Exact identifiers are found, an Edit is not proof of verified, and a package is not a file - #17
Merged
Merged
Conversation
…e note that contains it
…illed claim was verified
…the session touched
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Three 0.5.1 fixes, each with a test that fails on the old code.
Searching for WG-4471, ENG-1234 or 0.4.10 finds the note that contains it (34ae500)
ftsQuerystripped punctuation and glued the pieces into one token (WG4471), which the FTS index never holds — it indexedwgthen4471. A word with punctuation in it now becomes a phrase,"WG 4471", which matches the identifier as a unit and not a note that haswgand4471apart. With embeddings off, the old binary said "Nothing in the vault matches".A successful file edit is no longer proof that a distilled claim was verified (30b11a6)
The citation filter accepted any tool turn with status ok, and Claude Code reports an
Editas ok. So "the crash is gone (turn 3)", citing the edit that tried to fix it, was kept as verified. A verified claim now needs a cited turn whose tool ran a command (isShellTool, the existing per-harness list). Matching the runner names rather than excluding edit names fails safe: an unknown runner costs a claim, an unknown edit tool would let a false one through. The distiller prompt and theingest_distildescription now say a file edit is not a demonstration.A package or directory a command names is no longer listed as a touched file (e990ed3)
filesInCommandtook any argument with a slash as a file, sogo test ./cartputcartin Files, alongsidecddirectories and sed expressions. A slashed argument now needs an extension. An extensionless file named only in a shell command drops out of Files but stays in Commands.Verified
go build ./... && go vet ./... && gofmt -l . && go test ./...,go test -tags chaos ./chaos/...,staticcheck ./...— cleango test ./cart): the old binary kept 2 verified claims and listedcart; the new one kept 1, reported the dropped claim with its reason, and listed only cart.go