{"EL1-EMPIRICAL-OVER-CLAIMED-AND-RECOVERABLE": {"bucket": "covered", "cited_L": "L1", "stem": "EMPIRICAL-OVER-CLAIMED-AND-RECOVERABLE", "note": "Ground-truth-over-claimed-summary is L1-FABRICATED-LEDGER's exact shape (also P1-EMPIRICAL). The 'a dead RETURN loses the report, never the work' half is P11-RECOVERABLE-BY-DESIGN (scar L4-PAUSE-RESUME). Both halves already canon; no new concept."}, "EL2-FAST-IS-SUSPICIOUS": {"bucket": "covered", "cited_L": "L2", "stem": "FAST-IS-SUSPICIOUS", "note": "Same id-stem, same lesson, same shape (anomalous speed triggers an audit not a verdict; exonerate as readily as correct)."}, "EL3-POC-NEVER-CHANGES-ENTABENI-API-CONVENTIONS": {"bucket": "engagement-specific", "cited_L": null, "stem": "POC-NEVER-CHANGES-ENTABENI-API-CONVENTIONS", "note": "Entirely about Entabeni's own architecture: entabeni-api (Rails, Zendesk/TSYS/Multipass, host-based multi-tenancy), the CST/lodging POC boundary, named ports (ZendeskPort, CnsPort, OpenGDS/Core). No portable cross-project doctrine to extract beyond the generic 'a consumer adapts to the source of truth, never rewrites it' idea, which is architecture-specific detail, not a reusable kit rule. Confirms prior triage."}, "EL4-PAUSE-RESUME": {"bucket": "covered", "cited_L": "L4", "stem": "PAUSE-RESUME", "note": "Identical title and identical doctrine (pause = checkpoint + park on disk; resume = re-launch from ledger row; max loss = uncommitted working tree). This is P11-RECOVERABLE-BY-DESIGN's named scar."}, "EL5-USAGE-CAPS-AND-BUFFERED-LOGS": {"bucket": "covered", "cited_L": "L5", "stem": "USAGE-CAPS-AND-BUFFERED-LOGS", "note": "L5-USAGE-CAPS is the same incident shape (cap kills a worker in seconds, a build-failure-only watcher sees a quiet 'healthy' fleet; judge liveness by ground truth, never log stillness) and is P7-SILENCE-IS-NOT-SUCCESS's scar. The claude-code-specific buffered-stdout detail is an implementation nuance of the same rule, not a separate concept."}, "EL6-PRIOR-ART-AND-ENV-FRICTION": {"bucket": "covered", "cited_L": "L6", "stem": "PRIOR-ART-AND-ENV-FRICTION", "note": "[sensitive content omitted]"}, "EL7-THE-GATE-VARIES-BY-REPO": {"bucket": "covered", "cited_L": "L15", "stem": "THE-GATE-VARIES-BY-REPO", "note": "The general rule ('use the exact gate command from the unit's own profile, never a guessed equivalent; a repo with no enforced gate needs an explicit human/device proxy') is L15-VERIFY-VIA-OWN-COMMAND plus P3-GATES-NOT-VIBES. The enumerated per-repo commands (CST/lodging pnpm chain, entabeni-api bundle exec rake, CNS jest+nyc+Cypress, RN yarn+Maestro) are Entabeni repo-profile data, not kit doctrine."}, "EL8-DOC-ONLY-CONFIG-AND-STALE-DOCTRINE": {"bucket": "covered", "cited_L": "L19", "stem": "DOC-ONLY-CONFIG-AND-STALE-DOCTRINE", "note": "L19-PINNED-INSTRUCTIONS-GO-STALE covers 'even a pinned CLAUDE.md/AGENTS.md is a summary at a point in time; check canonical source before a doctrine-heavy expansion; when two sources conflict, surface it, don't silently pick.' L21-CORRECT-AGAINST-CANON covers the sibling case (a correction citing a drifted sibling instead of true canon). The lodging OPEN_GDS_ARI_SOURCE dead-flag and CST's packet/ledger -1 drift are Entabeni-specific instances of the same general rule."}, "EL9-SPEC-CHANGES-NEED-A-PUSH": {"bucket": "covered", "cited_L": "L9", "stem": "SPEC-CHANGES-NEED-A-PUSH", "note": "Identical title and identical doctrine (a spec change mid-flight ships with a re-read directive to live workers; audit compliance against ground truth). P5-SPEC-CANONICAL-ALIVE's named scar."