Repository navigation
The repo-graph freshness gate compares committed == fresh and passes when both are 356 triples short — assert a floor from binding.yaml entries / main's count, not self-consistency #3641
Description
What
the_tracked_repo_graph_is_fresh passes when contracts.nt (committed) == fresh extraction. Both being 356 triples short — as on #3350's branch after a one-character binding.yaml corruption (see the producer-side row) — passes. Equality with a fresh extraction is a consistency check, not a correctness check: it cannot see a shrink that the producer reproduces deterministically.
done_when
- the graph gate asserts a floor derived from the inputs, not from itself:
bound symbols resolved == binding.yaml entries(each of the 156 entries yields its node) and/ortriples ≥ origin/main's committed countunless the PR'sont-delta:line declares a shrink with the reason - mutation: remove one binding entry's node from the fresh extraction (or corrupt the input as on fix(format): ModelConstraints dropped the gated-DeltaNet shape, so 18 of every 24 Qwen3.5 layers were counted as if their tensors did not exist #3350) → RED naming the missing symbol(s); restore → GREEN
- the count and the source of the floor are printed in the gate's PASS line, so a reader sees what was compared
Refs #3350, the producer row (pv extract must refuse), #3639.
Activity
noahgift commentedon Sep 20, 2026
d4's discriminator, in their words, for the shape of the floor: the reason the
binding.yamldefect was catchable at all is that main's committed graph was a known-good 15,600 and the branch produced 15,244 — a count against a reference found it, wherecommitted == freshcould not. The tell is a drop against something outside the run, never agreement with yourself. Same discriminator as the stale-pvcase (#3635/#3600).noahgift commentedon Sep 23, 2026
milestone → 0.71.0 per docs/roadmaps/milestone-triage-2026-09-23.md (#4024), operator approval 2026-09-23 (relayed by aprender-cf). Reason: repo-graph freshness floor, with #3640: debt ratchet pillar C (#3997), slice 2
noahgift commentedon Sep 24, 2026
Closed by paiml-implement — folded into epic #2556 (triage evidence/triage-1623.md)