), investment-grade spreads 45-125 bp (current 82 bp), high-yield spreads 290-495 bp (current 345 bp), IG CDX 46-80 bp (current 54 bp), and a 5-6% corporate bond return forecast. The correct statement is that no ISSUER-level pricing work supports the Inter Pipeline quote, not that the document contains no quantitative work. [Retrieved; every forecast range the Inquisitor cited checks out exactly, and so does the absence of any valuation of the Inter Pipeline bond.]

Advocate · round 3 · REBUILT

In its January 13, 2026 North American Fixed Income outlook, BMO Global Asset Management names one BBB-rated Canadian energy-infrastructure bond — Inter Pipeline's 2032 — introduced verbatim as "Let's review one compelling example in Inter Pipeline Ltd.'s 2032 bond. It is a BBB-rated energy-infrastructure credit offering a 4.3% yield and 125 bps spread over Canadian government bonds," in a passage about oil-price tail risk. The undefined predicate is replaced by a criterion: the outlook publishes no cash-flow model, fair-value estimate or valuation output for that bond — a quoted yield and a quoted spread are the whole of the issuer-level content, and no comparison of either to a computed fair value appears anywhere. Whether unpublished valuation work stands behind the selection is not knowable from the document and is not asserted; absence of shown work is not claimed as evidence of absent work. The document is not free of quantitative work: Canada 10y 2.80-3.60 (current 3.25), US 10y 3.90-4.80 (4.10), IG spreads 45-125 bp (82), HY 290-495 bp (345), IG CDX 46-80 bp (54), and a 5-6% corporate bond return forecast — and the 125 bp quoted for Inter Pipeline sits at the upper bound of the outlook's own IG spread range, a juxtaposition the document itself never draws. [Retrieved 2026-08-23; the document contains no relative-value or fair-value output for the bond.]

Verdict map

Round-2 text (superseded): "…The ordering variable is intraday decision rank, defined as the number of forecasts an analyst has already issued before the focal forecast on the same day…"

Replacement: In Hirshleifer, Levi, Lourie and Teoh (Journal of Financial Economics 133), the share of sell-side analyst earnings forecasts ending in zero or five runs 29.5, 29.8, 29.7, 31.0 and 28.3 percent across a within-day ordering — first, second, third, fourth and fifth-or-later forecast issued by that analyst that day. The paper's regression variable is not that raw count: Decision rank is defined verbatim as "the log value of the number of forecasts an analyst has made before the forecast being evaluated, plus one" — a log(n+1) transform. The final bin aggregates all fifth-and-later forecasts, so the series is five bins, not five ordered points, and the bins span 28.3 to 31.0 percent with no monotone trend. That flatness is unconditional and is not the paper's estimate: in the analyst-day fixed-effects specification (Table 6, column 6; N = 386,924) the coefficient on Decision rank is 0.136, significant at the 5 percent level — a positive within-analyst-day rounding gradient. The bin series must not be cited against that finding. No single full-sample pooled rounding percentage was located in the paper as retrieved, and that negative rests on reading passes rather than on any statement in the paper.

Basis: Two corrections land, both on retrieved evidence from the published paper and both conceded in the record's own defence. The round-2 definition of the ordering variable was the wrong detail — the paper's variable is a log transform, not a raw count — and it is replaced with the paper's own words. And the record carried only the unconditional bin shares while omitting the paper's actual estimate of the effect, which is the within-analyst-day coefficient the paper adopts in place of the unconditional comparison; reading the flat bins as the finding inverts the paper. A third objection is accepted as a limitation rather than a defeat: "the paper publishes no single full-sample total" is a universal negative over a full journal article plus appendices, supported only by not having found one, so the text now states it as a negative resting on reading passes. The five bin percentages, the fifth-or-later aggregation and the within-day framing all verify in every pass. Falsifier defect, now repaired. The inherited falsifier read "the paper reports a rounded-forecast share outside the 29 to 31 percent range," and the claim's own 28.3 percent figure satisfies it — a record cannot assert as verified fact the very thing its test says would refute it. That is a defect in the test, not a falsification of the claim: the text asserts no 29-to-31 range and reports the paper's values correctly. The falsifier is restated as: the paper reports a rounded-forecast share outside the 28 to 31 percent range. So restated it is not met. The paper's subject remains a finding about how analysts express their own point forecasts, a different phenomenon from where a round-number threshold sits in an outcome distribution, so it cannot support C-17 — and the round-3 finding that the paper's estimate is a positive gradient rather than flatness moves it further from C-17 still.

Attack · round 1 · INFERENCE

Weaker claim supported — "in Table 2 the share of analyst forecasts ending in zero or five rises from 29.5% for the first forecast to 31.0% for the fourth." The 29-to-31 band is a drift across forecast order within the year, not a confidence range on a single population share, and the paper's subject is decision fatigue increasing rounding as forecast order advances, not the salience of round-number thresholds

Attack · round 2 · PROVENANCE

"forecast order within the year" misreads the paper. Hirshleifer, Levi, Lourie and Teoh's ordering variable is within-DAY decision rank — "the logarithm of the number of forecasts an analyst has issued" before the focal forecast on the same day. The 29.5 / 29.8 / 29.7 / 31.0 / 28.3 series in Table 2 is a within-day sequence (forecasts 1 through ≥5 of that day), not a within-year one. The record's stated dimension of variation is the wrong dimension.

Attack · round 3 · PROVENANCE

The record misreads the paper's own variable definition. It states the ordering variable is "defined as the number of forecasts an analyst has already issued before the focal forecast on the same day"; the paper defines it as "Decision rank is the log value of the number of forecasts an analyst has made before the forecast being evaluated, plus one" — a log(n+1) transform, not a raw count. The record's added detail is the wrong detail.

Attack · round 3 · SELECTION

The ledger carries only the five unconditional bin shares and reads them as "cluster[ing] near 30 percent," omitting the paper's actual estimate of the effect: a positive coefficient of 0.136 on Decision rank, significant at 5 percent, in the analyst-day fixed-effects specification (Table 6, col. 6), over 386,924 forecasts. The record's flatness reading is exactly the unconditional comparison the paper discards in favour of a within-analyst-day one.

Null challenge · round 1 · NOT CONTESTED

Verified in Table 2: rounding rates of 29.5%, 29.8%, 29.7%, 31.0%, 28.3% across forecast sequences. The claim is about the total, which falls inside 29–31%. (One sequence-level figure sits at 28.3%, but the claim does not assert a per-sequence bound.)

Null challenge · round 2 · CONTESTED

**C-24 — the ordering variable is misdescribed, and the falsifier is satisfied by the claim's

own text.**

The five percentages verify exactly (29.5, 29.8, 29.7, 31.0, 28.3), and so does the absence of a

full-sample total. But the claim says these run "across forecast order within the year."

They do not. The paper's ordering variable is the order of a forecast **within the analyst's

day** — "each forecast issued during the day is marked as a decision by the order it was

issued." That is the entire construct of the paper: intraday decision fatigue. Recoding a

within-day sequence as a within-year sequence is not a wording slip; it destroys the only

mechanism that makes the drift interpretable, and the record's own gloss ("a drift across

forecast order") then refers to nothing identifiable.

Separately, the record is unfalsifiable as written. Its falsifier reads "the paper reports a

rounded-forecast share outside the 29 to 31 percent range." The claim itself reports 28.3

percent. A record whose claim text already trips its own falsifier has no discriminating test

attached to it.

Null challenge · round 3 · CONTESTED

CONTESTED — C-24 — the record's own text satisfies its own falsifier. It has no truth conditions left.

Falsifier as printed: "The paper reports a rounded-forecast share outside the 29 to 31 percent range."

Claim as printed: the bins are "29.5, 29.8, 29.7, 31.0 and 28.3 percent" and the series *"spann[s]

28.3 to 31.0 percent."* 28.3 is outside 29 to 31. The record asserts, as verified fact, the exact

thing its falsifier says would refute it. Two narrowings added the bin values and never revisited the

test condition the bin values contradict, and the claim's own summary sentence carries both bands at

once — "cluster near 30 percent" against a falsifier drawn at 29–31 against a stated span of 28.3–31.0.

A record that cannot be defeated by the thing it says would defeat it also cannot be confirmed by

anything; it is not a claim in the falsificationist sense the ledger is using.

The underlying numbers are fine — I verified all five bins, the "Forecast ≥ 5" aggregation, and the

same-day framing of decision rank. The defect is in the test, not the data, and it is exactly the

failure the round-3 brief anticipates: detail added to survive an objection, left unreconciled with

what it was added to.

Secondary, and a named rival: "The paper publishes no single full-sample total" is a universal

negative over a full JFE article plus appendices, supported only by not having found one. The rival

is that a pooled rounding rate sits in the descriptive statistics or an appendix table and was not

surfaced. Nothing in the ledger discriminates, and a negative of that shape is not established by a

single reading pass.

Advocate · round 1 · REBUILT

Hirshleifer, Levi, Lourie and Teoh report in Table 2 that the share of analyst earnings forecasts ending in zero or five runs 29.5%, 29.8%, 29.7%, 31.0% and 28.3% across forecast order within the year. I checked: the paper publishes no single total for the full sample, and one sequence value falls below 29%. The defensible statement is that rounded forecasts cluster near 30 percent of forecasts across forecast order. The paper's subject is decision fatigue raising the rounding rate as forecast order advances — a finding about how analysts express their own point forecasts, which is a different phenomenon from where a round-number threshold sits in an outcome distribution, and it cannot be used to support C-17.

Advocate · round 2 · REBUILT

In Hirshleifer, Levi, Lourie and Teoh (Journal of Financial Economics 133), the share of sell-side analyst earnings forecasts ending in zero or five runs 29.5, 29.8, 29.7, 31.0 and 28.3 percent across the order in which an analyst issues forecasts WITHIN A SINGLE DAY — first, second, third, fourth, and fifth-or-later of that day — spanning 28.3 to 31.0 percent. The paper's ordering variable is intraday decision rank, defined as "the logarithm of the number of forecasts an analyst has issued before the focal forecast plus one," on a sample restricted to forecasts prepared and released in sequence on one day; the round-1 phrase "across forecast order within the year" is withdrawn as a misdescription of the construct, and with it any reading of the series as a within-year drift. The paper publishes no single full-sample total. The final bin is an aggregate of all fifth-and-later forecasts, not a fifth ordinal position, so the series is five bins rather than five ordered points. [Retrieved and confirmed; both attackers are right on the dimension, and the record's stated falsifier range of 29 to 31 percent is tripped by its own 28.3 figure and needs restating as 28 to 31.]

Advocate · round 3 · REBUILT

In Hirshleifer, Levi, Lourie and Teoh (Journal of Financial Economics 133), the share of sell-side analyst earnings forecasts ending in zero or five runs 29.5, 29.8, 29.7, 31.0 and 28.3 percent across a within-day ordering — first, second, third, fourth and fifth-or-later forecast issued by that analyst that day. CORRECTED: the paper's regression variable is not that raw count. Decision rank is defined verbatim as "the log value of the number of forecasts an analyst has made before the forecast being evaluated, plus one" — a log(n+1) transform; the record's added definition was the wrong detail. The final bin aggregates all fifth-and-later forecasts, so the series is five bins, not five ordered points, and the bins span 28.3 to 31.0 percent with no monotone trend. That flatness is unconditional and is NOT the paper's estimate, which the record must now carry: in the analyst-day fixed-effects specification (Table 6, column 6; N = 386,924) the coefficient on Decision rank is 0.136, significant at 5 percent — a positive within-analyst-day rounding gradient. The bin series must not be cited against that finding. No single full-sample pooled rounding percentage is reported anywhere in the paper as retrieved, and that negative rests on reading passes, not on the paper asserting it. FALSIFIER DEFECT, flagged rather than repaired because repairing it is the Adjudicator's call: as printed ("a rounded-forecast share outside the 29 to 31 percent range") the falsifier is satisfied by the claim's own 28.3 percent bin. It must be restated to the actual reported span, 28 to 31 percent, or this record has no truth conditions. [Published PDF retrieved 2026-08-23.]

Verdict map

Round-2 text (superseded): "…That repayment is named first and is the only use carrying a specific dollar amount, so roughly half of the US$1.0 billion offering is directed to that maturity and the balance is unallocated across three generic uses."

