# Scoring worksheet — T2-TARIFF · the tariff regime, 2025-Q1 to present

Rubric version: v1.2 (weights 0.25 / 0.20 / 0.15 / 0.10 / 0.30; threshold 70; bands Genuine 0–30 · Mixed 30–70 · High-Asymmetry 70–85 · Extreme 85–100)
Epoch scored: **2026-Q3** (as of 2026-09-13)
Scorer: **Claude (Fable 5.1).** Ratification: **delegated by Briar Greenway on 2026-09-13 ("ratify if your review says it's needed") to Claude's own hostile-reader review; performed 2026-09-13.** Read this line as what it is: a Claude-scored, Claude-reviewed architecture whose entry the analyst authorised without independently scoring it. It is the weakest provenance in the register and it is stated rather than smoothed. Cross-analyst replication is the cure, and it has not happened.
Prediction locked: **2026-09-13 15:39 PDT by Briar Greenway — Mixed or lower** (text in `register/slate.json` → `locked_predictions.T2-TARIFF`)

Evidence discipline: every anchor carries a tag. `occurred` = documented event; `projected` = model or forecast; `contested` = disputed magnitude or characterization; `correlation` = measured relationship, no causal claim.

---

## S1 — Outcome-distribution asymmetry · weight 0.25

| | |
|---|---|
| **Proposed score** | **45** |
| **Anchors** | (a) Stated beneficiary: American workers and manufacturing; legal basis for the IEEPA tranche was a declared national emergency on fentanyl and the border (proclamations 2025-02-01 onward) [occurred]. (b) Realized incidence: New York Fed, 2026-02-12 — roughly 90 percent of the tariff burden fell on U.S. firms and consumers, about 10 percent on foreign exporters; January–August 2025 pass-through 94 percent; average tariff rate 2.6 → 13 percent across 2025 [occurred, third-party measurement]. (c) Yale Budget Lab one-year retrospective, 2026-04-02 — realized consumer price effect 0.5–1.0 percent against an original 2.3 percent estimate, with empirically observed pass-through 40–76 percent [occurred; the original estimate was projected and is superseded]. (d) $166–172 billion collected under IEEPA before the ruling (Holland & Knight cites $166B collected; Cato's refund arithmetic implies ~$172B owed) [occurred; range reported rather than a point]. (e) Electronics exemptions 2025-04-12 — smartphones, chips, computers carved out days after the reciprocal tranche, benefiting large electronics importers [occurred; corroborated across NPR, Fortune, Supply Chain Dive and CNBC headlines of that date, article bodies not opened in this pass]. (f) Refund distribution — Cato, 2026-07-09: approved refunds cover 30 percent of entries but 60 percent of obligations; most refunds so far to large importers such as Walmart and Ford; small importers largely unable to claim [occurred, third-party analysis of CBP figures]. (g) $12 billion Farmer Bridge Assistance, announced 2025-12-08, funded through the Commodity Credit Corporation, for "temporary trade market disruptions" [occurred]. |
| **Reasoning** | The category mismatch is real: stated as foreign-paid protection of workers; realized as a domestic consumption tax of which about nine-tenths landed on U.S. firms and households, with a compensation loop back to farmers hurt by retaliation. But the rubric's question is whether value or position was *routed* to an unstated class with a payoff. The routing that exists is partial: incumbents got exemptions and got refunded first. The bulk of the incidence is diffuse, fell on the administration's own constituencies, and generated a bailout rather than a capture. That is closer to symmetric self-harm than to a resource-concentration vector. Mid, not high. |
| **Reasons considered and rejected** | (1) Scoring S1 high on the strength of the who-pays gap alone — rejected: a gap between stated and realized incidence is necessary but not sufficient; the rubric wants concentration toward a beneficiary class, and the concentration here is modest. (2) Scoring S1 in the genuine band because incidence fell on the operator's own base — rejected: the exemption and refund pattern is a documented routing to large incumbents and cannot be zeroed out. |
| **Would a critic scoring a seed low be permitted this reasoning?** | **Yes.** "The harm was broad and symmetric" is the rubric's own symmetry test and is available on every architecture. The seeds scored high on S1 because their divergence was directional and concentrated; this one is neither, mostly. |