}, "EL10-PERMISSION-VS-HUMAN": {"bucket": "covered", "cited_L": "L10", "stem": "PERMISSION-VS-HUMAN", "note": "L10-PERMISSION-FRICTION is the same incident shape almost verbatim (two apparent denials were a harness config deny-rule, not the human declining; pre-authorize the toolchain, record the posture once)."}, "EL11-USE-BEFORE-DONE": {"bucket": "covered", "cited_L": "L12", "stem": "USE-BEFORE-DONE", "note": "Identical title/doctrine to L12-USE-BEFORE-DONE (a unit can pass every code gate without ever being exercised as a product; split dogfood into a machine-runnable slice and a human-checklist slice). EL11's specific rule 'a DOGFOOD.md existing proves nothing was run' is L16-DOGFOOD-IS-RUN-NOT-CHECKED verbatim."}, "EL12-THE-LEDGER-AND-THE-BUS-CAN-BE-THE-FAULT": {"bucket": "covered", "cited_L": "L22", "stem": "THE-LEDGER-AND-THE-BUS-CAN-BE-THE-FAULT", "note": "L22-THE-BUS-CAN-BE-THE-FAULT is the near-identical title and doctrine (a control bus a worker can't parse steers nobody; malformed-bus rule: STOP, blockers=BUS-CORRUPT). L7-PROVE-THE-BUS covers the companion 'transient tool failure vs persistent tool defect, appender re-validates parse' rules."}, "EL13-PRODUCT-VOCAB-AND-DEVICE-TRUTH": {"bucket": "engagement-specific", "cited_L": null, "stem": "PRODUCT-VOCAB-AND-DEVICE-TRUTH", "note": "Dominated by Entabeni's specific resort hardware fleet: PAX/IM30 terminals, Magicard printer mis-encodes, duplicate-chip RFID cards, Lynx/Orca UHF power clamps, the RfidProvider single-connection invariant, EFS/CFS clock-safety-net timer. This is client hardware-domain knowledge, not portable doctrine -- confirms prior triage. Note: EL13's rule 1 alone (a monitor's keyword filter misfires on a domain's own normal vocabulary) restates the ALREADY-covered general pattern in L17-DOMAIN-VOCAB-FALSE-POSITIVES, so even that slice needs no promotion -- it's doubly not-needed (both engagement-scoped AND separately already covered)."}, "EL14-DOCUMENT-UNITS-NEED-RENDER-GATES": {"bucket": "promoted", "cited_L": "L83", "stem": "DOCUMENT-UNITS-NEED-RENDER-GATES", "note": "CONFIRMED PROMOTION (new finding, not in prior triage): L83-DOCUMENTS-NEED-MACHINE-GATES-AND-THE-JUDGE-NEEDS-JUDGING carries the comment '# 2026-07-14, Entabeni EL14' and reproduces the identical incident verbatim (eight HTML/markdown deliverables, overflowing divs/SVG/tables/popovers, a bounds checker whose first run mis-flagged 33 then corrected to 29 true overflows). This is a direct promotion the prior triage missed / folded into 'covered'."}, "EL15-CAPABILITY-READS-ARE-CLAIMS": {"bucket": "promoted", "cited_L": "L81", "stem": "CAPABILITY-READS-ARE-CLAIMS", "note": "CONFIRMED PROMOTION (new finding): L81-A-CAPABILITY-API-ANSWER-IS-A-CLAIM carries the comment 'harvested from the Entabeni fleet (EL15)' and reproduces the identical incident (jira-support-bff: mypermissions said DELETE_ISSUES=true, real DELETE returned 403). Prior triage likely folded this into 'covered' rather than 'promoted'."}, "EL16-DELEGATED-DONE-IS-A-LOCATION-CLAIM": {"bucket": "promoted", "cited_L": "L77", "stem": "DELEGATED-DONE-IS-A-LOCATION-CLAIM", "note": "CONFIRMED PROMOTION (new finding): L77-A-DONE-IS-A-CLAIM-ABOUT-A-LOCATION carries the comment \"Entabeni's own EL16 -- the sharpest one\" and reproduces the identical incident (jira-support-bff Phase 2, subagent worktree, 'gate green, 81 tests' true-but-wrong-location, recovered via git reflog --all). Prior triage likely folded this into 'covered' rather than 'promoted'; the god-kit author explicitly called it out as the sharpest scar in the batch."