The Portable Kit

A Five-Stage Audit and Transfer of the Leadership-Evaluation Framework

Inputs: the reference document (Was Donald Trump a Good Leader, 9 Aug 2026, 756 sources), the teaching companion (How to Judge a President), the underlying research files, and the scoring arithmetic (base weighted score 3.245 → reported 3.2). All new computations in this document were run and verified against that arithmetic.

Standing rules applied throughout: events over attributions; "unestablished" is an available verdict; when two credible numbers conflict, the definitional difference is surfaced before either is used.

Stage 1: Compression and Portability

Task: the minimal toolset that still works in 2035, under different institutions and technologies, ranked by expected value (EV) for reducing systematic error, where EV ≈ (frequency of the error the tool prevents) × (size of the error) × (probability an unaided evaluator commits it).

The ranked kit (ten tools)

1. Baseline discipline

• Operational form: no verdict on any governing statistic until you can state the inherited trend, the cycle position, and the counterfactual benchmark.
• Why ranked first: trend-attribution is the most frequent error in all leadership evaluation, committed by both directions with near-certainty when unaided; it single-handedly generated the illusory halves of this record (unemployment records that were decelerating trend-continuation; NATO spending that inflected in 2014).
• 2035-robust: fully, nothing about it is institutional.

2. Pre-committed rubric plus sensitivity analysis

• Form: declare criteria, weights, and a benchmark before scoring; then publish how far any coherent reweighting moves the verdict.
• Why second: verdict-first reasoning is the paradigm case of motivated cognition, and this is the only tool that makes it visible, the decisive finding here was that reweighting alone moves 3.2 only to 2.9–3.6 because all seven scores sit below the benchmark.
• 2035-robust: fully.

3. Enforcement tracing

• Form: for every rule, name the enforcer; for every enforcer, name their incentive; a rule whose chain terminates in the goodwill of the constrained party is a norm, not a rule.
• Why third: least intuitive tool with the highest institutional payoff (the GAO-finds → no-one-can-sue → Congress-abstains chain is invisible without it), and rising in importance toward 2035 as algorithmic and platform governance multiply rules-without-enforcers.

4. Own-goals test

• Form: first-pass grading against the leader's own stated objectives, quantified at announcement time.
• Why fourth: it de-ideologizes evaluation more cheaply than any other move (the tariff verdict requires no position on protectionism).

5. After-the-loss test

• Form: the most diagnostic window on any power-holder opens in the two weeks after an adverse ruling, vote, or election; classify the response as accept / appeal / reroute / defy.
• Evidence base here: three tariff regimes in eighteen months; a second birthright-citizenship order six weeks after the first was struck.
• 2035-robust: fully, losses are universal.

6. Definitional-difference-first

• Form: when credible sources publish conflicting numbers, locate the definition gap (5.9% vs 11.1% tariff rate = realized-vs-statutory; Atlanta Fed vs EPI wages = same-person panel vs fixed-percentile) before adopting either.
• Rising EV toward 2035 as synthetic content makes number-laundering cheaper.

7. Durability sort (ink / pencil / compounding)

• Form: classify every claimed achievement as statute-or-appointment (ink), decree (pencil), or self-growing liability (compounding: debt interest, verification gaps, systemic distrust); legacy = ink + compounding, with sign.

8. Unreviewable-powers reading

• Form: identify the discretion corners no institution can check (pardons, prerogative powers, founder super-voting shares) and weight conduct there heavily as a character measurement, because it is unconfounded by constraint.
• 2035-robust: every system keeps such corners.

9. Convergence-and-category calibration

• Form: confidence tracks (a) event vs attribution, (b) independence of converging methods, (c) whether a counterfactual is required; claims failing (c) get "unestablished," said aloud.
• Worked examples: tariff incidence earns high confidence (five independent methods converge); a Trump-specific polarization increment earns "unestablished" (no counterfactual exists).

10. Shared-premise critics as the highest-probative evidence class

• Form: seek the people who hold the leader's own premises and still object (the unitary-executive scholar against the unitary executive's use; the Federalist Society prosecutors who resigned).