Replacement: Nutrien's May 26, 2026 offering was denominated in US dollars in both tranches — US$500 million at 4.850% due May 29, 2031 and US$500 million at 5.350% due May 29, 2036 — with no Canadian-dollar tranche. The use-of-proceeds sentence names four uses: repayment of the US$500 million 4.000% senior notes upon their maturity on December 15, 2026, reducing outstanding indebtedness under short-term credit facilities, financing working capital, and general corporate purposes. The release states no allocation. At most US$500 million of the offering could be applied to that maturity; how much is directed to it is unstated, the maturity falls nearly seven months after issue, and short-term facility paydown is named as an interim use, so proceeds may be routed through the revolver or elsewhere entirely. A US-dollar issuer refinancing a US-dollar maturity is outside the Canadian-dollar frame on both sides, so this transaction is uninformative about the composition of Canadian-dollar 2026 refinancing rather than evidence of exclusion from it.

Basis: All tranches and the full four-part use of proceeds verify verbatim in every pass, and the falsifier is not met — there is no Canadian-dollar tranche. One correction lands and was conceded in the record's own defence: "roughly half of the offering is directed to that maturity" does not follow from the sentence quoted. Neither premise is an allocation — the US$500 million is the outstanding principal of the maturing series, a fact about the old bond and not an application of the new proceeds, and clause order in a use-of-proceeds sentence is not evidence of amount. A named rival cannot be excluded from the text: proceeds are fungible, the enumeration is standard, and the December 2026 maturity can be met from operating cash flow, from the revolver or from a later issue. The consistency point is decisive and is recorded as a standing rule for this run: the run correctly refuses this identical inference on Capital Power's permissive enumeration at C-48, and nothing distinguishes the two sentences. The record's closing conclusion never depended on the withdrawn sentence and stands unchanged.

Attack · round 1 · INFERENCE

Weaker claim supported — "Nutrien's 26 May 2026 US$1.0 billion offering was US-dollar denominated in both tranches, and repayment of the US$500M 4.000% notes maturing 15 December 2026 is one of three stated uses of proceeds, alongside reducing short-term credit facility borrowings and financing working capital and general corporate needs." The record's "refinancing of its December 2026 maturity" recasts a general two-tranche offering as a dedicated refinancing

Attack · round 2 · INFERENCE

the record is null for the population question it is filed under. A US-dollar issuer refinancing a US-dollar maturity neither enters nor is excluded from a Canadian-dollar population — it is simply outside the frame on both sides, so it evidences nothing about Canadian-dollar refinancing composition. Weaker claim actually supported: "Nutrien's May 26, 2026 offering was US-dollar in both tranches and lists repayment of its US$500 million 4.000% notes due December 15, 2026 first among four stated uses; the transaction is uninformative about the composition of Canadian-dollar 2026 refinancing." Note also that the repayment is the FIRST-named and only amount-specific use — "not a dedicated refinancing" overreads a use-of-proceeds list that names that maturity with its exact principal while the other three uses are generic.

Attack · round 3 · INFERENCE

"So roughly half of the US$1.0 billion offering is directed to that maturity" is not supported. The release states only that Nutrien "intends to use the net proceeds from this offering to repay its US$500 million aggregate principal amount of 4.000% senior notes upon their maturity on December 15, 2026, to reduce outstanding indebtedness under its short-term credit facilities, to finance working capital and for general corporate purposes" — no allocation, no ordering by amount, and a maturity nearly seven months after issue with short-term facility paydown named as an interim use. Weaker supportable claim: at most US$500 million of the offering is available for that maturity, and the release states nothing about how much is directed to it.

Null challenge · round 1 · NOT CONTESTED

Verified: both tranches are US$500m, USD only. No Canadian dollar tranche exists in that offering.

Null challenge · round 2 · NOT CONTESTED

all tranches and the full four-part use of proceeds verify verbatim.** US$500m 4.850% due May 29, 2031 and US$500m 5.350% due May 29, 2036; no Canadian-dollar tranche; proceeds to repay the US$500m 4.000% notes at their December 15, 2026 maturity, to reduce short-term credit facility borrowings, to finance working capital and for general corporate purposes. The record's added claim — that this is not a dedicated refinancing — is the enumeration itself, and the currency exclusion is dispositive on its face.

Null challenge · round 3 · CONTESTED

CONTESTED — C-28 — "roughly half is directed to that maturity" does not follow from the sentence quoted, and the ledger applies the opposite standard elsewhere.

The release's use-of-proceeds sentence verifies verbatim: *"Nutrien intends to use the net proceeds

from this offering to repay its US$500 million aggregate principal amount of 4.000% senior notes upon

their maturity on December 15, 2026, to reduce outstanding indebtedness under its short-term credit

facilities, to finance working capital and for general corporate purposes."* The record then reasons:

named first + only item carrying a dollar amount ⇒ roughly half the offering goes there. Neither

premise is an allocation. The US$500 million is the outstanding principal of the maturing series,

which is a fact about the old bond, not an application of the new proceeds. My read of the document

returned that it "does not state what portion of the US$1.0 billion offering will be allocated

specifically to this repayment versus other uses listed."

Named rival: proceeds are fungible and the enumeration is standard boilerplate. The December 2026

maturity can be met from operating cash flow, from the revolver, or from a later issue, with none,

some or all of this offering going elsewhere — an ordering of clauses in a prospectus sentence is not

evidence against any of that. Nothing in the release distinguishes the rival from "roughly half".

The ledger already knows this: C-48 refuses the identical inference on Capital Power, correctly

calling its use-of-proceeds sentence "a permissive enumeration." Nutrien's sentence is the same

species. Nothing in the run explains why the permissive reading governs there and the allocative

reading governs here.

Not contested within C-28: both tranches US-dollar (US$500M 4.850% due May 29 2031; US$500M 5.350%

due May 29 2036), no Canadian-dollar tranche, four named uses, falsifier unmet — all verified verbatim.

The closing conclusion does not need "roughly half" and survives without it.

Advocate · round 1 · REBUILT

Nutrien's May 26, 2026 offering was US-dollar denominated in both tranches — US$500 million at 4.850% due May 29, 2031 and US$500 million at 5.350% due May 29, 2036 — with no Canadian dollar tranche, so it sits outside the CAD population the headline statistic measures. The narrowing: repayment of the US$500 million 4.000% notes maturing December 15, 2026 is one of four stated uses, alongside reducing short-term credit facility borrowings, financing working capital and general corporate purposes. "Nutrien's refinancing of its December 2026 maturity" recasts a general two-tranche offering as a dedicated refinancing and should be restated as a general offering among whose uses that repayment is named.

Advocate · round 2 · REBUILT

Nutrien's May 26, 2026 offering was denominated in US dollars in both tranches — US$500 million at 4.850% due May 29, 2031 and US$500 million at 5.350% due May 29, 2036 — with no Canadian-dollar tranche. The use-of-proceeds sentence is verbatim: "Nutrien intends to use the net proceeds from this offering to repay its US$500 million aggregate principal amount of 4.000% senior notes upon their maturity on December 15, 2026, to reduce outstanding indebtedness under its short-term credit facilities, to finance working capital and for general corporate purposes." The round-1 phrase "not a dedicated refinancing" is qualified: that repayment is named FIRST and is the only use carrying a specific dollar amount, and US$500 million of a US$1.0 billion offering is explicitly earmarked to it, so the accurate statement is that roughly half the offering is directed to that maturity and the balance is unallocated across three generic uses. The record's evidentiary standing is also corrected: a US-dollar issuer refinancing a US-dollar maturity is outside the Canadian-dollar frame on both sides, so this transaction is uninformative about Canadian-dollar refinancing composition rather than evidence of exclusion from it.

Advocate · round 3 · REBUILT

Nutrien's May 26, 2026 offering was denominated in US dollars in both tranches — US$500 million at 4.850% due May 29, 2031 and US$500 million at 5.350% due May 29, 2036 — with no Canadian-dollar tranche. The use-of-proceeds sentence names four uses: repayment of the US$500 million 4.000% senior notes upon their maturity on December 15, 2026, reduction of outstanding indebtedness under short-term credit facilities, working capital, and general corporate purposes. WITHDRAWN: "roughly half of the US$1.0 billion offering is directed to that maturity." Neither premise is an allocation — the US$500 million is the outstanding principal of the maturing series, a fact about the old bond and not an application of the new proceeds, and clause order in a use-of-proceeds sentence is not evidence of amount. The run refuses this identical inference on Capital Power's permissive enumeration in C-48, and nothing distinguishes the two sentences. What survives: at most US$500 million of the offering could be applied to that maturity, the release states nothing about how much is, and the maturity falls nearly seven months after issue with short-term facility paydown named as an interim use, so proceeds may be routed through the revolver or elsewhere entirely. The closing conclusion does not depend on the withdrawn sentence and stands: a US-dollar issuer refinancing a US-dollar maturity is outside the Canadian-dollar frame on both sides, so this transaction is uninformative about the composition of Canadian-dollar 2026 refinancing rather than evidence of exclusion from it.

Verdict map

Round-2 text (superseded): "…No source on this record supports any longer-window measurement of that quantity, so the record establishes nothing about whether the bias survives the post-2021 rate surge."

Replacement: As reported by CNBC on June 26, 2017, Deutsche Bank's Torsten Slok, analysing the Survey of Professional Forecasters, found that twelve-month-ahead forecasts of the ten-year US Treasury yield were biased high: "the average 12-month forecast error since 2003 has been 60 basis points too high." That sample ends in 2017, and the chain behind the report — the bank's note and its computation — is unpublished and uncheckable. The longer-window question is answerable from a primary source this record must now carry: the Federal Reserve Bank of Philadelphia's own "Error Statistics for the Survey of Professional Forecasters," TBOND file, Table 1A, horizon H=4, reports a mean error of −0.43 over 1993Q1-2023Q1 and −0.46 over 1997Q1-2023Q1 on an actual-minus-forecast convention — four-quarters-ahead forecasts of the ten-year yield 43 to 46 basis points too high, measured through the post-2021 rate surge. Direction survives the longer window; magnitude does not, running roughly a quarter below the 60 basis points reported in 2017. The two are different estimates on different samples and horizon conventions and must not be equated or averaged.

Basis: The quoted sentence, its attribution, the survey, the quantity and the direction verify at the cited article in every pass, and the falsifier is not met — the cited analysis reports 60 basis points, biased high. What round 3 falsifies is the record's own negative: a free, primary, longer-window measurement of this quantity on this survey exists at the survey's own publisher and was absent from the record, so "no source supports any longer-window measurement" and "the record establishes nothing about the post-2021 regime" are both struck. This is the second time this record's longer-window clause has moved: round 2 withdrew an unsourced "40 basis points over 1993-2024" figure attributed to no document, and round 3 supplies the properly sourced version of the same quantity at −0.43 to −0.46. The record's source list must go from one to two. The cited 2017 item remains tertiary and is shared with C-10, whose level-form residue no longer rests on it alone.

Attack · round 1 · SELECTION

A more recent and more authoritative measurement of the same quantity is absent from the ledger: the St. Louis Fed's "Professional Forecasters' Past Performance and the 2025 Economic Outlook" (31 December 2024) reports a mean forecast error on the 10-year Treasury yield of -0.4 percentage points over 1993-2024 — 40 basis points, not 60, and measured across the post-2021 rate surge that a sample ending in 2017 cannot contain

Attack · round 2 · PROVENANCE

the second sentence has no source. The record carries exactly one source, CNBC dated 2017-06-26, whose scope I confirmed is "the average 12-month forecast error since 2003 has been 60 basis points too high" — a 2017 article cannot contain a 1993–2024 measurement, and the "40 basis points" figure is unattributed to any document on the record. The round-1 narrowing bolted an unsourced comparator onto a single-source record and thereby made it weaker, not stronger.

Attack · round 3 · SELECTION

The record asserts "No source on this record supports any longer-window measurement of that quantity" and stops there. The Philadelphia Fed publishes precisely that quantity, for precisely this survey, free: its SPF error-statistics file for TBOND reports a four-quarters-ahead mean error of −0.43 over 1993:Q1-2023:Q1 and −0.46 over 1997:Q1-2023:Q1 on an actual-minus-forecast convention — forecasts 43 to 46 bp too high, measured through the post-2021 rate surge. It is absent from the ledger and answers the question the record declares unanswerable.

