session 2026-08-24 afternoon: the guards were not on the path

Session record + MEMORY.md promote/demote (prior Active Session relocated
verbatim to MEMORY-reference.md, md5 d6a53ddb, slice-not-retyped and verified
present). 11 KG lines: 5 drift-patterns, 3 preventions, 3 facts. 3 skill-harvest
proposals (a wake chamber conditional -- third instance of the same miss; the
wrap's Instruments field to report OBSERVED first firing per PENDING-156 (c);
the daily-note format defined in two places). 1 FIX-lane index line.

Pulling thread: the beacon, 2026-08-25 14:00 CEST, run ONCE -- which had no
trigger of any kind until tonight.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01JQKeKY9T9d95KpvHwwok8T

claude/memory/knowledge-graph.jsonl

@@ -708,3 +708,14 @@
 {"subject": "/wake-up", "predicate": "reaches", "object": "0.71% of the memory corpus — 6,107 words of 859,803, always the newest, ordered by recency. Measured 2026-08-24. This is the gap thread-query.py exists to close, and the number the six-week trial is graded against.", "valid_from": "2026-08-24", "valid_to": null, "confidence": 1.0, "source_file": "session-2026-08-24-the-heap-got-a-dynamo.md", "extracted_at": "2026-08-24"}
 {"subject": "claude-code", "predicate": "drift-pattern", "object": "Read a CONFIG FILE as evidence of LIVE STATE, and reported a failure that had not occurred. Smart Connections' smart_env.json still recorded TaylorAI/bge-micro-v2 in a file written 14 seconds before the check; I treated the freshness of the mtime as evidence of currency and told the steward the model change had not taken. The plugin's own progress panel showed it embedding with Xenova/multilingual-e5-small at that moment. A config file is a PERSISTENCE ARTIFACT WITH A DEBOUNCE (re_import_wait_time: 13) — it answers 'what was last written', never 'what is running now'. The running process's own report is the substrate for a liveness question. The banked ladder rule 'verify the RUNNING BINARY's provenance, not source HEAD' is the identical principle in another domain and was not reached for.", "valid_from": "2026-08-24", "valid_to": null, "confidence": 1.0, "source_file": "session-2026-08-24-the-heap-got-a-dynamo.md", "extracted_at": "2026-08-24"}
 {"subject": "stating both possibilities instead of picking one", "predicate": "prevention", "object": "LIMITED A WRONG REPORT TO A SINGLE-DIRECTION CORRECTION. When smart_env.json disagreed with the steward's action, the report said explicitly 'that may simply be Obsidian not having flushed yet... this is the disk not having caught up, not it failed again — I cannot tell which from here, and I would rather say so than pick one.' The benign branch turned out correct. Because the uncertainty was declared rather than resolved by guess, the steward was not sent to undo work that was already right, and the record needed amending in one direction only. Hedging is usually noise; here the hedge was the load-bearing part.", "valid_from": "2026-08-24", "valid_to": null, "confidence": 0.9, "source_file": "session-2026-08-24-the-heap-got-a-dynamo.md", "extracted_at": "2026-08-24"}
+{"subject": "claude-code", "predicate": "drift-pattern", "object": "REPORTED BEFORE READING, twice in one afternoon. (1) Proposed giving the OCR pipeline its own workstream tracker; it is two leaves of conversion-runbook.yaml's decision_tree and the CapableHands authorization already lives in its `machines:` block — the steward corrected it in two words, 'read the runbook?'. My own standing rule says ANY chamber work OR TALK ABOUT IT begins with touchstone/constitution/charter/runbook, and the whole conversation was chamber talk. (2) Reported the hook-reachability defect as a new discovery; it had been on PENDING-95 since 2026-08-19 and cross-filed by the jurist on 08-20. Both would have been avoided by opening one file BEFORE composing rather than after.", "valid_from": "2026-08-24", "valid_to": null, "confidence": 1.0, "source_file": "session-2026-08-24-afternoon-the-guards-were-not-on-the-path.md", "extracted_at": "2026-08-24"}