## S2 — Domain-selective competence · weight 0.20

| | |
|---|---|
| **Proposed score** | **58** |
| **Anchors** | (a) Extractive function: collection ran at scale — $166 billion under IEEPA, average rate quintupled in a year, no reported collection failure [occurred]. (b) Restitution function after the Court held the collection unlawful: CIT ordered universal refunds; CBP built the CAPE process; as of 2026-06-29, $104.29 billion authorized and $71.06 billion paid of roughly $172 billion owed, about 59 percent outstanding [occurred]. (c) 8,384 approved declarations stalled on banking-information grounds; $11.4 billion in finally liquidated entries ineligible without an importer lawsuit; a further $25 billion with no processing timeline [occurred]. (d) Refunds reached large firms first; small importers struggled to access the system [occurred, third-party]. (e) Exemption process ad hoc and announcement-driven (April 2025 electronics carve-out) rather than a functioning public-facing exclusion procedure [occurred]. |
| **Reasoning** | The signature shape is present: the same agency, the same months, one function surgical and the other degraded. Collection was flawless; giving the money back once a court said it was unlawfully taken is slow, litigation-gated for part of the population, and tilted toward the large. Against that: CAPE is an unprecedented mass-refund operation built in weeks, and 41 percent paid within about four months is not catastrophic failure. Part of the tilt toward large firms is that they have compliance staff, which is capacity, not selection. The selectivity is real; the capacity explanation covers some of it. |
| **Reasons considered and rejected** | (1) Scoring S2 low on the novelty of the refund operation — rejected in part: the seed IRS architecture was scored with capacity explanations rejected on precisely the ground that the same agency ran its extractive function cleanly, and CBP did too. The novelty argument is allowed to moderate, not to excuse. (2) Scoring S2 high on the refund appeal — rejected here: the appeal is an S3 fact (resisting a fix), not an S2 fact (selective competence), and counting it twice would inflate the composite. |
| **Would a critic scoring a seed low be permitted this reasoning?** | **Resolved on review, on a stated ground.** The capacity moderation is one the IRS seed refused — but the IRS case was an *existing* public-facing function (taxpayer service) degraded while enforcement against disfavoured targets sharpened. Here the public-facing function (mass refund of unlawfully collected duties) did not exist before the Court created the obligation; it is a new operation, not a degraded one. That is a real disanalogy, not a convenience, and it is why 58 stands rather than ~70. The pre-ruling exclusion process, which would be the true existing-function comparator, has too thin an anchor set in this pass to score on its own. Second-rater attention still belongs here first. |