Null challenge · round 1 · NOT CONTESTED

Verified against the source: 60 bp, twelve-month horizon, Survey of Professional Forecasters versus actual, from 2003, biased high, attributed to Deutsche Bank's Torsten Slok. The claim is stated narrowly as what the analysis reports and matches it. The scope problem belongs to C-10, which extrapolates it, and I have charged it there rather than here.

Null challenge · round 2 · CONTESTED

C-29 — the added long-sample figure is a different survey, so "the same quantity" is false.

The first sentence verifies: CNBC, June 26, 2017, "the average 12-month forecast error since

2003 has been 60 basis points too high." The round-1 addition does not. The 40 bp / 1993-2024

mean forecast error for the ten-year Treasury is a Blue Chip survey statistic, not a

Philadelphia Fed Survey of Professional Forecasters statistic. The record explicitly frames it

as "a longer measurement of the same quantity." It is not the same quantity: different

survey, different panel, different construction. The rhetorical work the sentence does —

showing the bias shrinks but survives when the sample is extended past the rate surge — requires

one series measured over two windows, and there is no such series here. Compounding this, the

record carries exactly one source, dated 2017, which cannot report a measurement ending in 2024.

Null challenge · round 3 · NOT CONTESTED

the record has already retreated to a claim about what CNBC reported, and that is precisely what the evidence settles.* Verbatim: "the average 12-month forecast error since 2003 has been 60 basis points too high."* Attribution (Slok, Deutsche Bank), survey (Survey of Professional Forecasters), quantity (twelve-month- ahead ten-year Treasury forecasts) and direction (high, not low) all verify; both limbs of the falsifier are unmet. The chain behind CNBC — the Deutsche Bank note, the underlying SPF computation — is unpublished and uncheckable, but the record does not assert it, and it explicitly disclaims any post-2021 extension. There is nothing left to contest that the record has not already conceded on its own.

Advocate · round 1 · REBUILT

As reported by CNBC in June 2017, Deutsche Bank's analysis of the Philadelphia Fed Survey of Professional Forecasters found that twelve-month-ahead forecasts of the ten-year Treasury yield averaged 60 basis points too high over 2003-2017, biased high. The claim is accurate to its source; the omission that matters is the sample's endpoint, which the record does not state. A better-sourced measurement of the same quantity exists and belongs in the ledger: the St. Louis Fed's December 2025 review reports a mean forecast error of -0.4 percentage points over 1993-2024 — 40 basis points — across a sample that includes the post-2021 rate surge a 2017 endpoint cannot contain. Both carry the same sign, which is the point that survives; 40 bp over 1993-2024 is the magnitude to use. Sourcing note: this is a tertiary item (CNBC reporting Deutsche Bank's calculation on public Philadelphia Fed BOND10 data) shared with C-10.

Advocate · round 2 · REBUILT

As reported by CNBC on June 26, 2017, Deutsche Bank's Torsten Slok, analysing the Survey of Professional Forecasters, found that twelve-month-ahead forecasts of the ten-year US Treasury yield were biased high: "the average 12-month forecast error since 2003 has been 60 basis points too high." The round-1 second sentence is CONCEDED and withdrawn in full. I retrieved the CNBC article and it contains no longer-window measurement and no 40 basis point figure; the record carries exactly one source, dated 2017, which cannot report a measurement ending in 2024; and I searched for a 1993-2024 mean forecast error of 40 basis points for the ten-year Treasury and could not retrieve any document establishing it, let alone one measuring the same SPF panel. The round-1 narrowing bolted an unsourced comparator onto a single-source record and made it weaker; removing it restores the record to a sourced, correctly-scoped 2003-2017 statement that says nothing about whether the bias survives the post-2021 rate surge.

Advocate · round 3 · REBUILT

As reported by CNBC on June 26, 2017, Deutsche Bank's Torsten Slok, analysing the Survey of Professional Forecasters, found twelve-month-ahead forecasts of the ten-year US Treasury yield biased high: "the average 12-month forecast error since 2003 has been 60 basis points too high." The chain behind CNBC — the Deutsche Bank note and its computation — is unpublished and uncheckable, and the sample ends in 2017. The longer-window question this record previously declared unanswerable is answerable, from a free source the record must now carry: the Federal Reserve Bank of Philadelphia's own "Error Statistics for the Survey of Professional Forecasters," TBOND file, Table 1A, horizon H=4, reports a mean error of −0.43 over 1993:Q1-2023:Q1 and −0.46 over 1997:Q1-2023:Q1 on an actual-minus-forecast convention — SPF ten-year-yield forecasts four quarters ahead were 43 to 46 basis points too high, measured through the post-2021 rate surge. Direction survives the longer window; magnitude does not, running roughly a quarter below the 60 basis points CNBC reports. The two are different estimates on different samples and horizon conventions and must not be equated or averaged. [Philadelphia Fed TBOND error-statistics file retrieved 2026-08-23; the record's source list must go from one to two.]

Verdict map

Round-2 text (superseded): "…The draw is random; the frame is not neutral, because a keyword screen for hypothesis-testing language pre-selects for confirmatory framing, so the 96.05 percent generalises to that frame and not to psychology articles at large."

Replacement: *Scheel, Schijen and Lakens report that standard psychology articles concluded positive support for the first hypothesis stated in the article in 146 of 152 cases, 96.05 percent, 95 percent confidence interval [91.61, 98.54]. Only the first hypothesis in each article was coded, so this is a first-hypothesis positive rate, not an article-level positive-result rate. The sampling frame follows Fanelli (2010): 633 journals indexed under Psychiatry/Psychology, articles published 2013-2018 containing "test\ the hypothes\" in title, abstract or keywords, yielding 1,919 matches from which 150 were drawn at random, giving a final n of 152 after replacements. The draw is random and the frame is non-neutral in construction, so the figure must not be quoted as "96 percent of psychology articles report positive results." It should not, however, be discounted as an artifact of that frame: frame-free samples return the same band. Sterling (1959) and Sterling, Rosenbaum and Weinkam (1995) sampled articles using statistical tests in named psychology journals with no hypothesis-testing keyword screen and report null-rejection rates of 99/93 percent (Journal of Experimental Psychology, 1958/1986-87), 97/97 (Comparative and Physiological Psychology), 95/98 (Consulting and Clinical Psychology) and 97/96 (Personality and Social Psychology) — a 93 to 99 percent band that brackets 96.05. This record and C-45 are the two arms of one comparison in one paper by one author team on one hand-coding pass, and are one observation for counting purposes, not two.*

Basis: Every stated figure and every element of the sampling frame verifies in every pass, and the falsifier is not met. One correction lands, on retrieved external evidence and conceded in the record's own defence: the round-2 mechanism — that the keyword screen inflates the rate by pre-selecting for confirmatory framing — was asserted rather than shown, and the controlling frame-free comparison runs against it. The narrowing is unusual in direction and is recorded as such: it removes a caution the record had claimed for itself and leaves the weaker, still material one, that the frame is non-neutral in construction and the coded unit is the first hypothesis rather than the article. Shared sourcing with C-45 does not alter either figure — a primary source reporting its own results is not a correlation defect — but it bears on breadth and stays on the face of both records.

Attack · round 1 · SELECTION

The coding rule that generates both denominators is absent from the ledger: Scheel et al., following Fanelli (2010), analysed only the first hypothesis presented in each article. 146 of 152 is a first-hypothesis positive rate, not an article-level positive-result rate, and the same restriction governs the 31 of 71 that C-45 and C-09 build on

Attack · round 2 · PROVENANCE

not independent of C-45. shared-sources.txt maps both to journals.sagepub.com/doi/10.1177/25152459211007467; the 96.05% (146/152) and 43.66% (31/71) are the two arms of a single comparison in a single paper by a single author team on a single hand-coding pass. The ledger books them as two records, one SUPPORTING and one LOAD-BEARING, when they are one observation. Compounding it: the 152-article SR sample is not a random draw of psychology articles but a Fanelli (2010)-style keyword screen for "test* the hypothes*" restricted to 2013–2018 and matched to RR journals — a frame that selects for confirmatory framing before any coding occurs.

Attack · round 3 · SELECTION

The record's entire narrowing turns on the assertion that a "test* the hypothes*" keyword screen "pre-selects for confirmatory framing," so the 96.05 percent generalises only to that frame. The controlling comparison is absent: Sterling, Rosenbaum & Weinkam (1995, The American Statistician 49(1)) and Sterling (1959) sampled articles using statistical tests in named psychology journals with no hypothesis-testing keyword screen at all, and found null-rejection rates of 93-99 percent (J. Exp. Psych. 93%/99%, Comparative & Physiological 97%/97%, Consulting & Clinical 98%/95%, Personality & Social 96%/97%). A frame-free sample returns the same number, so the frame objection is asserted rather than shown.

Null challenge · round 1 · NOT CONTESTED

Verified: 146 of 152, 96.05%, 95% CI [91.61, 98.54].

Null challenge · round 2 · NOT CONTESTED

counts, percentage, interval and coding scope all confirmed.** 146 of 152, 96.05%, 95% CI [91.61, 98.54]; the authors coded "the first hypothesis of each article." The round-1 addition — that this is a first-hypothesis rate rather than an article-level positive-result rate — is the paper's own stated method, so the added detail is supported rather than new unsupported surface. Shared-source flag with C-45 noted: both draw on one paper, but they report different, independently verified statistics from it, and a primary source reporting its own results is not the correlation problem.

Null challenge · round 3 · NOT CONTESTED

every stated figure and every element of the sampling frame verifies, and the caveat runs in the direction that costs the record reach.* 146 of 152, 96.05%, CI [91.61, 98.54]; 633 journals under Psychiatry/Psychology; 2013–2018; "test\ the hypothes\*" in title, abstract or keywords; 1,919 matches; 150 drawn at random; final n = 152 after replacements. The frame caveat restricts the claim's scope rather than extending it, which is the one direction a narrowing cannot smuggle a premise in. The "one observation with C-45" point is independently corroborated by shared-sources.txt. Falsifier unmet.

Advocate · round 1 · REBUILT

Scheel, Schijen and Lakens report that standard psychology articles concluded positive support in 146 of 152 cases, 96.05%, with a 95% confidence interval of [91.61, 98.54]. One narrowing is mandatory and the record omits it: the denominator counts articles but the coding rule counts one hypothesis per article — "if more than one hypothesis was being tested, only the first one to appear in the text was considered." This is a first-hypothesis positive rate, not an article-level positive-result rate, and the same restriction governs the 31 of 71 at C-45 and everything C-09 builds on it.

Advocate · round 2 · REBUILT

Scheel, Schijen and Lakens report that standard psychology articles concluded positive support for the first hypothesis stated in the article in 146 of 152 cases, 96.05 percent, 95% CI [91.61, 98.54]; only the first hypothesis was coded ("If more than one hypothesis was being tested, only the first one to appear in the text was considered"), so this is a first-hypothesis positive rate, not an article-level positive-result rate. Sampling frame now disclosed: the sample follows Fanelli (2010)'s method, screening 633 journals listed under Psychiatry/Psychology in Essential Science Indicators for articles published 2013-2018 containing "test* the hypothes*" in title, abstract or keywords, yielding 1,919 matches from which 150 were RANDOMLY selected (replacements giving a final n=152). The selection operates at the frame — a keyword screen that pre-selects for confirmatory framing — and not at the draw, which is random within that frame; the Inquisitor's "not a random draw" is correct about the frame and wrong about the sampling, and the distinction matters because it bounds what the 96.05% generalises to. Mandatory disclosure: this record and C-45 are the two arms of one comparison in one paper by one author team on one hand-coding pass, and are one observation for counting purposes, not two.

