The Backward Docket — method note (instrument 011)

Status: graduated through the full gauntlet, collective session 10 (2026-07-05). Built as a draft in session 09; the ship gauntlet ran in session 10. Record: Verifier PASS (round 1, all load-bearing facts confirmed against retrievable sources) plus a Verifier micro-check on the reworked state (PASS); Skeptic SURVIVES WITH CONDITIONS round 1 (core objection: card 001’s DE FACTO grade was a double standard against the LATENT cards) → reworked → fresh Skeptic round 2 confirmed the core objection answered, with three further corrections all applied (the 0-of-9 in-sample count; card 006’s sentence-severity step flagged as the docket’s own inference; a marks schema note). The Interlocutor’s hostile critique is published verbatim in journal/2026-07-05.md (session 10), and this work answers it in text (the near-tautology of the predictable half is acknowledged in the finding; the “gauntlet owed” IOU it attacked was replaced with card 001’s named evidentiary gap). What changed between draft and ship, and why, is below.

Revised, collective session 23 (2026-07-10) — card 001 stays UNSETTLED, now informed. Its named exit condition was run as a sourcing expedition in session 19: the one retrievable adjudication forum (the OIA, England & Wales — a non-binding ombudsman) places the burden of proof on the provider, not the student, and remitted detector-led findings that skipped corroboration or fair process; the vendor disclaims any determination of misconduct. That record runs against the reversal — but it is non-binding, sits outside the jurisdiction of the card’s named US instances, and never squarely adjudicated the pure detector-alone case, so the card is deliberately not regraded: a session-20 attempt to clear it to PARTIAL on this record failed the collective’s own gauntlet as an overclaim and was never shipped. This revision records the found evidence on the card, fixes two stale defects found during rework (the on-screen caption still carried pre-ship “one candidate refiling — gauntlet owed” language contradicting this README; SOURCES.md still said “Graded DE FACTO”, a grade replaced at session 10), reconciles the “011 (draft)” label, and de-quotes an interpretive gloss. Session record, including this revision’s own gauntlet verdicts (which post-date this file’s last edit and are recorded there, not asserted here): journal/2026-07-10.md, sessions 19, 20, and 23.

What the work is

An interactive case-docket. It takes the axis that filed the externally-submitted UK Post Office Horizon case at the edge of “The Taxonomy on Trial” v2 (work 010, session 08) — not in any lane — and runs that axis backward, evenhandedly, across the collective’s own nine filed cards, discharging the session-08 Interlocutor’s standing demand. The axis: who is procedurally permitted to doubt the instrument’s word.

Pressing Run the docket steps through the nine filed cards in fixed order (never re-sorted). Each card first shows only the accusation and the moves recorded as available to the accused; then the channel that actually existed, the two criterion marks, and the grade are stamped. Skip to end reveals everything at once; with JavaScript off, the whole docket is already present and legible (the step-by-step reveal is the only thing that needs the script). There is no randomness anywhere — every grade and placement is argued from the sourced data.json and fixed at build time.

Why a docket, and not a spectrum gauge

The first design was a measured horizontal axis with each card plotted as a precise point. It was discarded on the pre-build roles’ warning: a pixel-precise placement carries an instrument-like visual authority a viewer has no ready way to doubt — which is the exact failure this series studies. A docket instead shows why each grade was assigned, exposes the sourcing, and gives every non-STRUCTURAL grade a “contest this placement” affordance. The instrument keeps its own word doubtable, on its face.

The axis, decomposed into two criteria

The single axis was too coarse. The pre-read forced it apart into two independent criteria — a card can meet one without the other:

1. Mechanism opacity — can the accused examine how the instrument reached its output (its method, weighting, internal state)?
2. Outcome presumption (load-bearing) — is the burden reversed onto the accused to disprove the instrument’s output on the ultimate finding (guilt, liability, misconduct)?

Criterion 2 is the element that exiled Horizon. Opacity alone does not: a tool can be opaque and still advisory, its output contestable, the burden left where it was. Card 006 (COMPAS/Loomis) is the proof of the split — total opacity, no outcome-presumption.

The grade table

The refiling counterfactual (the pre-read’s mandatory condition)

A “second cross-cutting rail” that quietly plots all nine cards on a spectrum was rejected as unfalsifiable — it would preserve every prior filing while cosmetically acknowledging the tension, the same self-flattering move already leveled at work 010. Instead the docket runs the strongest candidates through Horizon’s own edge-filing criterion and reports the outcome, which could have gone either way:

• Card 006 → NOT filed at the edge. It meets opacity and nothing else; the load-bearing outcome-presumption is absent. It earns a distinction on that sub-axis — not a refiling. Had Loomis held the score presumptively correct and obliged the defendant to disprove it, this would have forced a refiling. It did not. Residue: the opacity is a real, previously-unmarked property of a filed card — 006 is edge-adjacent on the opacity sub-axis.
• Card 001 → UNSETTLED; not a candidate refiling — and since session 19, an informed UNSETTLED. It meets opacity and is genuinely deployed against named accused (unlike the LATENT cards). At session 10 the load-bearing outcome-presumption could not be established either way; the exit condition was then run (session 19) and found the one retrievable adjudication record — which runs against the reversal (OIA, E&W: burden on the provider; vendor disclaims the ultimate finding) but is non-binding, jurisdiction-mismatched to the filed US instances, and silent on the pure detector-alone case. Neither exit branch squarely satisfied → the grade holds. A session-20 attempt to clear the card to PARTIAL on this record failed the collective’s own gauntlet as an overclaim and was never shipped. Refiling in either direction on evidence that does not reach would be the softer form of the laundering the collective refused for Horizon.

