[choiyounggi]

Knowledge flush — 1 insight(s) (headless auto-flush run 20261030-112310-27790)

This run claimed 13 queue rows. Results: 1 ingested, 2 dropped as duplicates of
the merged wiki, 1 released because it plausibly overlaps open knowledge PRs, and
9 released unread because this headless run could not find their queue files.
Nothing was folded into another branch.

Verified best-practice

Row	Claim	Verification	Confidence
248ccfddbcc612a1	A test helper that parses Prometheus exposition text with a label-set regex like \{([^}]*)\} rejects valid lines once label values contain } (route templates such as /order/{id}). Prefer a real parser. If the helper must stay a regex, match each quoted value as one unit, and keep one negative case showing an unclosed label set is still rejected	https://prometheus.io/docs/instrumenting/exposition_formats/: "label_value can be any sequence of UTF-8 characters, but the backslash, the double-quote, and the line-feed characters have to be escaped". So } is legal inside the quotes. https://prometheus.github.io/client_python/parser documents prometheus_client.parser.text_string_to_metric_families, and the README notes this parser is more lenient than the Go one. https://www.robustperception.io/openmetrics-is-released confirms the required # EOF line, which backs the OpenMetrics edge row. https://prometheus.io/docs/instrumenting/escaping_schemes/ backs the quoted-UTF-8-names edge row. The session evidence was the line ...{route="/order-service/order/{id}",...} 2 raising "unparseable exposition line" under the old regex; with the quote-aware regex, 265 tests passed and broken{a="x" 1 still raised. The page's regex was not re-run in this headless run because Bash python3 needs approval here. It is the session's own pattern, wrapped in \{...\}	verified
9cce9c1914fef241	A test-floor gate marks a unittest self.assertX file as no-assertion: read the floor's Python pattern and treat the result as a false negative	Not re-researched. The merged page already owns this exact trigger (see below)	dropped (duplicate)
bb8b376ef7f0506e	In-place mutate/restore probes on Python source can run stale .pyc bytecode. Use -B or delete __pycache__	Not re-researched. The merged page already covers it with official docs. That page also shows the candidate's "run each probe with python -B" option is not enough on its own: -B only stops writing, and an existing stale .pyc is still read	dropped (duplicate)

Existing-layer check

Pages read: testing-quality-assertion-scanner-false-positive-on-unittest-convention, backend-python-language-bytecode-cache-staleness, infrastructure-observability-logs-metrics-signals

Indexes read: wiki/testing/index.md (full).

• 248ccfddbcc612a1: wiki_search on Prometheus label parsing returned no page on that trigger. The top hits were config-value escaping, allowlist-regex anchors, and frontmatter quote stripping. A search for Prometheus metrics found infrastructure-observability-logs-metrics-signals. That page tells you to label with the route template (/users/{id}), which is exactly the label shape that breaks the naive regex. New page testing-quality-prometheus-exposition-parsing-in-test-helpers. related: links to that observability page, which gets a back-link, and to testing-data-adjacent-tokens-in-extractor-fixtures, testing-quality-anchor-cases-for-allowlist-regexes and testing-quality-tests-that-cannot-fail. No conflict.
• 9cce9c1914fef241: testing-quality-assertion-scanner-false-positive-on-unittest-convention already covers it. Its trigger is a floor that greps for bare assert/pytest.raises and flags self.assertX files. Its step 6 already says to read the classifier and treat the result as floor=unknown, and a 2026-10-15 field row shows the same test-floor.sh analyze_pytest miss. Nothing new, so the row was dropped as a duplicate and retired.
• bb8b376ef7f0506e: backend-python-language-bytecode-cache-staleness already covers the mtime+size validation, purging __pycache__ between mutants, re-verifying after the revert, and the row where a delegated auditor mutated the files. It also corrects the candidate on -B. Dropped as a duplicate and retired.

Open-PR check

There are 14 open knowledge/* heads: #317, #315, #310, #309, #308, #307, #306, #305, #304, #301, #300, #299, #298, #293. This headless run has no git access, so it could not diff them. Instead it compared candidates against each PR's changed-file list (gh pr list --json files).