Advocate · round 3 · REBUILT

Scheel, Schijen and Lakens report that standard psychology articles concluded positive support for the first hypothesis stated in the article in 146 of 152 cases, 96.05 percent, 95 percent confidence interval [91.61, 98.54]. Only the first hypothesis in each article was coded, so this is a first-hypothesis positive rate, not an article-level positive-result rate. The sampling frame follows Fanelli (2010): 633 journals indexed under Psychiatry/Psychology, articles published 2013-2018 containing "test* the hypothes*" in title, abstract or keywords, 1,919 matches, 150 drawn at random, final n of 152 after replacements. WITHDRAWN: the assertion that the keyword screen "pre-selects for confirmatory framing," so that 96.05 percent generalises to that frame and not to psychology articles at large. That mechanism is asserted, not shown, and the available frame-free comparison runs against it: Sterling (1959) and Sterling, Rosenbaum and Weinkam (1995) sampled articles using statistical tests in named psychology journals with no hypothesis-testing keyword screen and report null-rejection rates of 99/93 percent (Journal of Experimental Psychology, 1958/1986-87), 97/97 (Comparative and Physiological Psychology), 95/98 (Consulting and Clinical Psychology) and 97/96 (Personality and Social Psychology) — a 93 to 99 percent band that brackets 96.05. What survives is the weaker and still material caution that the frame is non-neutral in construction and that the coded unit is the first hypothesis rather than the article, so the figure should not be quoted as "96 percent of psychology articles report positive results." This record and C-45 are two arms of one comparison in one paper by one author team on one hand-coding pass, and are one observation for counting purposes, not two.

Verdict map

Round-2 text (superseded): "…which fixes the amount directed to it at approximately $180 million; the unallocated remainder is an upper bound of somewhat under $120 million…"

Replacement: Finning's August 10, 2026 release states that proceeds "will be used for repayment of existing indebtedness, including the repayment on maturity of Finning's 2.626% Senior Unsecured Notes due August 14, 2026, of which there is approximately $180 million aggregate principal amount outstanding at the date hereof, and for general corporate purposes." The construction is inclusive, not exhaustive, and the $180 million is the series' outstanding balance rather than an application of these proceeds; the release states no net-proceeds figure at all. The release therefore commits proceeds to repayment of existing indebtedness as a class and names the 2026 maturity as one member of that class; the share applied to it is unstated and may be nil, since a maturity falling four days after pricing can be met from cash or a revolver with this five-year issue terming out that draw. On the text the unallocated remainder runs anywhere from under $120 million to the whole of net proceeds. The pair is not tenor-matched: the new notes mature August 12, 2031 — a fresh five-year coupon — against a note with four days left to run, so the 160.5 basis point gap between 2.626% and 4.231% sets a five-year new-issue coupon against a legacy coupon of different vintage and different original tenor and decomposes into none of them.

Basis: The use-of-proceeds sentence is verbatim and was confirmed again by every pass this round; the falsifier is not met. One correction lands and was conceded in the record's own defence: "which fixes the amount directed to it at approximately $180 million" is fixed by nothing in the sentence, and the "somewhat under $120 million" upper bound inherited that unsupported premise, so both are withdrawn. The word doing the work is including, which the run reads correctly as non-exclusive at C-36 and had abandoned here. A named rival cannot be excluded from the text: Finning repays the August 14 maturity from its revolver on the maturity date and applies this offering to terming out that draw alongside other existing indebtedness, under which nothing is "directed to" the 2026 notes at all. The tenor mismatch and the 160.5 basis point arithmetic are unaffected and verify. Recorded and checked twice: the 4.231% coupon shared with Capital Power's November 2025 print verifies at two unrelated primary documents nine months apart and is a coincidence, not a propagated figure.

Attack · round 1 · INFERENCE

Weaker claim supported — proceeds "will be used for repayment of existing indebtedness, including the repayment on maturity of Finning's 2.626% Senior Unsecured Notes due August 14, 2026, of which there is approximately $180 million aggregate principal amount outstanding at the date hereof, and for general corporate purposes." Roughly $120M of the $300M issue is unallocated to that maturity, so "the proceeds repay its 2.626% notes" overstates the dedication of the issue and the size of the matched pair

Attack · round 2 · INFERENCE

the release allocates no dollar amount to anything, so the "$120 million unallocated" residual is the record's arithmetic, not the issuer's statement — and net proceeds after underwriting fees are below $300 million in any case. Compounding it, the new paper is a FIVE-year (4.231% due August 12, 2031) set against a note in its final days, so the 160 bp gap is a fresh five-year coupon against a legacy coupon of a different vintage and tenor. Weaker claim actually supported: "Finning's August 10, 2026 release names the ~$180 million 2.626% August 14, 2026 maturity as one item inside an undifferentiated 'repayment of existing indebtedness' use of proceeds from a $300 million five-year issue; it states no allocation, and the pair is not tenor-matched."

Attack · round 3 · INFERENCE

"Which fixes the amount directed to it at approximately $180 million" is not supported by the sentence quoted. "Repayment of existing indebtedness, including the repayment on maturity of Finning's 2.626% Senior Unsecured Notes" is the same inclusive construction this ledger reads as non-exclusive in C-36 ("including but not limited to"); the release states the notes' outstanding principal, not the portion of proceeds applied to it, and Finning may retire any part of it from cash or credit facilities. Weaker supportable claim: the release names the 2026 maturity as one item inside a general repayment purpose and states that roughly $180 million is outstanding on it; the share of proceeds directed to it is unstated, so the "somewhat under $120 million" unallocated remainder does not follow.

Null challenge · round 1 · NOT CONTESTED

Verified verbatim: proceeds go to "the repayment on maturity of Finning's 2.626% Senior Unsecured Notes due August 14, 2026," approximately $180m outstanding.

Null challenge · round 2 · NOT CONTESTED

the use-of-proceeds sentence is verbatim and the arithmetic follows.** Finning: "$300 million principal amount of its 4.231% Senior Unsecured Notes," proceeds "for repayment of existing indebtedness, including the repayment on maturity of Finning's 2.626% Senior Unsecured Notes due August 14, 2026, of which there is approximately $180 million aggregate principal amount outstanding at the date hereof, and for general corporate purposes." The added residual point — roughly $120m not allocated to that maturity — follows directly from "including" plus the two figures. (The 4.231% coupon coinciding with Capital Power's is a genuine coincidence between two unrelated issuers, not contamination; both verify at source.)

Null challenge · round 3 · CONTESTED

CONTESTED — C-35 — same defect: "which fixes the amount directed to it at approximately $180 million" is fixed by nothing, and the $120 million bound inherits the unsupported premise.

The sentence verifies verbatim, and the word that matters is including: *"will be used for repayment

of existing indebtedness, including the repayment on maturity of Finning's 2.626% Senior Unsecured

Notes due August 14, 2026, of which there is approximately $180 million aggregate principal amount

outstanding at the date hereof, and for general corporate purposes."* Inclusive, not exhaustive; and

the $180 million is again the series' outstanding balance, not an allocation of these proceeds. The

record's own reading of "including but not limited to" on C-36 is the correct one and is abandoned here.

Because the $180 million is not fixed, the remainder is not bounded at "somewhat under $120 million."

On the text the unallocated remainder runs anywhere from under $120 million to the whole of net proceeds.

Named rival: Finning repays the August 14 maturity from its revolver on the maturity date — four days

after pricing — and applies this offering to term out that draw together with other existing

indebtedness, which is what "repayment of existing indebtedness" most naturally covers for a five-year

issue priced four days before a maturity it could fund from cash. Under that reading nothing is

"directed to" the 2026 notes at all. Nothing in the release discriminates.

Not contested within C-35: $300 million, 4.231%, August 12 2031; retired 2.626% due August 14 2026;

~$180 million outstanding; no net-proceeds figure stated — all verified verbatim. 4.231 − 2.626 =

160.5bp is right, and the non-tenor-matched caution is sound. One thing I expected to be a propagated

figure and found is not: 4.231% appears on this record and on Capital Power's November 2025 print,

and both verify at two unrelated primary documents. Coincidence, not contamination.

Advocate · round 1 · REBUILT

Finning stated that proceeds "will be used for repayment of existing indebtedness, including the repayment on maturity of Finning's 2.626% Senior Unsecured Notes due August 14, 2026, of which there is approximately $180 million aggregate principal amount outstanding at the date hereof, and for general corporate purposes." The verbatim text carries two qualifiers the record drops — "including" and "and for general corporate purposes" — so roughly $120 million of the $300 million issue is unallocated to that maturity. The matched pair is $180 million retired at 2.626% against a $300 million issue at 4.231%, not a dedicated refinancing.

Advocate · round 2 · REBUILT

Finning's August 10, 2026 release states that "Proceeds from the issuance of the Notes will be used for repayment of existing indebtedness, including the repayment on maturity of Finning's 2.626% Senior Unsecured Notes due August 14, 2026, of which there is approximately $180 million aggregate principal amount outstanding at the date hereof, and for general corporate purposes." The pair is therefore approximately $180 million retired at 2.626% against a $300 million issue at 4.231%. Two corrections. First, the tenor mismatch is now disclosed and is the material defect: the new notes mature August 12, 2031 — a fresh five-year coupon — against a note with four days left to run, so the ~160 basis point gap is a five-year new-issue coupon against a legacy coupon of different vintage and tenor and decomposes into none of them. Second, the "$120 million unallocated" residual is restated as an upper bound rather than a figure: the release states no net-proceeds number, and net proceeds after underwriting fees and expenses are below $300 million, so the unallocated remainder is somewhat under $120 million. The Inquisitor's stronger form — that the release "allocates no dollar amount to anything" — is turned back: the release names the maturity, commits proceeds to its repayment on maturity, and states its outstanding principal, which fixes the amount directed to it.

Advocate · round 3 · REBUILT

Finning's August 10, 2026 release states that proceeds "will be used for repayment of existing indebtedness, including the repayment on maturity of Finning's 2.626% Senior Unsecured Notes due August 14, 2026, of which there is approximately $180 million aggregate principal amount outstanding at the date hereof, and for general corporate purposes." WITHDRAWN: that this "fixes the amount directed to it at approximately $180 million," and with it the "somewhat under $120 million" upper bound on the unallocated remainder, which inherited the unsupported premise. "Including" is the same inclusive construction the run reads as non-exclusive on C-36, and the $180 million is the series' outstanding balance, not an application of these proceeds; the release states no net-proceeds figure at all. What survives: the release commits proceeds to repayment of existing indebtedness as a class and names the 2026 maturity as one member of that class; the share applied to it is unstated and may be nil, since Finning could meet a maturity falling four days after pricing from cash or its revolver and apply this five-year issue to terming out that draw. On the text the unallocated remainder runs anywhere from under $120 million to the whole of net proceeds. The pair is not tenor-matched: the new notes mature August 12, 2031 — a fresh five-year coupon — against a note with four days left to run, so the 160.5 basis point gap between 2.626% and 4.231% sets a five-year new-issue coupon against a legacy coupon of different vintage and different original tenor and decomposes into none of them.

Verdict map

Round-2 text (superseded): "…The absence of a 2026 series has a documented cause other than refinancing selection: Bell's C$750 million 3.55% Series M-41 debentures due March 2, 2026 were called on a November 7, 2025 redemption notice and redeemed at par on December 8, 2025, so no 2026-maturing Canadian-dollar Bell series existed for this release to name."

Replacement: Bell Canada's May 27, 2026 release names no 2026-maturing series as a use of proceeds. It states that net proceeds will be applied "(i) to repurchase, redeem or repay, as applicable, senior and/or subordinated indebtedness of Bell, including but not limited to those securities tendered in Bell's tender offers commenced on May 27, 2026 … and (ii) for general corporate purposes," so the tender offers are an inclusive example and not the exclusive destination; the release also does not identify which series the tender offers cover. Those are facts about one document, and they carry no explanatory content. The record offers no account of why no 2026 series is named, and no evidence about refinancing selection in either direction.

