Published result

Evidence-path guidance before and after rejection

A plain-language summary comes first, followed by the full text of the result's reproduction pack.

Result summary and pack text · reproduce/lab-shaping-dual-path-guidance-screen-20261108/README.md

In plain terms

Evidence-path guidance reached 7 of 10 ready, with one regression against the stored release, and was dropped.

Verdict key: Yes, The result supports the change or claim that was tested. No, The result does not support it, so nothing changes. Unresolved, The evidence is not strong enough either way. Superseded, A later test replaced this result.

Full text of the reproduction pack

The text below is the pack's README as published in the repository (a system folder name may show without its leading slash, so that the site's privacy scan passes).

On ten small requests, the stored release reached 6 ready Intents and a candidate reached 7. Two cases changed from not ready to ready and one from ready to not ready. A second unpatched repetition produced exactly the same paired outcomes. The candidate was dropped because its regression fails the registered rule and the net gain does not exceed observed variation. No questions were asked. This is a screen, not a qualified delivery change.

All three discordances occurred in shaping; their cause is unclear because stochastic variation is also observed. In the loss the candidate used bare tracked paths in a recorded decision but then failed a later carried-decisions check; the stored run recovered after an Intent-format refusal. In both gains the stored release ended at the evidence check while the candidate recovered after decision-check refusals. The path instruction is consistent with the form observed in the loss, but the trace does not establish a causal gain. The decision, evidence and readiness gates remained unchanged.

The candidate gave the same guidance both initially and after rejection: use a separate bare inspected tracked path for each file evidence item, without line ranges or prose in the path. Both arms used the same model, low effort and no answerer; the control was stored. Times include failed attempts. The shared credit balance did not move visibly; attributable spend is unknown. No held-out or complex requests, Build stage or hidden checks ran; landing and hidden outcomes are unmeasured. Small Go and Node workspaces were reconstructed; byte identity with the earlier study is unverified.

The release uses Kogen source and shaping source. The candidate adds only the guidance described above.

Rerun

Run python3 report.py to reproduce paired counts. To repeat the model screen, use the small-request setup, pinned release and same ten requests; add the described initial and feedback guidance, keep all gates, and shape each request with the same model, low effort and no answerer. Record readiness, questions, refusal causes and elapsed time.