}, "EL17-PARALLEL-LANES-CANNOT-SHARE-A-BRANCH": {"bucket": "covered", "cited_L": "L72", "stem": "PARALLEL-LANES-CANNOT-SHARE-A-BRANCH", "note": "Identical id-stem/title and identical doctrine already in canon: L72-PARALLEL-LANES-CANNOT-SHARE-A-BRANCH (one branch per concurrent unit; worker never self-merges; orchestrator merges after gate passes)."}, "EL18-DESTRUCTIVE-GIT-IS-NEVER-A-RESUME": {"bucket": "covered", "cited_L": "L65", "stem": "DESTRUCTIVE-GIT-IS-NEVER-A-RESUME", "note": "[sensitive content omitted]"}, "EL19-GATE-IS-THE-ONLY-DEFINITION-OF-DONE": {"bucket": "covered", "cited_L": "L73", "stem": "GATE-IS-THE-ONLY-DEFINITION-OF-DONE", "note": "L73-A-RED-GATE-MAY-ONLY-BE-A-STALE-INSTALL matches EL19's rule verbatim ('a red gate may be a stale install first: module-not-found is an environment hypothesis before a code hypothesis; every fresh worktree needs its own install'). The umbrella claim ('the gate is the only definition of done') is P3-GATES-NOT-VIBES; 'a done marker is a claim, audit against ground truth' is P1/L1. No new concept; EL19's machine-checkable ledger line-format detail (`landed: <ID> <sha>` at line start) is an Entabeni tooling implementation detail."}, "EL20-EVERY-GUARD-MUST-BE-RED-FLIP-PROVEN": {"bucket": "covered", "cited_L": "L20", "stem": "EVERY-GUARD-MUST-BE-RED-FLIP-PROVEN", "note": "L20-MUTATING-PROBES-AND-FLAKY-GATES rule 'never reactively edit a judging tool mid-investigation; a change to a correctness-judging tool must be tested against a KNOWN-GOOD case first' plus L51-TEST-THE-JUDGE-WHERE-IT-LIES ('every addition to a judging tool ships with its own known-good failure case') plus L83's rule 2 ('a freshly authored judge is untrusted until judged; run against known-good and known-bad fixtures') collectively cover this doctrine already. No new concept."}, "EL21-MOCKS-MUST-BE-SIZE-FAITHFUL": {"bucket": "covered", "cited_L": "L59", "stem": "MOCKS-MUST-BE-SIZE-FAITHFUL", "note": "L59-THE-MOCK-IS-INSTANT-AND-REALITY-IS-NOT's second rule states this near-verbatim: 'MOCKS ALSO COMPRESS SIZE, and that is the subtler killer: a fixture is small, so no test ever produces an output near the response ceiling... Test the BIG path: one fixture near the true ceiling, and size the ceiling to the WORKLOAD.' The jira-support-bff/lodging-api specific fixture guidance is Entabeni repo detail under the same general rule."}, "EL22-THE-REGISTER-MUST-BE-INJECTED": {"bucket": "covered", "cited_L": "L68", "stem": "THE-REGISTER-MUST-BE-INJECTED", "note": "L68-THE-REGISTER-IS-INJECTED-OR-IT-DOES-NOT-EXIST is the near-identical title and doctrine (a standing register must be pasted into every worker dispatch, not remembered). L71-A-TEMPLATE-UPDATED-IS-NOT-A-FLEET-UPDATED covers EL22's companion rules exactly (fixing the template does not fix the live fleet copy; on re-entry, diff the running apparatus against the template before issuing new directives)."