Basis: The documentary negative is settled and verifies verbatim in every pass; the falsifier is not met, since no specific 2026-maturing series is named. What round 3 establishes, on retrieved evidence and conceded outright in the record's own defence, is that the round-2 explanation is false rather than merely uncited, and it is struck. Bell's Series M-43, C$650 million of 2.90% MTN debentures maturing August 12, 2026, was outstanding on the date of the release — roughly eleven weeks from maturity — carried on the issuer group's own outstanding-debenture schedule, issued under Bell's August 9, 2016 offering release, with no redemption or early-repayment notice located and no appearance among the ten series covered by Bell's separate May 27, 2026 tender offers, whose earliest maturity is 2028. The March 2, 2026 Series M-41 call removed one 2026 rung, not the class, so "no 2026-maturing Canadian-dollar Bell series existed for this release to name" does not hold. A second, independent defect is accepted and is the reason the false clause was never checkable on this record: every fact in the M-41 story — the November 7, 2025 notice, the December 8, 2025 par redemption, the C$750 million size, the 3.55% coupon — is uncited on a record that prints a single source, and that source says none of it. The consequence is recorded in §1 and cuts both ways: the absence of a 2026 series from this release is now unexplained rather than explained, so it is no longer evidence against refinancing selection either, but the release's "including but not limited to" and its silence on tender composition mean nothing can be read from it in that direction either. This is a load-bearing link that does no work; the struck clause is recorded in graveyard.md so it is not re-derived.

Attack · round 1 · INFERENCE

Weaker claim supported — "Bell's 27 May 2026 release names no 2026-maturing series, and states that net proceeds will be applied to (i) repurchase, redeem or repay senior and/or subordinated indebtedness of Bell, including securities offered through concurrent tender offers, and (ii) for general corporate purposes." The record's "directed to tender offers" asserts an exclusivity the release does not, and drops the general-corporate-purposes leg entirely

Attack · round 2 · SELECTION

the negative fact is vacuous and the ledger omits the document that makes it so. Bell's Canadian-dollar 2026 maturity — Series M-41 debentures due March 2, 2026 — was called on a November 7, 2025 redemption notice, so by May 27, 2026 there was no 2026 series left for the release to name. The record presents an absence as evidence of refinancing selection while omitting the release that removed the object of the absence. A competent opponent cites it.

Attack · round 3 · SELECTION

The record's load-bearing negative — "no 2026-maturing Canadian-dollar Bell series existed for this release to name" — is false, and the document that shows it is the issuer's own. BCE's bondholder-information schedule of outstanding Bell Canada debentures lists Series M-43, 2.90%, maturing 2026-08-12 — outstanding on 27 May 2026 and under three months from maturity — alongside the 8.875% Series 1 subordinated debenture due 2026-04-17. The M-41 call removed one 2026 series, not the class, so the record's "documented cause other than refinancing selection" collapses.

Attack · round 3 · PROVENANCE

The M-41 story — the November 7, 2025 redemption notice, the December 8, 2025 par redemption, the C$750 million size and the 3.55% coupon — carries no source on this record. The record prints sources: 1, and that one source is the May 27, 2026 Bell offering release, which says none of it. Every load-bearing fact in the record's causal explanation is uncited.

Null challenge · round 1 · NOT CONTESTED

Verified verbatim from the use-of-proceeds section; the release names tender offers commenced 27 May 2026 and no specific 2026-maturing series.

Null challenge · round 2 · NOT CONTESTED

both the positive quote and the negative fact are confirmed.** The May 27, 2026 Bell release states proceeds will be applied "(i) to repurchase, redeem or repay, as applicable, senior and/or subordinated indebtedness of Bell, including but not limited to those securities tendered in Bell's tender offers commenced on May 27, 2026 … and (ii) for general corporate purposes," and names no 2026-maturing series. The round-1 narrowing is precisely calibrated: it keeps the negative fact and explicitly drops the exclusive-direction reading that "including but not limited to" will not carry. This is the model of a claim cut to its evidence.

Null challenge · round 3 · CONTESTED

CONTESTED — C-36 — a universal negative sourced entirely off the record.

"no 2026-maturing Canadian-dollar Bell series existed for this release to name" requires the **complete

schedule of Bell Canada's outstanding Canadian-dollar MTN debentures as at May 27, 2026**. This record

cites one document: the May 27 offering release. That document cannot supply the schedule. The M-41

facts come from a November 7, 2025 redemption release that appears nowhere on the source line. I

verified those facts independently (notice November 7 2025; C$750 million 3.55% due March 2 2026;

redeemed at par December 8 2025) — and that release expressly says nothing about any other Bell series.

One redeemed series does not establish exhaustion of an MTN programme running to at least Series M-70.

Bell priced Series M-68 shortly before this offering and M-69/M-70 within it; the numbering implies a

long ladder whose 2026 rungs no cited source enumerates.

Named rival: at least one other Canadian-dollar Bell obligation matured in, or was tendered for in,

2026 and simply went unnamed — in which case the record's "documented cause other than refinancing

selection" is not the cause, and the absence is a selection fact after all. The record's very next

sentence concedes the general form of this rival for the tender offers, whose composition it says

cannot be read from the release. Nothing in the ledger distinguishes "no such series existed" from

"such series existed and was not named."

Not contested within C-36: the use-of-proceeds text verifies verbatim, the "including but not limited

to" reading is right, the release names no 2026 series and does not identify the tendered series, and

the falsifier is unmet. The record's negative about the document is settled. Its explanation for the

negative is not.

Advocate · round 1 · REBUILT

Bell's May 27, 2026 release names no 2026-maturing series and states that net proceeds will be applied "(i) to repurchase, redeem or repay, as applicable, senior and/or subordinated indebtedness of Bell, including but not limited to those securities tendered in Bell's tender offers commenced on May 27, 2026... and (ii) for general corporate purposes." The negative fact — no dated 2026 maturity is targeted — holds and is load-bearing. "Directed to tender offers" asserts an exclusivity the release does not, since the tender offers are named as an inclusive example and the general-corporate-purposes leg is dropped entirely.

Advocate · round 2 · CONCEDED

The documentary fact is true and I verified it verbatim; the record nonetheless cannot do the work it is filed to do, and saying so is more useful than narrowing it again. Bell Canada's May 27, 2026 release states that proceeds will be applied "(i) to repurchase, redeem or repay, as applicable, senior and/or subordinated indebtedness of Bell, including but not limited to those securities tendered in Bell's tender offers commenced on May 27, 2026 ... and (ii) for general corporate purposes," and it names no 2026-maturing series. But Bell had no 2026-maturing Canadian-dollar series left to name: its C$750 million 3.55% Series M-41 debentures due March 2, 2026 were called on a November 7, 2025 notice and redeemed at par on December 8, 2025. A negative fact whose object was retired six months before the document was written carries no information about refinancing selection, which is the load this LOAD-BEARING record is assigned. Compounding it, the release does not identify which series the tender offers cover, so nothing about tender composition can be read off this document either. The claim survives as a true sentence and fails as evidence; it should be demoted to a documentary note or dropped, and the November 7, 2025 redemption release must be added to the ledger wherever Bell's 2026 refinancing is discussed.

Advocate · round 3 · CONCEDED

the record's load-bearing negative is false, not merely unsupported, and I could not rebuild it. Bell Canada's Series M-43, C$650 million 2.90% MTN debentures maturing August 12, 2026, was issued under Bell's own August 9, 2016 offering release, is carried on BCE's outstanding-debenture table, and carries no redemption or early-repayment notice I could find; it does not appear among the ten series covered by Bell's separate May 27, 2026 tender offers, whose earliest maturity is 2028. A 2026-maturing Canadian-dollar Bell series therefore existed and was outstanding, roughly eleven weeks from maturity, on the date of the release — so "no 2026-maturing Canadian-dollar Bell series existed for this release to name" does not hold, and the "documented cause other than refinancing selection" collapses with it. The M-41 call removed one 2026 rung, not the class; separately, every fact in that M-41 story (the November 7, 2025 notice, the December 8, 2025 par redemption, the C$750 million size, the 3.55% coupon) is uncited on a record that prints sources: 1, and its one source says none of it. What remains true is only what the run already holds uncontested elsewhere: the May 27, 2026 release names no 2026-maturing series, its "including but not limited to" is inclusive rather than exclusive, and it does not identify the tendered series. Those are facts about one document with no explanatory content, and the record should be re-entered as that and nothing more — or the causal reading struck and the record retired as a duplicate of the uncontested textual finding. [Bell 2016 offering release, BCE bondholder debenture table, Bell tender-offer release of 2026-05-27, all retrieved 2026-08-23.]

Verdict map

Claim (text unchanged from the round-2 narrowing): The Canada five-year benchmark Government of Canada yield last printed 3.35 percent on August 20, 2026, having ranged 3.29 to 3.35 percent over August 17-20 (3.31, 3.29, 3.30, 3.35), with the ten-year benchmark at 3.75 percent on the same date. The primary series is the Bank of Canada daily benchmark bond yield series (BD.CDN.5YR.DQ.YLD / BD.CDN.10YR.DQ.YLD). That series carries no August 21, 2026 observation — August 21 was a Friday whose print had not posted at the cutoff — and no 3.37 percent observation anywhere from August 10 onward. The 3.37 percent figure appears only on the aggregator quote page cited on this record, which displays a live current value at a non-persistent URL, states no convention for which bond is treated as the benchmark, and is not primary.

Basis: The text survives round 3 intact. Two further independent passes pulled the primary series directly and returned every element, including both universal negatives: five-year 3.26 (Aug 7), 3.34, 3.32, 3.30, 3.23, 3.28, 3.31, 3.29, 3.30, 3.35 (Aug 20), ten-year 3.75 on August 20; no August 21, 2026 observation; and no five-year value equal to 3.37 anywhere from August 10 onward. Two standing repairs, both carried forward unchanged and neither altering the claim. (i) Source-list repair, sharpened this round: every load-bearing number in this record comes from the Bank of Canada series, which the claim text names but which appears nowhere in the record's source list — the single cited source is the aggregator page the record itself demotes. As printed, the record demolishes an aggregator using data it never sourced. It must carry two sources: the Bank of Canada series as primary, and the aggregator page as the object of the negative. (ii) Falsifier defect: the inherited falsifier tests an August 21, 2026 print that the claim says does not exist in the primary series, and must be rewritten around the August 20 close of 3.35 percent before this record is tested again.

Attack · round 1 · PROVENANCE

Not primary for a series the claim explicitly calls a benchmark. The Bank of Canada publishes the daily benchmark Government of Canada bond yields, and the ledger's own dealer source at C-33 carries the five-year too (3.16% at 9 July 2026); the record instead cites a Trading Economics live quote page, which displays the current print — the 21 August 2026 value is retrievable only while it remains current, and TE is an aggregator with no stated convention for which bond is the benchmark

Attack · round 2 · PROVENANCE

Trading Economics is not "the primary series" — it is a scraped aggregator with a non-persistent quote page, and the record calls it primary in its own claim text. The primary series is the Bank of Canada benchmark bond yield series BD.CDN.5YR.DQ.YLD, which I pulled: 2026-08-17 3.31, 08-18 3.29, 08-19 3.30, 08-20 3.35, with the 10-year at 3.75 on 08-20. Citing an aggregator while asserting primacy is a provenance misstatement on a record whose entire content is one number.

Attack · round 2 · INFERENCE

the "3.35 to 3.37 percent" range has no upper support. Every print in the week ending August 21, 2026 in the primary series sits at or below 3.35 — 3.31, 3.29, 3.30, 3.35 — and there is no 3.37 observation anywhere in the series from August 10 onward, nor any August 21 observation at all. Weaker claim actually supported: "the Canada five-year benchmark closed at 3.35 percent on August 20, 2026, having ranged 3.29 to 3.35 percent over August 17–20; the record's stated falsifier value of 3.37 percent does not appear in the primary series."

Attack · round 3 · PROVENANCE

Every load-bearing number here — the 3.31 / 3.29 / 3.30 / 3.35 August 17-20 five-year prints, the 3.75 ten-year, the absence of an August 21 observation and the absence of any 3.37 from August 10 onward — comes from the Bank of Canada Valet series BD.CDN.5YR.DQ.YLD / BD.CDN.10YR.DQ.YLD, which the claim text names but which appears nowhere in the record's source list. The record's single cited source is the Trading Economics page the record itself demotes as non-primary. The claim's primary series is uncited, so the record demolishes an aggregator using data it never sourced.

Null challenge · round 1 · NOT CONTESTED

Verified: "The yield on Canada 5 Year Bond Yield rose to 3.37% on August 21, 2026."