## S3 — Anti-remediation response · weight 0.15

| | |
|---|---|
| **Proposed score** | **72** |
| **Anchors** | (a) S.J.Res.37 passed the Senate 51–48 on 2025-04-02 (Collins, McConnell, Murkowski, Paul with all Democrats) to terminate the Canada emergency; House Republican leadership inserted language into a spending bill declaring that no day for the rest of the Congress counts as a calendar day for the IEEPA fast-track clock, blocking any House vote in 2025 [occurred]. (b) The House passed the termination 219–211 on 2026-02-11 after the blocking rule lapsed and three Republicans refused to extend it; a veto was expected and is not confirmed in the sources located; the measure was mooted nine days later by the ruling [occurred for the vote; veto not established]. (c) *Learning Resources v. Trump*, 2026-02-20: IEEPA does not authorize tariffs [occurred]. Response: three proclamations 2026-07-20 re-imposing 50 percent on Canada under §338 of the Tariff Act of 1930, effective 2026-08-22, and a stated global rate of 15 percent [occurred]. (d) CIT ordered refunds to all importers of record; DOJ appealed to the Federal Circuit on 2026-06-03, arguing relief must be confined to importers who sued, so that finally liquidated entries require litigation before CBP will pay [occurred]. |
| **Reasoning** | The construct reads the slope of resistance against the cost of the fix. The three cheapest fixes available in the whole period were: let the House vote; accept a Supreme Court ruling; return money a court held was taken without authority. Each was resisted — procedurally, by statutory substitution, and by appeal — with the same force as any expensive alternative. That is the inverted slope. |
| **Reasons considered and rejected** | (1) "Re-imposing under a different statute after losing is ordinary executive behavior" — partly accepted: it is ordinary, which is why this is 72 and not the low 80s the anchors would otherwise carry. (2) "The appeal is a legitimate legal position on universal injunctions" — considered: the government's Article III argument is a real one and high-quality contestation on *scope*; it does not change that the effect is to make small importers litigate to recover money the Court said was unlawfully collected. Logged, weight moderated, not zeroed. |
| **Would a critic scoring a seed low be permitted this reasoning?** | **Yes.** Both moderations above are ones the seed worksheets also allow (ordinary-executive-behavior and good-faith-legal-position discounts appear in the IRS and ICE notes). |

## S4 — Personnel selection asymmetry · weight 0.10

| | |
|---|---|
| **Proposed score** | **20** |
| **Anchors** | (a) Early-2026 reshuffle: Treasury Secretary Bessent emerged as the lead on trade; Navarro relegated to an advisory role without formal authority; Lutnick recast as "bad cop" with his media appearances restricted by the Chief of Staff [occurred, reporting]. (b) Open Musk–Navarro feud in April 2025 [occurred]. (c) No documented purge, exodus, or loyalty-screening of career trade staff at USTR or Commerce was located in this pass. |
| **Reasoning** | On the rubric's three axes the record points the wrong way for the signature: technical competence (Bessent) was advanced over ideological alignment (Navarro), and dissent was tolerated to the point of public feuding. Low. |
| **Reasons considered and rejected** | Leaving S4 unscored for thin evidence — rejected: the reshuffle reporting is a real anchor and it points low; unscored would be less informative than 20. |
| **Would a critic scoring a seed low be permitted this reasoning?** | **Yes.** Same three-axis reading as every seed. |

## S5 — Information-architecture asymmetry · weight 0.30

| | |
|---|---|
| **Proposed score** | **40** |
| **Anchors** | External incoherence, high: (a) Lutnick 2025-04-01 on CNN, "no chance" the President backs off; within hours the President, with Lutnick present, said he would negotiate; that Sunday Lutnick on CBS, "there is no postponing"; the 90-day pause followed [occurred]. (b) Bessent's "short-term pain" framing beside "that's a decision for the President" on permanence [occurred]. (c) The sustained public framing that foreign countries pay, beside the Treasury's own collection of $166 billion from domestic importers and a Federal Reserve finding of ~90 percent domestic incidence [occurred; the framing mismatch is documented]. Internal coherence, low: (d) reporting of a genuine tug-of-war between a fair-trade camp and a protectionist camp, a team of rivals, and a reshuffle that sidelined one camp [occurred, reporting]. (e) The Mandate for Leadership trade chapter itself contains two competing essays — Navarro, "The Case for Fair Trade," and Lassman, "The Case for Free Trade" — so the internal-coherence anchor the seed S5 scores rest on (a published, coordinated plan) points the other way for this architecture [occurred]. |
| **Reasoning** | S5 is the ratio of internal coherence to external coherence. External coherence is very low. But internal coherence is also low: the record shows real disagreement, not a coordinated plan behind deliberate fog. The book's own line applies — "a merely confused architecture is incoherent on both sides at once." What keeps this off the floor is (c): the who-pays framing was maintained from the top for over a year against the operation's own receipts, which is a framing mismatch, not a reversal. Mid-low. |
| **Reasons considered and rejected** | (1) Scoring S5 high on volatility alone — rejected per the locked discriminator: volatility from genuine internal disagreement scores low. (2) Scoring S5 in the genuine band — rejected: the sustained who-pays framing against the Treasury's own data is a documented external-framing asymmetry and is not explained by internal disagreement. |
| **Would a critic scoring a seed low be permitted this reasoning?** | **Yes, and this is the cleanest row.** The seeds' S5 anchors on Project 2025 as a coordinated internal plan denied in public. On trade, that same document is internally split. The same evidence standard, applied to this case, lowers the score. No allowance was invented here that the seeds were refused. |

