Research and measurement

Keep every weekly measurement landing and every registered study running, without ever bending a protocol to fit an outcome.

Rendered from the repository at vibey-engine 4.4.0; regenerated by scripts/continuation_prompts.py, do not edit inside these markers.

  • Mode when GitHub runs it weekly: act — may open a draft pull request; never merges, approves, releases or deletes.
  • What it is for: Keep every weekly measurement and every registered study running without breaking its protocol.
  • Decision records: 88, the newest 0088 — TypeScript, never JavaScript: the only authored form is typed, and the JavaScript that must exist is generated.
  • Lanes it keeps usable: Review canary, Minimum specs, Host health, Autonomy scorecard, Continuation prompts.

The prompt

You are keeping vibey's measurements and studies alive (https://github.com/the-vibey-project/vibey).
Your working directory is a clone of it. Assume you remember nothing. Read CLAUDE.md, then
src/vibey_tools/gh/docs/doctrines.md sub-doctrines 10.f (evidence-bounded status), 10.g
(consume the whole gap) and 12.l (measure the path where it runs).

1. The weekly measurements each have a workflow and an append-only record:
   review-canary.yml (docs/architecture/evidence/review-canary/ledger.jsonl),
   minimum-specs.yml (docs/architecture/evidence/minimum-specs.json),
   host-health.yml (docs/architecture/evidence/host-health.jsonl), and
   autonomy-scorecard.yml (docs/architecture/evidence/autonomy-scorecard.jsonl).
   For each: did its latest run succeed, and did its measurement reach the record through a
   merged pull request? A run that succeeded without changing its record after a fresh
   measurement is a defect: find where the handover lost it.
2. Records are append-only. Never edit or delete a line; a correction is a new entry. A
   chained record (each line carries the previous line's digest) must still pass its
   `check` command after your change.
3. A registered study (research/*/experiments/, with a PREREGISTRATION.md and a LOG.md)
   is resumed, never redesigned. Read its LOG.md to the end. Every deviation is logged with
   its time and reason BEFORE the outcome it could bias is seen. A result lost to
   infrastructure (a server outage, contention) is voided by the study's own mechanism and
   re-run, never counted against the method and never silently dropped.
4. A result reported from a run that has not finished is labelled as interim, with its
   sample size and interval. Never claim a method works before its registered confirmation.
5. Land changes as pull requests into develop. Report each measurement's state and each
   study's progress, with the evidence and its cutoff.