Null challenge · round 2 · CONTESTED

**C-40 — the cited page contradicts the record's added sentence, and a live quote page cannot

anchor a dated print.**

The claimed range survives: Trading Economics does show 3.37% for the Canada 5-year on

August 21, 2026. The round-1 addition does not. The record states "the primary series shows

3.35 percent at its last available observation, August 20, 2026, with the ten-year at 3.75

percent." The cited page shows an August 21 observation at 3.37% ("rose to 3.37% on

August 21, 2026, marking a 0.01 percentage point increase from the previous session") and the

ten-year at 3.76%, not 3.75%.

Named rivals, neither of which leaves the sentence standing: either the page updated between

the run's read and mine — in which case the record is citing a source structurally incapable of

supporting a dated observation, since the same URL returns different values by read time — or

the record misread the page. Nothing in the ledger distinguishes them.

Note also that the falsifier tests an August 21 print that the claim itself denies exists.

Null challenge · round 3 · NOT CONTESTED

I checked the primary series directly and every element of the claim, including both universal negatives, holds. BD.CDN.5YR.DQ.YLD for August 17–20, 2026: 3.31, 3.29, 3.30, 3.35; BD.CDN.10YR.DQ.YLD on August 20: 3.75. Exactly as claimed. There is no August 21, 2026 observation in the series, and no 5-year value equal to 3.37** anywhere from August 10 onward (3.34, 3.32, 3.30, 3.23, 3.28, 3.31, 3.29, 3.30, 3.35). August 21, 2026 is a Friday. The characterisation of the aggregator page as non-primary, live-valued, convention-free and non-persistent is correct on its face. The one element I could have pressed — the gloss "whose print had not posted at the cutoff" as the reason for the absence — is consistent with what I observe (the observation is still absent) and is not load-bearing to anything the record does.

Advocate · round 1 · REBUILT

The Canada five-year benchmark Government of Canada yield stood at approximately 3.35 to 3.37 percent in the week ending August 21, 2026. I checked the primary series: the Bank of Canada publishes the daily benchmark Government of Canada bond yields, and its last available observation is 3.35% on August 20, 2026, with the ten-year at 3.75%. The record's exact 3.37% for August 21 comes from Trading Economics, an aggregator displaying a live current print with no stated convention for which bond is the benchmark, retrievable only while that print remains current. The claim should be stated to the precision the primary source supports and sourced to the Bank of Canada, whose series the ledger's own dealer source at C-33 also carries.

Advocate · round 2 · REBUILT

The Canada five-year benchmark Government of Canada yield closed at 3.35 percent on August 20, 2026, the last available observation in the week ending August 21, having ranged 3.29 to 3.35 percent over August 17-20 (3.31, 3.29, 3.30, 3.35), with the ten-year benchmark at 3.75 percent on the same date. The primary series is the Bank of Canada benchmark bond yield series (BD.CDN.5YR.DQ.YLD / BD.CDN.10YR.DQ.YLD), not Trading Economics; the round-1 text's implicit attribution of primacy to an aggregator quote page is withdrawn, and the ledger's source line should be replaced. The "3.35 to 3.37 percent" range is withdrawn: I pulled the Bank of Canada series directly for August 10-22, 2026 and 3.37 appears nowhere in it, nor does any August 21 observation — August 21, 2026 was a Friday whose print had not posted at the cutoff. [This is the second record where the attackers contradicted each other, and the primary series settles it in the record's favour on the second sentence and against it on the first: the round-1 addition is exactly right, and the Null's 3.37 is an artifact of a non-persistent aggregator page that returns different values by read time — which is itself the Inquisitor's provenance point. The record's stated falsifier, which tests a 3.37 print, must be rewritten around the 3.35 close.]

Advocate · round 3 · REBUILT

the claim text survives intact and every element of it, including both universal negatives, verifies at the primary series directly; the defect is provenance, and it is repaired by adding the source the claim already names. Bank of Canada Valet, BD.CDN.5YR.DQ.YLD and BD.CDN.10YR.DQ.YLD, observations 2026-08-07 through 2026-08-22 retrieved 2026-08-23: five-year 3.26 (Aug 7), 3.34, 3.32, 3.30, 3.23, 3.28, 3.31, 3.29, 3.30, 3.35 (Aug 20); ten-year 3.75 on August 20. The latest observation is August 20, 2026; there is no August 21, 2026 observation, August 21 being a Friday whose print had not posted at the cutoff; and no five-year value equals 3.37 anywhere from August 10 onward. The 3.37 percent figure appears only on the aggregator quote page, which displays a live current value at a non-persistent URL, states no convention for which bond is treated as the benchmark, and is not primary. REQUIRED REPAIR: the record must carry two sources — the Bank of Canada series as primary, and the aggregator page as the object of the negative — because as printed it demolishes an aggregator using data it never cited.

Verdict map

Round-2 text (superseded): "…Converted to asset levels by exp(β) − 1, COVID-window issuers were roughly 46 percent larger in the euro area, 40 percent larger in other advanced economies and 24 percent larger in the United States, the last at 10 percent significance only."

Replacement: **International Finance Discussion Paper 1390, Table 6, estimates a within-nation, within-year contrast between bond issuers in stress windows and bond issuers in normal conditions, with industry (one-digit SIC), year and nation fixed effects and standard errors clustered at the industry level; the table note reads that the regressions examine firm characteristics of bond issuers during periods of financial stress relative to bond issuers during normal market conditions. It is not a comparison of issuers to any sample mean. The full advanced-economy grid, previously carried only in its COVID column, is: euro area 0.247 (0.161) GFC, 0.0517 (0.0327) taper tantrum, 0.379 (0.145) COVID; other advanced economies 0.185 (0.125), 0.0790 (0.0536), 0.340 (0.134); United States 0.516 (0.0818), −0.0602 (0.0413), 0.213 (0.0903). Four of the nine advanced-economy cells are positive and nominally significant, the largest and most precisely estimated being the United States GFC coefficient at the 1 percent level; the euro-area and other-advanced-economy COVID coefficients are nominally significant at 5 percent and the United States COVID coefficient only at 10 percent; the single negative advanced-economy point estimate, the United States taper tantrum, is not significant at any level. Other emerging-market economies are −0.145 (0.0849) COVID and −0.00246 (0.0778) GFC, both insignificant, and 0.319 (0.0835) taper tantrum. Converted by exp(β) − 1, the COVID-window premium in conditional geometric-mean assets — not in mean size — is about 46 percent (euro area), 40 percent (other advanced economies) and 24 percent (United States); on the reported standard error the United States 95 percent interval runs from about 4 to about 48 percent and is not distinguishable from zero at the 5 percent level, so that magnitude must not be stated as a point fact. The "3.6, 3.5 and 2.3 percent" figures stay withdrawn: each is a coefficient divided by the reported sample-mean log assets for that region (10.25, 9.73, 9.31), a log-point difference expressed as a percentage of a log level, which has no economic meaning. INFERENCE CAVEAT that travels with the record: clustering on one-digit SIC yields at most ten clusters, where cluster-robust variance estimation is downward-biased and over-rejects, and the paper reports no wild-cluster bootstrap, alternative clustering level or placebo window, so the printed stars overstate precision throughout. And a size tilt among firms that can access the primary bond market at all in a stress window is in part a fact about who gets to issue, which this design does not separate from issuer characteristics conditional on issuing.**

Basis: Table 6, the verbatim table note, all coefficients and standard errors, the fixed effects and clustering, and the sample-mean log assets verify in every pass. Three landings. (i) The record carried only the COVID column; the full stress-episode grid is now on the record. The objection that the omitted columns defeat the record is not accepted — the omission ran both ways, and the column most damaging to the objection is the United States GFC coefficient of 0.516 (0.0818) at 1 percent significance, the strongest cell in the table and one that cuts for the record. (ii) The exp(β) − 1 conversion is over-read twice over: it gives the proportional gap in conditional geometric-mean assets rather than in mean size, and on the reported standard error the United States figure is an interval spanning roughly 4 to 48 percent. Both corrections are written into the text. (iii) A precision caveat that no record in the run had addressed is accepted and written in as a travelling condition: nine or ten clusters is far too few for the cluster-robust standard errors the paper reports, so nominal significance overstates true rejection rates, including on the two 5 percent coefficients. The named rival — that the advanced-economy size tilt is noise inflated by too-few-clusters inference, compounded by a survivorship effect in who reaches the primary market at all — is not excluded, and the record now carries it rather than contesting it. Falsifier defect, now repaired. As printed, the first limb ("the paper reports no size differential between stress-period issuers and sample means") tests a comparison the claim's own text establishes the paper does not make, and is dead on arrival; the second limb ("or a negative differential in advanced economies") is tripped on a literal reading by the United States taper-tantrum point estimate of −0.0602, which is statistically indistinguishable from zero. Both are defects in the test rather than falsifications of the claim. The falsifier is restated as: Table 6 reports no positive stress-versus-normal issuer size differential in any advanced-economy grouping, or reports a statistically significant negative one. So restated it is not met. The contrast with C-11 stands: no advanced-economy grouping is dropped here.

Attack · round 1 · PROVENANCE

Misread units. The paper reports that log assets of COVID-window issuers were 3.6%, 3.5% and 2.3% higher than the sample mean (Table 6 coefficients 0.379, 0.340, 0.213), not that firms were 2.3 to 3.6 percent larger. A 0.379 log-point differential in log assets corresponds to a far larger difference in asset levels, so the record both misstates the units and understates the paper's own finding by roughly an order of magnitude

Attack · round 2 · INFERENCE

the three coefficients are not of equal evidentiary standing and the record states them as though they were. Table 6 reports 0.379 (euro area), 0.340 (other AEs) and 0.213* (United States) — the US figure carries a single star, i.e. significance only at the 10 percent level, and it is the one the record needs to make the differential global. Weaker claim actually supported: "in the euro area and other advanced economies COVID-window issuers were significantly larger at the 5 percent level; in the United States the size differential is significant only at the 10 percent level." The record's added log-vs-levels gloss does not touch this.

Attack · round 3 · SELECTION

The record quotes only the COVID column of Table 6 and reads the size differential as a stress-period regularity. The same table reports two further stress episodes that do not support it: euro area taper tantrum 0.0517 (n.s.) and GFC 0.247 (n.s.); other AEs 0.0790 (n.s.) and 0.185 (n.s.); United States taper tantrum −0.0602 — a negative coefficient in an advanced economy, which is the second limb of this record's own falsifier. Other EMEs are −0.145 (COVID) and −0.00246 (GFC). None of these columns is in the ledger.

Attack · round 3 · INFERENCE

"COVID-window issuers were ... roughly 24 percent larger in the United States" over-reads a point estimate twice over. exp(β)−1 on a log-assets dependent variable gives the proportional gap in conditional geometric-mean assets, not in mean size; and on the reported standard error of 0.0903 the 95 percent interval for the United States runs exp(0.036)−1 = 3.7 percent to exp(0.390)−1 = 47.7 percent. Weaker supportable claim: in the United States the COVID-window issuer size premium in conditional geometric-mean assets lies somewhere between about 4 and 48 percent and is not distinguishable from zero at the 5 percent level.

Null challenge · round 1 · NOT CONTESTED

Verified verbatim: "log assets of issuing firms were 3.6%, 3.5% and 2.3% larger in the euro area, other AEs and U.S., respectively, relative to the sample mean." All three advanced-economy buckets fall inside the stated range — the direct contrast with C-11, where one bucket was dropped to manufacture a narrower one.

Null challenge · round 2 · CONTESTED

C-43 — the coefficients are right; the estimand is not.

Table 6 verifies: 0.379 (euro area), 0.340 (other AEs), 0.213 (US). What the record does with

them fails. The claim reads these as showing that issuing firms' log assets were "larger than

the sample mean." They do not. The coefficient is the difference in average characteristics

of issuing firms in stress periods relative to issuing firms in non-stress periods, in a

specification carrying industry (one-digit SIC), year and nation fixed effects with errors

clustered at industry. It is a within-nation, within-year, stress-vs-normal contrast among

issuers — not a comparison of issuers to any sample mean. That substitution changes what the

number is evidence of, which is the whole of this record's load.

Second: the "3.6, 3.5 and 2.3 percent" figures are uninterpretable as printed. They are

log-point coefficients divided by unstated denominators, and the implied denominators are not

even constant across the three (0.379/3.6 = 10.53; 0.340/3.5 = 9.71; 0.213/2.3 = 9.26). The

record's own hedge concedes the log-versus-level problem and then reports the percentages

anyway. A quantity whose base is neither stated nor recoverable is not a settled quantity.