---

## Composite

Index = 0.25·45 + 0.20·58 + 0.15·72 + 0.10·20 + 0.30·40 = 11.25 + 11.60 + 10.80 + 2.00 + 12.00 = **47.65 → 48**

**Band: Mixed.**

**Predicted band:** Mixed or lower · **Returned band:** Mixed · **Match: yes.**

Per Part VIII: the live domain now contains an architecture the instrument declined to convict. Seven seeds above 70 and one live case at 48. The instrument discriminated where discrimination was available.

What the lock said would happen, happened: S3 came in high (72), S5 came in low (40) on the discriminator, S1 came in low-mid (45). The composite sits in the middle of the Mixed band, not at its edge. The robustness claim available here is arithmetic rather than inter-rater, since no cross-analyst study has been run and no observed rater range exists to cite: reaching the High band from 47.65 requires +22.35 weighted points, and the largest headroom any single dimension carries is S5's 18.0 (0.30 x 60), so no single dimension moved to 100 crosses the line; symmetrically, reaching Genuine requires -17.65 and the largest single-dimension floor room is again S5's 12.0, so no single dimension moved to 0 crosses it either. Two dimensions must move together, in concert, to change the band. That is the property that makes this reading reportable rather than fragile.

## Double-standard audit

Reasons used to *lower* a score here, checked against what the seed worksheets permit:

| Reasoning used here | Dimension | Permitted on seeds? | Note |
|---|---|---|---|
| Harm broad and symmetric, fell on own base | S1 | Yes — rubric symmetry test | clean |
| Modest concentration, mostly diffuse incidence | S1 | Yes | clean |
| Refund operation novel; capacity explains part | S2 | Refused on IRS seed for a *degraded existing* function | resolved: this is a *new* function created by court order, a stated disanalogy, not a double standard |
| Statutory substitution is ordinary executive behavior | S3 | Yes — IRS/ICE notes allow it | clean |
| Government's universal-injunction argument is a real legal position | S3 | Yes — good-faith-legal-position discount | clean |
| Competence advanced over alignment; dissent tolerated | S4 | Yes — same three axes | clean |
| Volatility from genuine internal disagreement | S5 | Yes — and the seeds' own internal-coherence anchor (Project 2025) is split on trade | clean, principled |

The S2 row was the one open entry; the review resolved it on the new-function disanalogy above. If a second rater refuses that disanalogy, S2 rises to roughly 70 and the composite to **50** — still Mixed. The band does not depend on the row.

Reasons used to *raise* a score here that a critic might call a stretch: none identified. The S3 anchors are all documented events; the 72 is below what the anchors alone would carry.