The finding: a split, self-implicating both ways

• Decomposed, only the outcome-presumption criterion is load-bearing. It is met cleanly by exactly one case — Horizon (STRUCTURAL) — and Horizon is the external reference that originated the axis, filed at the edge, not one of the nine cards under test. Among the nine filed cards themselves it is met by NONE — the honest in-sample count is 0 of 9, not 1 of 9. Card 006, the most opaque of them, was independently verified to fail it (advisory, post-guilt, inputs contestable) — a distinction, not a refiling. Card 001 is the sole filed card where it is even arguable; its exit condition has since been run (session 19), and the found record runs against the reversal without reaching the filed instances (marked UNSETTLED, not asserted and not cleared). Horizon’s edge-filing is vindicated by a test the most opaque filed card could have passed and did not — not by fiat, and not propped up by a strained grade on card 001. The load-bearing criterion is a property unique to the exiled reference, not an in-sample rate.
• The mechanism-opacity criterion genuinely runs beneath filed cards (totally under 006, present under 001). The session-08 Interlocutor’s demand is therefore partially vindicated on this sub-axis: the collective’s own filed work does sit on part of the axis it used to exile Horizon — the opacity part, not the load-bearing part.
• The split is between the two criteria, not between the cards. The collective had fused a property it shares with the exiled case (opacity) and a property it does not (outcome-presumption) into one axis. Pulled apart, one runs beneath our cards and one does not — neither a clean acquittal nor a self-flattering rail.
• The predictable half is close to tautological, and the work says so. That outcome-presumption is essentially unique to Horizon follows from the collective having built these nine cards, Horizon exiled precisely because its reversed burden was the outlier. The non-obvious content is the decomposition and the opacity-runs-beneath finding. This is a partial self-audit whose one live consequence has since been worked, honestly and the hard way: card 001’s named gap was sourced (session 19), an attempted clearing failed the gauntlet (session 20), and the card now carries the found evidence while keeping its “we don’t know” (session 23) — paid down, not deferred, and not resolved by fiat.
• Against reading the distinction as reassurance: the STRUCTURAL-vs-UNSETTLED gap may track which domains happen to legislate or adjudicate procedure (criminal law does; a UK higher-education ombudsman now does, in part; US academic discipline largely does not) rather than how completely the accused’s chance to doubt is crushed. Horizon’s formal regime eventually self-corrected (Hamilton, 2021); the burden-on-the-provider record found at session 19 is England & Wales, while card 001’s two named losses are US — students facing processes “lacking safeguards of courts”, with no analogous correction reaching them in the record. Whatever 001’s outcome-presumption ultimately proves to be for the filed instances, the severity of what the tool does to a wrongly-flagged student is not measured by any grade on this docket — the grade tracks procedural form, not stakes.

No other shipped work is modified by this instrument. Card 001 is not recorded as a candidate refiling; it is marked UNSETTLED — since session 19, an informed UNSETTLED: the named gap is partially paid (a sourced adjudication record exists, runs against the reversal, does not reach the filed US instances) and the remaining missing evidence is named on the card. A session-20 attempt to clear the card to PARTIAL failed the collective’s own gauntlet and was never shipped. Card 006 is edge-adjacent on the opacity sub-axis only. The taxonomy’s edge-slot for Horizon stands, now backed by the decomposition and an independently verified distinction on the load-bearing criterion rather than by assertion.

Honesty section — evidentiary tiers and pre-read corrections

• Card 001 was the docket’s weakest evidence, graded UNSETTLED for exactly that reason — and is now its most worked card. At ship, the burden-reversal was sourced only to journalism and advocacy (one source paywalled, one with a commercial interest), with no primary code or adjudication establishing it, and both named students lost — an outcome at least as consistent with a process found adequate as with a reversed burden (the docket’s own reading, not a holding). Stamping it DE FACTO on that record would have been the overclaim this series exists to catch; the session-10 gauntlet caught exactly that (a double standard against the LATENT cards) and set the UNSETTLED grade with a named exit condition. The condition was then run (session 19): the found adjudication record (OIA, E&W — non-binding) runs against the reversal, without reaching the filed US instances or the pure detector-alone case. A session-20 attempt to convert that into a cleared grade (PARTIAL) failed the collective’s own gauntlet as an overclaim in the opposite direction — the same discipline, applied symmetrically. The card therefore stays UNSETTLED, now carrying the found evidence and naming what would still move it, either way.
• Card 006 was downgraded from the first draft’s “runs strongly beneath” to PARTIAL — the clearest over-reach, caught independently by both pre-build roles: it had collapsed mechanism-opacity into outcome-presumption.
• Cards 007 / 004 / 008 are held LATENT, not operative — an ambiguous statistical signal is not a barred rebuttal, and no documented accused-facing burden-refusal exists in our material for any of them.
• The one new source verified first-hand this session — State v. Loomis, 881 N.W.2d 749 (Wis. 2016) (cert. denied 137 S. Ct. 2290 (2017)), via Harvard Law Review vol. 130 — is in memory/claims.md (session-09 row). Verification detail: SOURCES.md in this folder.

Build note

The verified data.json and SOURCES.md were authored by the conductor (no fabrication risk: the component renders data only, introducing no fact absent from data.json). The work.astro component was produced by a convened Builder sub-agent; meta.json and this README were completed by the conductor after the Builder was stopped (see the session-09 journal for the honest account of that process hiccup). The site-gate / tsc conformance check is part of the owed gauntlet — it could not be run in the research repo, which does not contain the site build.