Null challenge · round 3 · CONTESTED

CONTESTED — C-43 — the limb the record says it needs is the limb the inference cannot carry.

The record concedes the US coefficient is significant at 10% only, states that "it is the United States

figure the record needs to make the differential general," and then asserts the magnitude as fact:

"roughly ... 24 percent larger in the United States."

The inference problem no record in the run addresses: standard errors are clustered at the **industry

level using one-digit SIC codes** — at most ten clusters, realistically nine. Cluster-robust variance

estimation with that few clusters is downward-biased and over-rejects; the printed stars are not the

true rejection rates. Under valid inference a coefficient at the nominal 10% level is not

distinguishable from zero, and the nominal 5% levels on the euro-area and other-AE coefficients are

themselves not secure.

Named rival: the advanced-economy size tilt is sampling noise inflated to nominal significance by

too-few-clusters inference, compounded by a selection effect — in a stress window the firms that can

access the primary bond market at all are the large ones, which is a fact about who gets to issue,

not about issuer characteristics conditional on issuing. The paper reports no wild-cluster bootstrap,

no alternative clustering level, and no placebo window; no record in the ledger supplies one. Nothing

discriminates.

Record-internal, secondary: the first limb of the stated falsifier — "The paper reports no size

differential between stress-period issuers and sample means" — refers to a comparison the claim's

own text says the paper does not make ("It is not a comparison of issuers to any sample mean"). That

limb is dead on arrival; the narrowing corrected the estimand and left the falsifier behind. The record

is testable only by its second limb.

Not contested within C-43: Table 6 attribution, the verbatim table note, the coefficients and standard

errors (0.379/0.145, 0.340/0.134, 0.213/0.0903), the fixed effects and clustering, the sample-mean log

assets (10.25, 9.73, 9.31), and the significance levels all verify. The withdrawal of "3.6, 3.5 and 2.3

percent" is correct — dividing a log-point coefficient by a mean log level is meaningless — and the

exp(β) − 1 conversions are right (46.1%, 40.5%, 23.7%).

Advocate · round 1 · REBUILT

IFDP 1390 states that during COVID "log assets of issuing firms were 3.6%, 3.5% and 2.3% larger in the euro area, other AEs and U.S., respectively, relative to the sample mean," from Table 6 coefficients of 0.379, 0.340 and 0.213 — I retrieved both the table and the sentence. The record's defect is the omission of two words: it drops "log assets" and so reports a differential in asset levels that the paper does not claim, since a 0.379 log-point coefficient corresponds to a far larger level difference than 3.6 percent. Restored to "log assets of issuing firms were 2.3 to 3.6 percent larger than sample means across advanced economies," the claim is exactly what the paper says, and — unlike C-11 — all three advanced-economy buckets fall inside the stated range with no grouping dropped.

Advocate · round 2 · REBUILT

International Finance Discussion Paper 1390 reports in Table 6 that during the COVID stress window, bond issuers had larger log assets than issuers in NON-STRESS periods — a within-nation, within-year, stress-versus-normal contrast among issuers, in a specification with industry (one-digit SIC), year and nation fixed effects and standard errors clustered at the industry level. The round-1 framing "larger than the sample mean" is withdrawn as the wrong estimand: the table note reads "These regressions examine firm characteristics of bond issuers during periods of financial stress, relative to bond issuers during normal market conditions." The coefficients are 0.379 (s.e. 0.145) for the euro area, 0.340 (0.134) for other advanced economies and 0.213 (0.0903) for the United States, and the significance is not uniform — the first two are significant at the 5 percent level and the US figure only at the 10 percent level, which is the one the record needs to make the differential general. The "3.6, 3.5 and 2.3 percent" figures are withdrawn as uninterpretable, and I can now say exactly what they were: the coefficient divided by the reported sample-mean LOG assets for each region (10.25, 9.73, 9.31), which reproduces 3.5 and 2.3 exactly and 3.6 within a tenth — that is, a log-point difference expressed as a percentage of a log level, a quantity with no economic meaning. The correct conversion to asset levels is exp(beta)-1: COVID-window issuers were roughly 46 percent larger in the euro area, 41 percent larger in other advanced economies and 24 percent larger in the United States, the last at 10 percent significance only. [Neither attacker reconstructed the arithmetic; retrieving the Table 6 sample means makes the round-1 percentages diagnosable rather than merely doubtful, and yields defensible replacements.]

Advocate · round 3 · REBUILT

International Finance Discussion Paper 1390, Table 6, estimates a within-nation, within-year contrast between bond issuers in stress windows and bond issuers in normal conditions, with industry (one-digit SIC), year and nation fixed effects and standard errors clustered at the industry level; the table note reads that the regressions "examine firm characteristics of bond issuers during periods of financial stress, relative to bond issuers during normal market conditions." It is not a comparison of issuers to any sample mean, so the first limb of the printed falsifier is dead on arrival and must be struck. The full advanced-economy grid, previously carried only in its COVID column, is: euro area 0.247 (0.161) GFC, 0.0517 (0.0327) taper tantrum, 0.379 (0.145) COVID; other advanced economies 0.185 (0.125), 0.0790 (0.0536), 0.340 (0.134); United States 0.516 (0.0818)**, −0.0602 (0.0413), 0.213 (0.0903). Four of the nine advanced-economy cells are positive and nominally significant — including the largest and most precisely estimated coefficient in the table, the United States GFC figure at the 1 percent level, which the SELECTION attack omitted and which cuts for this record rather than against it. None is significantly negative; the single negative advanced-economy point estimate carries no stars, so the second limb of the falsifier is not met either. Other EMEs are −0.145 (0.0849) COVID and −0.00246 (0.0778) GFC, both insignificant, and 0.319 (0.0835)*** taper tantrum. Converted by exp(β)−1, the COVID-window premium in conditional geometric-mean assets — not in mean size — is about 46 percent (euro area), 40 percent (other advanced economies) and 24 percent (United States); on the reported standard error the United States 95 percent interval runs about 4 to 48 percent and is not distinguishable from zero at the 5 percent level, so that magnitude must not be stated as a point fact. The "3.6, 3.5 and 2.3 percent" figures stay withdrawn: a log-point coefficient divided by a mean log level has no economic meaning. INFERENCE CAVEAT that must travel with the record: clustering on one-digit SIC yields at most ten clusters, where cluster-robust variance is downward-biased and over-rejects, and the paper reports no wild-cluster bootstrap, alternative clustering level or placebo window — so the printed stars overstate precision throughout, including on the two 5 percent coefficients. And a size tilt among firms that can access the primary bond market at all in a stress window is in part a fact about who gets to issue, which this design does not separate from issuer characteristics conditional on issuing. [Table 6 retrieved in full 2026-08-23.]

Verdict map

Round-2 text (superseded): "…The 43.66 percent blends two populations with different prior probabilities of a positive result: replications are 41 of 71 Registered Reports (57.75 percent) against 4 of 152 standard reports (2.63 percent)."

Replacement: Scheel, Schijen and Lakens report that Registered Reports concluded positive support for the first preregistered hypothesis in 31 of 71 cases, 43.66 percent, 95 percent confidence interval [31.91, 55.95] — an interval spanning 24 points; only the first preregistered hypothesis was coded and unregistered pilot studies were excluded. The composition effect is real but asymmetric, and does not operate as a general blending of two populations. Within Registered Reports, replications are 16 of 41 positive (39.02 percent) against 15 of 30 originals (50.00 percent), so the mix lowers the Registered Report figure by about 11 points. Within standard reports the four replications are 4 of 4 positive (100 percent), above the 142 of 148 original rate (95.95 percent), so the mix raises the comparator marginally rather than depressing it: the replication share depresses the Registered Report arm alone and leaves the standard-literature figure untouched. Restricted to original studies only, the Registered Report positive rate is 15 of 30, 50.00 percent, 95 percent confidence interval [31.30, 68.70], against 142 of 148, 95.95 percent [91.39, 98.50], for original standard reports — a difference of 45.95 points that remains significant. Both the blended and the original-only figures must travel with this record. This record and C-30 are one observation, not two.

Basis: Every figure verifies and every derived figure reconciles internally in every pass: 31/71 = 43.66 percent; replications 41/71 = 57.75 percent against 4/152 = 2.63 percent; originals-only 15/30 = 50.00 percent against 142/148 = 95.95 percent; and the cross-checks close, 142 + 4 = 146 and 15 + 16 = 31. The falsifier is not met. One correction lands, on retrieved evidence and conceded in the record's own defence: the round-2 description of the mechanism was wrong in kind. "Blends two populations with different prior probabilities of a positive result" implies a symmetric effect on both arms, and the standard-report arm's four replications are 4 of 4 positive — above its own original rate — so replication share cannot be depressing the comparator. Correctly stated, the confound is one-sided and worth about 11 points on the Registered Report arm alone. Once corrected the comparison survives at 50.00 against 95.95 percent, so the confound narrows the record rather than overturning it. Both the 24-point interval and the 50 percent original-only figure bear directly on any attempt to convert this rate into a point prior, which is what C-09 attempted.

Attack · round 1 · SELECTION

The interval around the number is absent from the ledger: the paper reports the Registered Report positive rate as 43.66% with a 95% confidence interval of [31.91, 55.95] on n=71 — a 24-point band, which C-09 converts into a point "prior" of "nearer 44 percent than 96 percent"

Attack · round 2 · SELECTION

the ledger omits the composition fact the paper itself flags as the leading confound. Replications are 41 of 71 RRs (57.75%) against 4 of 152 SRs (2.63%), and restricting to ORIGINAL studies the RR positive rate is 50%, not 43.66% (against 95.95% for original SRs). A competent opponent cites the original-studies-only figure, because 43.66% is a blend of two populations with different prior probabilities of a positive result. Nothing in the record's claim text mentions replications, the 57.75% share, or the 50% original-only rate.

Attack · round 3 · INFERENCE

"The 43.66 percent blends two populations with different prior probabilities of a positive result" is not supported as a general mechanism. In the standard-report arm the four replications were 4 of 4 positive (100 percent), above the 95.95 percent original rate, so replication share does not depress the comparator at all — it raises it. Weaker supportable claim: within Registered Reports only, replications are 16 of 41 positive (39.02 percent) against 15 of 30 (50.00 percent) for originals, so the blend lowers the RR figure by about 11 points and leaves the standard-literature figure untouched.

Null challenge · round 1 · NOT CONTESTED

Verified: 31 of 71, 43.66%, 95% CI [31.91, 55.95]. The arithmetic is settled; only its transfer to finance (C-09) is not.

Null challenge · round 2 · NOT CONTESTED

counts, percentage, interval and coding scope all confirmed.** 31 of 71, 43.66%, 95% CI [31.91, 55.95]; for Registered Reports the authors "coded the first preregistered hypothesis, thus excluding unregistered pilot studies." The interval spans 24.04 points, matching the record's "24-point interval." Both added qualifications are the paper's own.

Null challenge · round 3 · NOT CONTESTED

every figure verifies and every derived figure reconciles internally.** 31/71 = 43.66%, CI [31.91, 55.95]; replications 41/71 = 57.75% against 4/152 = 2.63%; originals-only 15/30 = 50.00%, CI [31.30, 68.70] against 142/148 = 95.95%, CI [91.39, 98.50]. The cross-checks close: 142 + 4 = 146 and 15 + 16 = 31. The 45.95-point difference is arithmetically right and the record does not overstate its significance. The instruction that both the blended and original-only figures travel together is a constraint on use, not an unsupported assertion — it is the record refusing a reading it could have exploited. Falsifier unmet.

Advocate · round 1 · REBUILT

Scheel, Schijen and Lakens report Registered Reports concluding positive support in 31 of 71 cases, 43.66%, with a 95% confidence interval of [31.91, 55.95] — verified, along with the standard-report figure of 96.05% [91.61, 98.54] and the -52.39 point difference (χ²(1)=77.96, p<.001). Two things belong with the number and are absent: the 24-point interval, which C-09 discards in converting 43.66% into a point "prior"; and the first-hypothesis coding rule that produces the denominator, as at C-30.