+{"subject": "claude-code", "predicate": "drift-pattern", "object": "A BANKED MEMORY NOTE TOLD ME A VERIFICATION WAS POINTLESS, AND IT WAS FALSE. feedback-chamber-work-ground-in-constitution-charter-runbook stated the conversion runbook 'does not yaml-parse — it is human-read'. It parses. I ran the check anyway and it caught a defect I had just introduced (inner double quotes inside a double-quoted YAML scalar) that would have shipped a runbook that no longer loads. A note that licenses SKIPPING a check is the most expensive kind to get wrong, because its failure mode is invisible by construction. Superseded in place with strikethrough, not silently rewritten. The class — notes that license skipping checks — has no census.", "valid_from": "2026-08-24", "valid_to": null, "confidence": 1.0, "source_file": "session-2026-08-24-afternoon-the-guards-were-not-on-the-path.md", "extracted_at": "2026-08-24"}
+{"subject": "claude-code", "predicate": "drift-pattern", "object": "A PROXY CENSUS PRODUCED A TIDY ANSWER THAT WAS WRONG IN BOTH DIRECTIONS AT ONCE. Classifying mechanisms as hook-wired vs prose-wired by asking 'is it named in a SKILL.md?' gave a clean 3/3/3 split. A caller search corrected two of nine, in OPPOSITE directions: governance-drift-check.py looked prose-wired and is executed by wake-digest.py; vault-links.py looked code-referenced and is only cited in a docstring comment. A proxy does not err in one direction you can correct for. The tidy first answer is the one that would have been filed.", "valid_from": "2026-08-24", "valid_to": null, "confidence": 1.0, "source_file": "session-2026-08-24-afternoon-the-guards-were-not-on-the-path.md", "extracted_at": "2026-08-24"}
+{"subject": "claude-code", "predicate": "drift-pattern", "object": "A BROKEN TEST HARNESS NEARLY BECAME A REPORTED BUG. Live-testing daybook-cue.py with `echo '{...\\n...}'` under zsh, whose echo interprets backslash escapes, produced malformed JSON; the hook correctly fail-opened (exit 0) and I read the silence as the hook failing. Rebuilt the payload with json.dumps before drawing any conclusion. The instrument under test was fine; the instrument doing the testing was not, and its failure mode was indistinguishable from the target's.", "valid_from": "2026-08-24", "valid_to": null, "confidence": 1.0, "source_file": "session-2026-08-24-afternoon-the-guards-were-not-on-the-path.md", "extracted_at": "2026-08-24"}
+{"subject": "claude-code", "predicate": "drift-pattern", "object": "FIXED A DEFECT BY EYE AND MISSED A SECOND INSTANCE OF IT. Repairing the YAML break, I found one pair of inner double quotes, replaced it, and re-ran — still broken. A census of all inner-quote pairs in the block found the second. Finding one instance is not fixing the defect; the count comes first.", "valid_from": "2026-08-24", "valid_to": null, "confidence": 1.0, "source_file": "session-2026-08-24-afternoon-the-guards-were-not-on-the-path.md", "extracted_at": "2026-08-24"}
+{"subject": "running the check a memory note called unnecessary", "predicate": "prevention", "object": "CAUGHT A DEFECT INTRODUCED INTO A GOVERNED FILE MINUTES EARLIER. The banked note said the conversion runbook does not yaml-parse, which would have made the parse check pointless. Running it anyway revealed that (a) the file DOES parse and (b) my own edit had just broken it. The note was corrected as a result. A memory layer is a witness, not a notary — the CLAUDE.md rule that a conflict between layers is a verification trigger fired here in the strongest form: the substrate contradicted the note AND my edit at once.", "valid_from": "2026-08-24", "valid_to": null, "confidence": 0.95, "source_file": "session-2026-08-24-afternoon-the-guards-were-not-on-the-path.md", "extracted_at": "2026-08-24"}
