lab-report

name

lab-report

description

Answer questions about the shared HyperKit experiment lab in lab/ without changing anything: what experiments exist, what is running, what the latest results and decisions were, who owns what, which follow-ups are up for grabs, and whether a proposed idea overlaps an existing claim. Use when asked "what were the results of the last experiments", "what experiment was this", "what is running right now", "summarize eNNN", "what did we learn about <topic>", "what follow-ups are open", "did anyone already run <idea>", or for a lab status report. Read-only: reads REGISTRY.md, experiment.md files, JOURNAL.md, and (only for raw detail) Athena/Grafana.

Lab Report (read-only)

Answer from the committed record. Do not modify any file in lab/; if the record looks stale or inconsistent, say so instead of fixing it.

Reading order

1. lab/REGISTRY.md — index of every experiment: id, owner, status, benchmark, claim, budget, spent.
2. lab/experiments/<id>/experiment.md — per-experiment truth. Front matter for state (status, updated, spent_usd); sections for substance (Hypothesis, Design, Out of scope, Results, Decision, Follow-ups).
3. lab/JOURNAL.md — newest-first cross-experiment log; only entries that affected shared plans.
4. sweep.lock.json — exact frozen cell matrix, generations, repo SHA.
5. Grafana (run_id = experiment id) / Athena (sweep_id partition) — live progress and raw per-shard data. Go here only when the question needs numbers beyond the Results section.

Question -> source

Question	Answer from
results of the last experiments	registry rows analyzed, sorted by updated in front matter; then each Results + Decision
what was experiment X	its experiment.md: claim, Hypothesis, Design, Results, Decision
what is running / planned	registry status column (running, locked, proposed)
what exactly did X run	experiment.py + sweep.lock.json
open follow-ups	Follow-ups sections marked up-for-grabs
did anyone already run	claim column + Out-of-scope sections; note near-misses, not just exact matches
how much have we spent	spent_usd (analyzed) and budget_usd (in flight) across front matter
live progress of a sweep	Grafana filtered to run_id = <id>; do not write it back into git

Answer style

• Lead with the direct answer, then cite experiment ids and quote the Decision line where relevant.
• Report Results numbers with their confidence intervals as written; never restate a point estimate without its interval.
• Distinguish clearly between analyzed conclusions, in-flight sweeps, and merely proposed claims.
• If two sources disagree (e.g. registry vs front matter), report the discrepancy — front matter wins.