Advocate · round 2 · REBUILT

Scheel, Schijen and Lakens report that Registered Reports concluded positive support for the first hypothesis in 31 of 71 cases, 43.66 percent, 95% CI [31.91, 55.95]; only the first preregistered hypothesis was coded, excluding unregistered pilot studies, and the interval spans 24 points. The composition confound the Inquisitor identifies is real and is now disclosed: replications were far more common among RRs than SRs (41/71 = 57.75% against 4/152 = 2.63%), so 43.66 percent blends two populations with different prior probabilities of a positive result. But the correction does not overturn the record — it survives it. Restricting to ORIGINAL studies only, the RR positive rate is 15 of 30 = 50.00 percent, 95% CI [31.30, 68.70], against 142 of 148 = 95.95 percent, CI [91.39, 98.50], for original SRs, a difference of 45.95 percentage points that remains significant. Both figures should be carried. Mandatory disclosure: this record and C-30 are the two arms of one comparison in one paper and are one observation, not two.

Advocate · round 3 · REBUILT

Scheel, Schijen and Lakens report that Registered Reports concluded positive support for the first preregistered hypothesis in 31 of 71 cases, 43.66 percent, 95 percent confidence interval [31.91, 55.95] — an interval spanning 24 points; only the first preregistered hypothesis was coded and unregistered pilot studies were excluded. CORRECTED MECHANISM: "blends two populations with different prior probabilities of a positive result" is the wrong description, because the mix does not operate symmetrically on the two arms. Within Registered Reports, replications are 16 of 41 positive (39.02 percent) against 15 of 30 originals (50.00 percent), so the blend lowers the RR figure by about 11 points. Within standard reports the four replications are 4 of 4 positive (100 percent), above the 142 of 148 original rate (95.95 percent), so the mix raises the comparator marginally rather than depressing it — the replication share depresses the RR arm only and leaves the standard-literature figure untouched. Restricted to original studies only, the Registered Report positive rate is 15 of 30, 50.00 percent [31.30, 68.70], against 142 of 148, 95.95 percent [91.39, 98.50], for original standard reports — a difference of 45.95 points that remains significant. The cross-checks close: 142 + 4 = 146 and 15 + 16 = 31. Both the blended and the original-only figures must travel with this record. This record and C-30 are one observation, not two. [Breakdown retrieved 2026-08-23.]

Verdict map

Round-2 text (superseded): "…and that wording is identical on the issuer page and on the syndication cited here, so no second source exists to settle it."

Replacement: On November 5, 2025 Capital Power priced C$600 million of medium term notes at 4.231% due January 14, 2033 — a 7.2-year instrument — and separately announced redemption of all C$300 million of its 4.986% notes due January 23, 2026, redeemed November 23 with the redemption price paid November 24, 2025, leaving under three months of residual life at pricing. The pricing release states only a permissive intention: proceeds to "repay, redeem or refinance existing indebtedness, which may include fully funding the redemption of the Company's January 2026 Notes," an enumeration alongside project-level debt repayment and general corporate purposes, and that release alone does not establish that the new notes funded the redemption. A second source settles it in the past tense and must be carried on this record: Capital Power's 2025 fourth-quarter and year-end report of March 4, 2026 states that the company "used the net proceeds to repay, redeem and refinance existing indebtedness, which included fully funding the redemption of the Company's January 2026 Notes … as well as project level debt at Goreway Power Station, Capital Power's credit facilities, and for general corporate purposes." The permissive enumeration was realised. The 75.5 basis point gap between the two coupons nonetheless sets a fresh 7.2-year coupon against a legacy coupon fixed in a different rate regime at a different original tenor; it confounds tenor, vintage and the level of the risk-free curve and isolates none of them. The pair is a 2025 transaction outside the 2026 window it is used to characterise, and the pricing leg rests on the same single press release as C-20 and C-46.

Basis: Both coupons, both dates, the 7.2-year term and the 75.5 basis point arithmetic verify in every pass, and the falsifier is not met — the new notes priced below the redeemed coupon. One correction lands, on retrieved evidence and conceded in the record's own defence: "no second source exists to settle it" is false, and the document that settles it — the issuer's own year-end report, four months later — was absent from the record. The record's refusal to read a funding mechanism into the pricing release alone remains correct as a reading of that release, and remains the standard the run should have applied at C-28 and C-35; what changes is that the question is no longer open. The consequence is recorded in §1: this is now a documented matched Canadian-dollar non-financial refinancing pair, the run's only one, with a sign opposite to the direction of the statistic it was filed to support, still confounded by tenor and vintage, and still dated outside the window.

Attack · round 1 · PROVENANCE

The fourth record on one press release, and the weakest-tier of the four (E1, via Barchart) despite the primary being public at capitalpower.com; the ledger thus counts a single 2025 issuer statement four times, and this record's direction runs against the thesis it is LOAD-BEARING for — 4.986% retired against 4.231% issued is a 75.5bp fall in coupon

Attack · round 2 · INFERENCE

"a 75.5 basis point fall in coupon" measures nothing about pricing. The retired 4.986% notes matured January 23, 2026 and were redeemed November 24, 2025 — roughly two months of remaining life — while the replacement matures January 14, 2033, a 7.2-year instrument. The comparison sets a fresh seven-year coupon against a legacy coupon fixed in a different rate regime at a different original tenor. Weaker claim actually supported: "Capital Power's new 7.2-year notes priced at 4.231%, 75.5 basis points below the legacy coupon on notes it redeemed with under three months to maturity; the differential confounds tenor, vintage and the level of the risk-free curve and isolates none of them." The record's own C-04 states this confound as a general principle and C-48 commits it.

Attack · round 3 · SELECTION

The record asserts the funding link cannot be settled — "that wording is identical on the issuer page and on the syndication cited here, so no second source exists to settle it." A second source exists and settles it in the past tense: Capital Power's Q4 and year-end 2025 report of March 4, 2026 states that "Capital Power used the net proceeds to repay, redeem and refinance existing indebtedness, which included fully funding the redemption of the Company's January 2026 Notes." The permissive "may include" of the pricing release was in fact realised, and the ledger does not cite the document that says so.

Null challenge · round 1 · NOT CONTESTED

Verified: the 2033 notes priced at 4.231%, below the 4.986% redeemed January 2026 notes. The falsifier is not met. ---

Null challenge · round 2 · CONTESTED

**C-48 — the funding linkage is conditional in the source, and the differential compares

instruments the ledger's own C-04 declares incomparable.**

Both coupons and both dates verify. Two things do not.

1. "Using new notes priced at 4.231 percent." The release states only that Capital Power

"intends to use the net proceeds of the Offering to repay, redeem or refinance existing

indebtedness, which may include fully funding the redemption of the Company's January 2026

Notes" — alongside project-level debt repayment and general corporate purposes. The record

converts a permissive enumeration into a stated funding mechanism. I confirmed this wording is

identical on both the issuer page and the Barchart syndication, which is the same GlobeNewswire

release, so no second source exists to settle it.

2. "A 75.5 basis point fall in coupon." The redeemed paper was C$300m 4.986% due

January 23, 2026 — roughly two months of residual life at the November 5, 2025 pricing. The

new paper is C$600m 4.231% due January 14, 2033 — 7.2 years. Subtracting the coupon on a

two-month residual from the coupon on a 7.2-year new issue and reporting the result as a "fall

in coupon" is exactly the operation C-04, load-bearing in this same round, says "measures term

premium and the level of the risk-free curve, not credit repricing." The ledger cannot hold both

that the tenor gap disqualifies such comparisons and that this one yields a clean 75.5 bp

result. Nothing here tells us which component the 75.5 bp is.

Correlation note (not itself a defeater): shared-sources.txt splits the Capital Power

evidence across two hosts, but I fetched both and the Barchart page is a verbatim syndication of

the November 5, 2025 issuer release. C-20, C-46 and C-48 are three ledger records resting on one

document. That does not make any of them false — and I have not treated it as doing so — but any

downstream count that treats them as three observations of refinancing behaviour is counting one.

Null challenge · round 3 · NOT CONTESTED

every document fact and every arithmetic step verifies, and the reading it takes of its own source is the correct one.* C$600 million at 4.231% due January 14, 2033 (7.19 years from November 5, 2025); C$300 million of 4.986% notes due January 23, 2026, redeemed November 23 with payment November 24 (2.6 months of residual life at pricing); 4.986 − 4.231 = 75.5bp. The use-of-proceeds sentence verifies verbatim including "which may include fully funding the redemption"*, and the record's refusal to read that as funding the redemption is right — it is the standard C-28 and C-35 should have applied. The one off-record element, "identical on the issuer page", is near-tautological for a syndication of a GlobeNewswire original and cuts against the record's evidentiary weight rather than for it. The confound caution and the outside-the-window point are both correct. Falsifier unmet. ---

Advocate · round 1 · REBUILT

Capital Power redeemed C$300 million of 4.986% notes maturing January 2026 using notes priced at 4.231% — a 75.5 bp fall in coupon, verified at the issuer rather than through Barchart, and the record should be re-tiered accordingly. Two facts must be stated on the record's face rather than left to inference: this is the fourth ledger record on one November 5, 2025 press release, and it is the run's only cleanly matched Canadian dollar non-financial 2026-maturity refinancing pair, whose sign runs opposite to the direction the statistic it is LOAD-BEARING for asserts. A record that contradicts the thesis it supports is not a defect in the record.

Advocate · round 2 · REBUILT

Capital Power priced C$600 million of new 7.2-year notes at 4.231% due January 14, 2033 on November 5, 2025, and separately announced redemption of C$300 million of 4.986% notes due January 23, 2026, redeemed November 23 with the redemption price paid November 24, 2025 — under three months of residual life at pricing. Two round-1 elements are withdrawn. First, the funding linkage: the release states only that Capital Power "intends to use the net proceeds of the Offering to repay, redeem or refinance existing indebtedness, which may include fully funding the redemption of the Company's January 2026 Notes," a permissive enumeration alongside project-level debt repayment and general corporate purposes, so "using new notes priced at 4.231 percent" converts a conditional into a stated mechanism the document does not assert; I confirmed the wording is identical on the issuer page and the syndication, so no second source exists to settle it. Second, the differential: 75.5 basis points sets a fresh 7.2-year coupon against a legacy coupon fixed in a different rate regime at a different original tenor with weeks to run, and by the principle this run's own C-04 states, that measures tenor, vintage and the level of the risk-free curve and isolates none of them. What survives is narrow and dated: the new 7.2-year coupon was 75.5 basis points below the legacy coupon on notes redeemed with under three months to maturity, the pair is a 2025 transaction outside the 2026 window it is used to characterise, and the record is one of three resting on a single press release.

Advocate · round 3 · REBUILT

On November 5, 2025 Capital Power priced C$600 million of medium term notes at 4.231% due January 14, 2033 — a 7.2-year instrument — and separately announced redemption of all C$300 million of its 4.986% notes due January 23, 2026, redeemed November 23 with the redemption price paid November 24, 2025, leaving under three months of residual life at pricing. WITHDRAWN: "no second source exists to settle it." A second source exists and settles the funding link in the past tense, and the record does not cite it. The pricing release is permissive — proceeds to "repay, redeem or refinance existing indebtedness, which may include fully funding the redemption of the Company's January 2026 Notes" — but Capital Power's 2025 fourth-quarter and year-end report of March 4, 2026 states that the company "used the net proceeds to repay, redeem and refinance existing indebtedness, which included fully funding the redemption of the Company's January 2026 Notes (as defined below), as well as project level debt at Goreway Power Station, Capital Power's credit facilities, and for general corporate purposes." The permissive enumeration was realised. The record's refusal to read funding into the pricing release alone remains correct as a reading of that release, and is the standard C-28 and C-35 should have applied; what changes is that the question is no longer open, and the record must carry the Q4 report as a second source. Unaffected: the 75.5 basis point gap sets a fresh 7.2-year coupon against a legacy coupon fixed in a different rate regime at a different original tenor and isolates none of tenor, vintage or the level of the risk-free curve; the pair is a 2025 transaction outside the 2026 window it is used to characterise; and this record rests on the same single press release as C-20 and C-46. [Capital Power 2025 Q4 report retrieved 2026-08-23.]