}, "EL23-RECONCILE-THE-APPARATUS-NOT-JUST-THE-DOCTRINE": {"bucket": "promoted", "cited_L": "L122", "stem": "RECONCILE-THE-APPARATUS-NOT-JUST-THE-DOCTRINE", "note": "The known promotion, confirmed by direct read: L122-RECONCILE-A-FORKED-INSTANCES-APPARATUS-NOT-JUST-ITS-DOCTRINE carries the comment '2026-07-18, promoted from think-like-entabeni EL23 (live 2026-07-14)' and reproduces the identical tools/*.sh incident (gate.sh not sourcing project.env, audit-done.sh passing on the bare word 'walked', ATOMS_TOTAL_CMD substituting only {slug}, PyYAML-missing misread as bus corruption, pool.sh lacking session-detach/lease/pinned-CWD)."}, "EL24-A-SMALL-WINDOW-CHANGES-THE-CADENCE-NOT-THE-METHOD": {"bucket": "engagement-specific", "cited_L": null, "stem": "A-SMALL-WINDOW-CHANGES-THE-CADENCE-NOT-THE-METHOD", "note": "Confirms prior triage. The underlying MECHANISM this EL invokes (kernel/small-window.yaml, SW1-SW7: boot-thin, thresholds scale down, resume-as-you-go, delegate the window, atoms fit the session, know the reset cadence, revalidate detached state) is already full canon in the god kit, itself citing L94/L99/L88/L90/L100 as its scars -- and is a god-kit PRIVILEGE that Entabeni merely pins as a mandatory requirement for this specific machine. The EL entry itself is a machine/engagement-specific application record ('this machine's plan is below Pro/Max, so small-window mode is ON here') rather than new doctrine, so it should not be promoted as a fresh L -- the doctrine it leans on is already there under a different, more general name."}, "EL25-GREEN-GATE-IS-NOT-A-RENDERED-PAGE": {"bucket": "covered", "cited_L": "L87", "stem": "GREEN-GATE-IS-NOT-A-RENDERED-PAGE", "note": "L87-THE-CHECK-YOU-SKIPPED-IS-THE-ONE-THE-HUMAN-RUNS-FIRST is the same lesson shape, dated 2026-07-14 -- also a JiraSupport-family incident (green gate: 81 tests, eslint, tsc, all clean; the UI never actually rendered; a nav-key/route-segment mismatch broke every tab) predating this EL's 2026-07-17 recurrence (formatMessage i18n crash on tab render). Same doctrine: gate-green is a precondition, not proof a page renders; declare an unrendered surface a loud verification gap. L16-DOGFOOD-IS-RUN-NOT-CHECKED and L58-THE-WALK-IS-A-DEFECT-HUNT are adjacent covering lessons. NOTE: this EL shares the literal numeric id 'EL24' with EL24-A-SMALL-WINDOW-CHANGES-THE-CADENCE-NOT-THE-METHOD in the source think-like-entabeni/kernel/lessons.yaml -- an id collision in that file (the second one was very likely meant to be EL25); flagged here as a source-data defect, not something this ledger can resolve."}, "TL1-CONTROLLED-VOCABULARY-IS-CLOSED-NEVER-IMPROVISED": {"bucket": "engagement-specific", "cited_L": null, "stem": "CONTROLLED-VOCABULARY-IS-CLOSED-NEVER-IMPROVISED", "note": "Terminus's own closed taxonomy of review error-categories and dispositions is domain data specific to that review tool, not a portable orchestration rule."}, "TL2-GROUND-TRUTH-VERIFICATION-OVER-NARRATIVE": {"bucket": "covered", "cited_L": "L23", "stem": "GROUND-TRUTH-VERIFICATION-OVER-NARRATIVE", "note": "Same governing rule as L23-ACCEPT-THE-ARTIFACT-NOT-THE-SOURCE and P1-EMPIRICAL: execute/inspect the real artifact, never accept a log line or pasted report about it."}, "TL3-DECLARED-ENVIRONMENT-MUST-MATCH-DEMONSTRATED-NEED": {"bucket": "covered", "cited_L": "P1", "stem": "DECLARED-ENVIRONMENT-MUST-MATCH-DEMONSTRATED-NEED", "note": "'A metadata/mode flag is a claim until run and observed' is P1-EMPIRICAL applied to a config flag; sibling of L81-A-CAPABILITY-API-ANSWER-IS-A-CLAIM."