Discarded as non-portable

• The specific 18/15/20/10/10/17/10 weights (context-tuned).
• The 5.0 = "average of Reagan–Biden" benchmark (cohort-specific, see Stage 3, Finding 5).
• Named instruments (V-Dem, Gallup), portable form is "perception thermometers vs event ledgers."
• The mandate-as-constraint decision (portable is the obligation to decide it explicitly, not the answer).
• All US-specific enforcement plumbing (portable form is tool 3 itself).

Stage 2: Stress-Testing the Framework

Method. Hold the documented facts fixed. For each criterion, state the strongest coherent alternative definition a sophisticated defender or critic could adopt, who actually holds it, and the verdict movement it buys (Δ = weight × score change). Then separate two channels: Channel A ("the numbers changed"), pending evidence resolving one way or the other; Channel B ("the interpretation changed"), definitions or weights moving with facts fixed. All arithmetic machine-verified.

2.1 Per-criterion alternatives

2.2 The three lenses, isolated

• Outcomes-only lens (econ 45 / natsec 25 / crisis 30): ≈ 4.1
• Stewardship-only lens (institutions 50 / long-term 35 / character 15): ≈ 2.4
• Mandate-only lens: 7–8 (two victories, a non-consecutive comeback, coalition realignment, but this lens answers "was he chosen?", a different question)

2.3 The two channels, quantified

Reading: the evidence channel moves the verdict by at most ±0.5; the interpretation channel by at most ±1.3; and even the full favourable stack (every pending fact resolving his way, the strongest coherent redefinitions of the two lowest criteria, and the mandate converted from a constraint into a heavily-weighted criterion) lands at 4.99, a hair under the average-president line. This independently reproduces, by a different route, the reference document's 5.2 corner (which additionally reweighted).

2.4 Two additional stress tests the document did not run

First-term-only

Objection: the verdict is dragged by nineteen unresolved months. Rescoring on first-term evidence alone (fresh estimates, each ±0.5: econ 5.5, natsec 4.0, institutions 3.5, appointments ≈7 at one-third weight, 2020-election conduct ≈2 at two-thirds, crisis 4.0, comm 4.0, long-term 4.0, character 3.0) yields ≈ 4.1, corridor roughly 3.6–4.6.

Mandate-as-criterion

Converting the framework's most contested exclusion into a heavily-weighted inclusion (15%, scored 7.5) moves the base verdict from 3.25 to 3.88. Even the strongest structural concession to the defence is worth about +0.6.

Stage 3: Methodological Audit

The document's standards, applied to the document. Ordered by severity. For each: the standard, the strain, and whether the verdict survives the correction.

Finding 1: "Repeatedly obeyed" overstates the document's own compliance evidence

Severity: highest. The Executive Summary states courts "repeatedly ruled against him and were repeatedly obeyed." The document's own §3.4.2 records a judicial finding of "willful and intentional noncompliance" (Abrego Garcia), a probable-cause contempt finding, and a 57-of-165 accusation count. The body holds the tension carefully; the summary resolves it in the defence's favour. This is the document committing a mild version of the error it names: treating formal compliance in the marquee cases as general fidelity. Verdict survives (the institutional score already prices this), but the summary sentence should read "and, in the highest-stakes cases, obeyed."

Finding 2: Perception-index salience contradicts the stated evidence hierarchy

Severity: high. The methodology demotes V-Dem-class indices to "thermometers of expert assessment" to be paired with hard events; yet the V-Dem reclassification appears in the Executive Summary's driver paragraph, the document's most-read sentences, with marquee placement. The companion repeats the pattern (Bright Line Watch's Canada-88/US-57 in its Canadian section). The stated hierarchy and the rhetorical hierarchy diverge. Correction: summary placement should be reserved for the event ledger; indices belong in the supporting tier. Verdict survives, the document itself demonstrates (§3.4.2) that the institutional case stands with the indices struck.

Finding 3: Deflator inconsistency in the median-household indictment

Severity: moderate, and instructive. The document attributes the 2026 headline-inflation spike to an exogenous oil shock (correctly), then deflates second-term real wages by headline CPI (+0.80% real over 17 months) in the affordability indictment. Consistency requires a core-deflated check. Running it: headline CPI over the window annualizes to ≈3.0%, core PCE to ≈3.2%, the gap nets out to roughly zero over this particular window, so the indictment survives the check empirically. But the document never ran the check, and under a slightly different window it could have flipped the sub-finding. This is the cleanest example of "the interpretation was fine, the hygiene was not."

