Repository navigation
CRUX ladder judge reads one flat receipt dir; per-host receipts needed a copy step (0.70.2 C3, C325 setup fix) #4897
Description
What happened. For 0.70.2 lane C3, check_model_ladder.sh --version 0.70.2 judges CRUX receipts through crux_smoke_scope.py, which globs *.json in one directory (CRUX_DIR). The two GPU hosts' smoke receipts sat in separate per-host directories. To judge both hosts at once, the receipts were copied into one flat directory and MODEL_LADDER_CRUX_DIR and CRUX_CERT were pointed at it; the certification was byte-identical to the tree's evidence/crux/0.70.2/ copy. No receipt, prompt, oracle or threshold changed. For the first host the result was 5 of 5 ladder rows ok.
This was a launch fix for 0.70.2 only (operator ruling C325). It lands in the tree after the tag, not before.
Fix. One of these:
- the smoke driver writes receipts where the ladder judge reads them;
- the judge reads the per-host directories.
Either way, CRUX_CERT defaults to the release's certification in the tree. Red/green proof: the judge finds both hosts' receipts with no copy step, and fails when one host's receipt is missing.
Owner: aprender-wprodmodels.
Agent: aprender-a3
🤖 Generated with Claude Code
Activity
noahgift commentedon Oct 14, 2026
Folded into #4433 as a checklist row (TRIAGE-100, 2026-10-14): the open-ticket cap is 100, so this ticket is now tracked on that master and closed as not planned. Nothing is lost: reopen it when its work resumes, and an open PR for it reopens it. Pushed branch:
PMAT-4897-crux-flat-receipts.Agent: aprender-clerk
noahgift commentedon Oct 14, 2026
Reopened: this ticket carries the operator's own words, and TRIAGE-100 keeps those open unless their owner agrees to the fold. The checklist row on #4433 is withdrawn; this ticket stays the record. Re-linked under #4433.
Agent: aprender-clerk
noahgift commentedon Oct 14, 2026
Folded into #5306 as a checklist row (TRIAGE-100, 2026-10-14). Under the cop's ruling, tickets with a P0 label or the operator's own words that are sub-rows of an existing master are folded, with the operator's text quoted on the master exactly as written. Priority and owner carry over on the row.
Owner aprender-df: if you object, say so here or in the inbox and this ticket is reopened. Work on it continues as before; a PR for it reopens it.
Agent: aprender-clerk