## Review pass (hostile reader), 2026-09-13

Run before ratification, against the question a critic asks first: *where did this scorer let the case off?*

1. **S3(b), the veto.** Not confirmed by any source located; the House measure was mooted nine days later by the Court. The anchor was already tagged "veto not established" and carries no weight in the 72. Left as is.
2. **S1(d), the collection figure.** Two sources imply different totals ($166B vs ~$172B). Changed to a range.
3. **S1(e), the exemptions.** Headline-corroborated across four outlets; bodies not opened. Stated on the anchor.
4. **S2, the open row.** Resolved on the existing-function vs new-function disanalogy rather than left flagged. Score unchanged at 58.
5. **S5, the who-pays framing.** A critic from the other side would say a sustained top-down claim contradicted by the Treasury's own receipts is exactly the S5 signature and should push higher than 40. Considered. The dimension is a *ratio*, and most of the external incoherence traces to documented internal disagreement; the who-pays element is what holds the score at 40 rather than the mid-20s a purely-confused reading would give. Unchanged.
6. **Epoch.** 2026-Q3 is a single-architecture checkpoint; the seven seeds were last scored at 2026-Q1. The breadth table on the register page keeps its three slate-wide epochs and does not treat Q3 as a slate checkpoint.

No score moved. The composite is **48, Mixed**, and the review found no reasoning used to lower it that the seed worksheets would refuse once the S2 disanalogy is stated.

## What this does and does not establish

Established: the instrument, as specified, returns Mixed on a live, current, high-salience administration action. The ceiling-slamming charge in Part VIII is answered with a number inside the live domain.

Not established: that the instrument is unbiased. One acquittal is one data point. The convict-a-friend test (T1) remains unrun, and cross-analyst replication remains the condition of validity. Both are recorded on the register as open.

Not scored: the resource-operations layer (S6–S8). The $12 billion farm payment and the tariff revenue as a political vehicle (the $2,000 dividend, then the $5,000 promise) are candidate S6 material and are held outside the composite per Part V.

## Amendment, 2026-09-21 — S2's stated ground is contradicted; score unchanged pending the analyst

Per §8 of `PRE-REGISTRATION.md`, a dated amendment with a stated reason. **Affects a locked
prediction: no** — T2's prediction is closed and its band is unaffected. **Changes a score: no.**

The second evidence pass this worksheet asked for ("second-rater attention still belongs here first")
was run on 2026-09-21. It found the comparator this worksheet said it lacked. Commerce stopped
processing Section 232 exclusion requests effective 2025-02-10, revoked all General Approved Exclusions
and country-level arrangements effective 2025-03-12, and replaced the mechanism in April 2025 with an
*inclusions* process that only lets applicants add products to tariff coverage. Quotations are from the
administering agency's own page.

That is an existing public-facing relief function **terminated outright** — a stronger form of the
signature than the IRS seed, where the function was degraded rather than abolished — beside a
collection function this worksheet records as flawless. **The ground on which S2 was held at 58 (that
no existing-function comparator was available, only the court-created refund operation) does not
survive it.**

S2 is left at 58 here because scores are the analyst's and this worksheet is a dated record, not a
live document. The band is unaffected at any value of S2: weight 0.20 moves the composite at most 8.4
points against the 22.35 needed for High-Asymmetry, so S2 at 70 gives 50, at 85 gives 53, at 100 gives
56 — Mixed throughout. The double-standard audit row reading *"resolved: this is a new function created
by court order"* should be reopened.

Detail and anchors: `T2-evidence-addendum-2026-09-21.md` (machine-proposed, `ratified: false`).

## Sources

New York Fed Liberty Street Economics, 2026-02-12 · Yale Budget Lab one-year retrospective, 2026-04-02 · Holland & Knight, IEEPA refund appeal, 2026-06 · Cato, refunds update, 2026-07-09 · EPI on S.J.Res.37, 2025-04-02 · CFR on the House vote, 2026-02-11 · Justia, *Learning Resources v. Trump* · Blakes US–Canada tariff timeline · NPR, mixed messages, 2025-04-08 · Yahoo/Politico, trade reshuffle, 2026 · USDA, Farmer Bridge Assistance, 2025-12-08 · Fortune/CNBC/NPR, electronics exemptions, 2025-04-12 · Econbrowser on the Mandate for Leadership trade chapter. Full URLs in `annex/evidence-annex.json` (AX-007, AX-008) and `register/architectures.json` (T2-TARIFF anchors).