Finding 4: Endpoint choice inflates a headline fiscal number

Severity: moderate. "Deficit rose 68 percent" uses FY2016 → FY2019. A defender choosing FY2017 (the last budget substantially Obama's) gets +48 percent. Both are defensible; the document picks the larger without flagging the choice, while elsewhere (fiscal-year attribution, −6.55% vs −8.63%) it does flag convention-dependence. The %-of-GDP framing (3.1→4.6) is endpoint-robust and should carry the claim.

Finding 5: Asymmetric circumstance adjustment

Severity: moderate, conceptually the deepest. The framework circumstance-adjusts the economy meticulously (91-month expansion, inheritance table) but applies no equivalent adjustment to the institutional criterion, where a defender would argue the environment was also unprecedented (record emergency-docket volume, nationwide injunctions, two assassination attempts, prosecutions later dismissed). The document's implicit answer, much of that environment was self-generated, is defensible but asserted rather than argued, and the cohort benchmark ("average of Reagan–Biden") embeds the same asymmetry: it adjusts economic starting conditions across the cohort but not institutional stress. A rigorous v3 would either adjust both or explicitly defend adjusting only one.

Finding 6: A latent double-count in the fiscal indictment

Severity: moderate. §2.3 lowers national security's weight partly to avoid double-counting with long-term positioning, but deficits and debt appear as negatives in both the economy criterion (§3.1.7) and long-term positioning (§3.8), with no equivalent disclosure. Rough magnitude: if fiscal deterioration were charged only once, the economy score plausibly rises ~0.5, moving the verdict ≈ +0.09. Small, but the inconsistency is real.

Finding 7: The own-goals verdict is time-sensitive and one sub-goal has since moved

Severity: low-moderate. "Tariffs failed on their own terms" bundles three goals. Manufacturing employment (failed, twice) and incidence (failed, high confidence) are stable. But the trade deficit narrowed materially in 2026 ($932B → ≈$728B twelve-month), which the document reports and then under-weights in the summary formulation. Honest form: "failed two of three stated goals; the third is mixed and recent." Verdict effect: negligible; framing effect: real.

Finding 8: The companion promotes n = 1 inductions to general signatures

Severity: moderate for the companion, nil for the document. The "five-tell cluster" for communicator-without-steward, and the "$0 investments page" as a diagnostic, were induced from a single case; the page is a website artifact, colorful, evidentially thin. Teaching artifacts compress; compression became overclaim. Fix: label induced heuristics with their n and their disconfirming test (are there high-decree, high-churn leaders who were durable stewards? Wartime cases are candidates).

Shared-premise critics who would still dissent from the corrected document

• Deficit hawks would press Finding 6.
• The originalists who won the tariff case would reject the "facts about courts" discount, arguing that litigating in good faith and complying is what constitutional fidelity looks like from the inside (they would score institutions nearer 3 than 2).
• The C-SPAN-postponement logic, that contemporaneous expert judgment of a sitting president is punditry, indicts the document's own timing, a point it acknowledges (§6.4) but does not fully absorb.

Stage 4: Transfer and Generation

4.1 The template: a concentrated-power CEO (one page)

Same architecture, three substitutions: the benchmark cohort becomes peer large-cap founder/CEOs; the enforcer map becomes board → shareholders → courts (Delaware/securities) → regulators; the mandate analogue becomes shareholder votes (with the same rule, ratification legitimizes, it does not grade).

Weights justification mirrors the original: governance = the internal controls only the CEO can maintain; durability heavy because compounding liabilities (leverage, key-person concentration) outlast tenure; outcomes capped at 20% because sector beta dominates raw returns.