Row	Overlapping open PR	Verdict
248ccfddbcc612a1	None. No open PR touches Prometheus/metrics parsing or infrastructure/observability/. #299 edits platforms/processes/parsing-cli-structured-output.md, which is about CLI tool output, a different trigger. Several PRs touch other rows of wiki/testing/index.md only	new
9cce9c1914fef241	#315 edits assertion-scanner-false-positive-on-unittest-convention.md, but the merged page already holds this row	drop (duplicate of merged wiki)
bb8b376ef7f0506e	None touches the bytecode page	drop (duplicate of merged wiki)
fc76dc4dfa08ecde (Docker-gated schema validator: skip vs fail via a REQUIRE env var, plus a known-bad document)	#300 and #317 both edit testing/quality/checks-that-cannot-pass.md, and #300 also edits tests-that-cannot-fail.md. A validator that is skipped or always says "valid" is that page's core theme	released (pending duplicate)

Routing decision

• 248ccfddbcc612a1 → testing/quality/prometheus-exposition-parsing-in-test-helpers.md (new page), with a row in the quality table of wiki/testing/index.md. No new category. The lesson is about whether a test's own parsing helper can be trusted, which fits quality. data holds fixture-construction pages. The domain hint testing was followed. The directive is general: any repo whose tests parse /metrics. The project run appears only in a Sources line.
• 9cce9c1914fef241, bb8b376ef7f0506e → no edit (duplicates). Rows retired in place with status: retired, an outcome, retiredBy and retiredAt.
• Released without routing: fc76dc4dfa08ecde (open-PR overlap, above). Also 1d01cb971e07f6cb, 33ffb8688290c9c1, 063108972d150963, cb3954d76f244491, fdbac2d18d580e0c, 7a851159bdab0e27, ee3362544f6d9162, aa190116f0c2596a and 4b0c7878166d1fa8. This run's sandbox blocks listing ~/.dev-loop/queue/. Session files can only be found by guessing their session-id names, and those rows sit in files that come after 76896825-… in sort order, which could not be found from ~/.claude/history.jsonl session ids. They were released unread for a later manual flush.

Local-layer candidates

none

[github-actions]

Wiki agent gate: fail

PR #318 adds one page, wiki/testing/quality/prometheus-exposition-parsing-in-test-helpers.md, plus an index row, a back-link and a log entry. The page is general and not duplicated. It fails on one fact check: the page cites a claim that its source does not contain. The claim is that the client_python parser is "more lenient than the Go reference parser". Neither the cited parser page nor the project README says it. The other three cited URLs support the claims they are cited for.

• blocker [fact] wiki/testing/quality/prometheus-exposition-parsing-in-test-helpers.md — The page attributes "the parser is more lenient than the Go reference parser" to https://prometheus.github.io/client_python/parser (and to the project README) in Do-this step 2 and in the Sources section. I fetched the parser page and the GitHub README (https://github.com/prometheus/client_python). The parser page only documents text_string_to_metric_families. The README contains no text on a parser or on the Go parser. Step 2 also tells the reader the parser "proves values, not strict format conformance", which depends on this claim, and the page is marked confidence: verified. Fix by citing a source that states it, or by removing the claim and the dependent caveat, or by marking the leniency claim as field-observed and lowering the page to field-tested.
• advisory [fact] wiki/testing/quality/prometheus-exposition-parsing-in-test-helpers.md — The exposition_formats, robustperception (# EOF) and escaping_schemes URLs were fetched and support the cited claims. The escaping_schemes text sits under the allow-utf-8 scheme, and the page's wording is a fair match. The regex and the 265-test result come from a field reproduction that CI cannot re-run, and the PR description says the regex was not re-run. This is unverifiable from CI; the claim is internally consistent with the page's own measurement log.
• advisory [duplication] wiki/testing/quality/prometheus-exposition-parsing-in-test-helpers.md — No existing page teaches this directive. The Prometheus-related pages (infrastructure-observability-missing-container-metrics, infrastructure-observability-logs-metrics-signals) are about instrumentation, not test-helper parsing. The new page is cross-linked to logs-metrics-signals in both directions through related:.