+{"subject": "verifying another party's finding before accepting it", "predicate": "prevention", "object": "CONFIRMED A CORRECTION RATHER THAN DEFERRING TO AUTHORITY, AND MADE IT USABLE. The jurist, reading both grounding hooks, found that PENDING-95 Amendment 1's claim — that the pre-commit gate 'reuses the current rule' — was false, and that the gate is a strict superset. Rather than accept it on the jurist's authority I constructed the discriminating case: an edit stripping the only marker from a marked file. Exit 0 at the PreToolUse hook, exit 1 at the commit gate. The finding is now confirmed by the party it corrects, which is what let the standing instruction be rewritten as ASYMMETRIC (pre-commit is the floor, do not equalise downward) rather than as a vague 'apply to both'.", "valid_from": "2026-08-24", "valid_to": null, "confidence": 0.95, "source_file": "session-2026-08-24-afternoon-the-guards-were-not-on-the-path.md", "extracted_at": "2026-08-24"}
+{"subject": "demanding a positive control before trusting a new gate", "predicate": "prevention", "object": "STOPPED A DECORATIVE TRIGGER BLOCK FROM BEING FILED FOR AN IRREVERSIBLE ACT. The beacon derivation — authorized, dated, run-once — had NO trigger of any kind. Filing a DEFERRED-DECISION block that merely LOOKED well-formed would have reproduced the exact failure being fixed. Setting the trigger date temporarily to the past proved the checker announces it BY NAME as COME DUE; restoring the real date returned 'none due'. Same discipline applied to the pre-commit grounding gate (constitutional+ungrounded BLOCKS, constitutional+grounded PASSES, non-constitutional PASSES) before it was committed.", "valid_from": "2026-08-24", "valid_to": null, "confidence": 0.95, "source_file": "session-2026-08-24-afternoon-the-guards-were-not-on-the-path.md", "extracted_at": "2026-08-24"}
+{"subject": "the jurist", "predicate": "drift-pattern", "object": "Asserted TWICE in one session that its governance tools were unavailable; they were deferred behind tool_search and never called. Self-caught and self-reported. Mirror image of the executor's 2026-08-23 claim that Claude.app has no filesystem access. Both are negative claims about the OTHER party's capabilities made with no positive control, by parties who each hold the doctrine forbidding it. Filed to PENDING-89 as a same-direction miss: neither party checks what the other can actually do, and it cost real work in both directions.", "valid_from": "2026-08-24", "valid_to": null, "confidence": 1.0, "source_file": "session-2026-08-24-afternoon-the-guards-were-not-on-the-path.md", "extracted_at": "2026-08-24"}
+{"subject": "daybook-cue.py", "predicate": "was-fixed-to", "object": "Trigger changed from note SIZE (note_size < SKELETON_CEILING) to STALENESS (note untouched > 30 min while work lands), on the steward's ruling 'the daily note needs to be appended to until the end of the day. Period. There is no design choice to make.' Matcher widened Write|Edit -> Bash|Write|Edit. Self-silences on compliance because appending resets the note's mtime. Fired twice in anger the same evening, having never fired once in the ~7 hours it was live with the old wiring.", "valid_from": "2026-08-24", "valid_to": null, "confidence": 1.0, "source_file": "session-2026-08-24-afternoon-the-guards-were-not-on-the-path.md", "extracted_at": "2026-08-24"}
+{"subject": "the violin/music-in-the-XXI-century treatise", "predicate": "is-gated-by", "object": "conversion-runbook.yaml known_gaps.research — polytonic Greek / Latin / fraktur / long-s OCR. Written at both ends 2026-08-24 (runbook 7868bd5 + project-chamber-versioned-releases) because the dependency existed only in the steward's head: the runbook knew 'first historical treatise' as a trigger without knowing which, and the works end knew the treatise waited on 'the Chamber' without knowing on which gap. Partly crossed: Docling+OCR on scanned French, ~11 min on the M4, readable but NOT verbatim-clean; UNTESTED on polytonic Greek and fraktur. The treatise's material is live at vault 06. Projects/Pattern, Presence, Practice — 177 notes / 158,313 words — and has no repo by design.", "valid_from": "2026-08-24", "valid_to": null, "confidence": 1.0, "source_file": "session-2026-08-24-afternoon-the-guards-were-not-on-the-path.md", "extracted_at": "2026-08-24"}