4.2 Brief application: Elon Musk at Tesla

The reasoning chain, tools in brackets:

• Baseline: Tesla's returns and its 2024 delivery decline (~1.79M, the first annual drop, with BYD overtaking on BEV volume in late 2023) mean nothing raw; graded against the EV sector's simultaneous demand slowdown, execution reads mid-pack rather than catastrophic, a genuine moderation of the bear case, exactly as baseline discipline moderated the Trump-economy bear case.
• Own-goals, Timestamped targets, one million robotaxis by 2020 (announced 2019, delivered zero by early 2025), and the 20-million-vehicles-by-2030 goal reportedly dropped from the 2024 impact report (moderate confidence; verify), fail the own-goals test on the announced schedule, independent of any view about the technology.
• After-the-loss, the isomorph, The Delaware Court of Chancery rescinded the $55.8B pay package (January 2024); the response was not acceptance or mere appeal but rerouting: a shareholder re-ratification, reincorporation from Delaware to Texas, and a public "be careful where you incorporate" posture, structurally identical to the tariff sequence (lose on the instrument, rebuild the same end under a friendlier authority). The court declined to reinstate despite ratification (December 2024); appeal pending as of the knowledge window.
• Enforcement tracing: The enforcer chain has a documented weak link: directors settled excess-compensation claims by returning ~$735M (2023), and the 2018 SEC consent decree's supervision terms produced years of contested compliance, detection robust, consequence intermittent.
• Unreviewable corners: Conduct on a personally-owned platform, unconstrained by any board, is the pure character read the template asks for.
• Durability sort: Ink: manufacturing scale, Supercharger network, energy business. Pencil: valuation premised on announced-not-delivered autonomy. Compounding: key-person concentration, explicitly priced by the principal himself ("uncomfortable growing Tesla in AI/robotics without ~25% voting control," January 2024).

4.3 The Canadian-PM variant

Needs only substitutions, not redesign: benchmark = post-1980 PMs circumstance-adjusted; enforcer map = confidence convention, PBO, Auditor General, courts under the Charter, the Crown's reserve powers; mandate analogue = seat count vs vote share (a minority PM's "mandate" discount is the exact analogue of the plurality/majority distinction); unreviewable corners = prorogation advice and appointments. Building and scoring it is the retention exercise.

Stage 5: Self-Improvement Loop

Limitation 1: Exogeneity rulings are not automatically propagated

The oil shock was ruled exogenous for inflation-blame, then a headline-CPI deflator was used for real wages three subsections later (Stage 3, Finding 3). The pattern: a judgment made in one cell of an analysis does not automatically recompute the other cells it touches.

• Adjustment: whenever a factor is classified as exogenous/endogenous, mechanically list every other computation in the document that factor enters, and either re-run each under the classification or state why it is immune. This is a checklist behaviour, not an insight, which is exactly why it was missed.

Limitation 2: Salience allocation drifts from the stated evidence hierarchy

The methodology ranked events above perception indices, then gave an index marquee placement in the summary, and repeated the pattern in the companion (Stage 3, Findings 1–2). Summaries are where motivated reasoning hides after the body has been disciplined, because summary-writing feels like compression rather than inference.

• Adjustment: audit executive summaries as claims, not as prose, every sentence in a summary must cite the evidentiary tier of its strongest support, and only top-tier items get driver-paragraph placement.

Limitation 3: Single-case inductions promoted to general signatures without labelling the n

The "five-tell cluster" and the corridor heuristics were built from one presidency and shipped as portable tools without their disconfirmation tests attached (Stage 3, Finding 8; Stage 4's Musk case partially, but only partially, replicates one tool).

• Adjustment: every induced heuristic ships with three fields: n, the nearest disconfirming case searched for, and the observation that would retire it. Where no search has happened, the heuristic is labelled "candidate," not "signature."

The additional hardening heuristic

Related notes

• Was Donald Trump a Good Leader, the 24,000-word reference document this kit was induced from
• How to Judge a President, the teaching companion (nine lessons, Canadian lens)
• Baseline discipline · Pre-committed rubric · Enforcement tracing · Own-goals test · After-the-loss test
• Definitional-difference-first · Durability sort · Unreviewable-powers reading · Convergence-and-category calibration · Shared-premise critics
• Flip-list · Actor-swap
• Tornetta v Musk, the corporate isomorph of the rerouting pattern
• Presidential greatness surveys, methodology, partisan-composition problem and the Republican-respondent breakout
• Motivated reasoning, countermeasures, where the flip-list and actor-swap belong in a broader toolkit