# Precedents, policy, and commitment

Canonical: https://www.meshqu.com/research/precedents-policy-and-commitment

We re-ran the same procurement decisions through an AI agent with the governance context broken apart piece by piece — to find out which piece was doing the work.

Published: 2026-06-09T00:00:00.000Z

## Key facts

- 3.5% vs 0% vs 0% — Precedents alone produce far less commitment than predicted. The precedents-only arm returned DENY on 3.5% of records (10 of 283) against a locked 20% floor, while the verdict-stripped arm and the density control returned none at all. (The prediction is falsified, but the mechanism is not refuted: the precedents-only arm is still the only condition producing any DENY at all. The reading the corpus supports is that accumulated governance context amplifies commitment rather than precedents carrying it alone.)
- 60.7% retention, 3.7pp delta — Removing the anti-sycophancy clause barely changed anything. The variant without it retained 60.7% of the earlier experiment's committed set (65 of 107) against a 65% falsification floor — a 3.7 percentage-point difference from the condition that had the clause. (The paper's reading is that the policy text's structural cues were already driving the reversion; the nudge is a discipline reinforcement, not the causal agent. It does not conclude that the clause is useless.)
- 88 of 100 — Inversion-blindness reproduces at scale. With the policy operator reversed, 88 of 100 records produced the same verdict as under the unperturbed policy. (Recorded as Under-tested rather than confirmed or falsified: 88% sits two points below the locked confirm floor and no falsification band was locked, and the locked vocabulary does not permit a post-hoc rule. The paper treats that restraint as a methods contribution in its own right.)
- 93% and 100% — Both model families reason the same way under inversion. The category for reasoning solely against rule intent — rather than noticing the inversion — dominated at 93% on the primary model and 100% on Claude Opus 4.7.
- 23% vs 80% — Verdict behaviour does not travel with reasoning behaviour. On the same record-matched corpus GPT-5.4 committed to a verdict on 23 of 100 records and Opus 4.7 on 80 of 100 — a 57-point gap in decisive rate, and a 46-point gap on the same-as-unperturbed rate (88% against 42%) that falsified the prediction of a task-class result on that axis. (The paper's implication for practitioners is narrow and specific: reasoning evaluations may generalise across models while verdict evaluations may not, so cross-model deployment in regulated decisioning should test both axes rather than assuming one implies the other.)
- 1 confirmed · 3 falsified · 2 under-tested — Across six pre-registered predictions, one was confirmed, three were falsified and two could not be adjudicated because their locked criteria specified a confirm band with no falsification band. (All predictions, thresholds and artefacts were unchanged from the lock. The paper's position is that the Under-tested calls are the only honest ones available, and that the asymmetric lock is itself a lesson for the next design.)
- 1,332 / 1,332 PASS — Every receipt in the run verifies: 1,332 receipts across six arms, produced in 80 minutes 33 seconds, with a clean verifier pass on all of them.

{/* The PDF remains the version of record. The summary, method, findings
    and limitations above are frontmatter, not body prose — they render
    through the research template's structured blocks (PSA-040) so that
    each one is separately labelled, quotable and checkable against the
    section of the paper it came from. The MDX body stays empty by
    design: this page frames and summarises the publication, it does not
    become it. */}