}, "TL4-POLICY-GATE-BEFORE-QUALITY-REVIEW": {"bucket": "engagement-specific", "cited_L": null, "stem": "POLICY-GATE-BEFORE-QUALITY-REVIEW", "note": "The review-order discipline (policy/category gate before quality pass) plus 'category masking' naming are specific to terminus's task-review pipeline; P3 governs whether a gate exists, not this sequencing."}, "TL5-EXCEPTIONS-ARE-NAMED-AND-NARROW": {"bucket": "engagement-specific", "cited_L": null, "stem": "EXCEPTIONS-ARE-NAMED-AND-NARROW", "note": "Terminus's enumerated 'mechanical CI unblocker' exception list and grandfathering-needs-a-real-start-date rule are corrections-channel policy specific to that review queue."}, "TL6-NEWEST-DATED-GUIDANCE-WINS-ON-CONFLICT": {"bucket": "covered", "cited_L": "L19", "stem": "NEWEST-DATED-GUIDANCE-WINS-ON-CONFLICT", "note": "L19-PINNED-INSTRUCTIONS-GO-STALE rule 3 states the identical resolution (newer explicit supersedes; surface which source won); TL6 applies it to terminus's dated channel updates."}, "TL7-A-DECISION-COMMAND-IS-EXECUTED-NOT-NARRATED": {"bucket": "engagement-specific", "cited_L": null, "stem": "A-DECISION-COMMAND-IS-EXECUTED-NOT-NARRATED", "note": "Running ./done accept/revise/archive on an explicit operator command, forwarding params verbatim, is terminus's own CLI-completion procedure; no canonical lesson states this."}, "TL8-ROUTE-BY-FAILURE-CLASS-NOT-SYMPTOM": {"bucket": "covered", "cited_L": "L73", "stem": "ROUTE-BY-FAILURE-CLASS-NOT-SYMPTOM", "note": "Same rule as L73-A-RED-GATE-MAY-ONLY-BE-A-STALE-INSTALL: classify a failure's ORIGIN (infra vs artifact) before forming a verdict, apply the origin-specific remedy."}, "TL9-EVERY-CHECKLIST-ITEM-GETS-AN-EXPLICIT-DISPOSITION": {"bucket": "engagement-specific", "cited_L": null, "stem": "EVERY-CHECKLIST-ITEM-GETS-AN-EXPLICIT-DISPOSITION", "note": "Terminus's fixed 3-way disposition walked over its own review checklist is that tool's audit procedure; related in spirit to L26/P1 but not the specific per-item-disposition discipline."}, "TL10-SECONDARY-SOURCES-NEVER-SUBSTITUTE-FOR-PRIMARY-EVIDENCE": {"bucket": "covered", "cited_L": "P1", "stem": "SECONDARY-SOURCES-NEVER-SUBSTITUTE-FOR-PRIMARY-EVIDENCE", "note": "Core rule is P1-EMPIRICAL's shape, reinforced by L84 and L23; the bundled-equivalent clause is a mechanism-level elaboration under that same covered rule (parallel to EL18 under L65)."}, "TL11-OPERATIONAL-CONSTRAINTS-ARE-NOT-QUALITY-VERDICTS": {"bucket": "engagement-specific", "cited_L": null, "stem": "OPERATIONAL-CONSTRAINTS-ARE-NOT-QUALITY-VERDICTS", "note": "Terminus's revision-queue-cap-is-not-a-quality-blocker rule is that review process's capacity policy; related in spirit to L92 but a different pairing of axes (throughput vs quality), so kept instance-specific."}, "TL12-DECISION-OUTPUT-FOLLOWS-A-FIXED-TEMPLATE-CONVERSATIONALLY": {"bucket": "engagement-specific", "cited_L": null, "stem": "DECISION-OUTPUT-FOLLOWS-A-FIXED-TEMPLATE-CONVERSATIONALLY", "note": "The fixed heading/field-order/conversational-note conventions for terminus's decision artifacts are that tool's own output format; L71 addresses a different problem (template fix reaching the fleet)."}, "TL13-A-MISSING-RUBRIC-IS-AN-AUTOMATIC-REVISE-CHECKED-NOT-JUDGED": {"bucket": "engagement-specific", "cited_L": null, "stem": "A-MISSING-RUBRIC-IS-AN-AUTOMATIC-REVISE-CHECKED-NOT-JUDGED", "note": "Rejected: the rubric-none-is-a-mechanical-Rubric-blocker rule and the re-run-the-hard-gate-checklist-before-ACCEPT discipline are terminus review-pipeline procedure; the general mechanical-gate doctrine its incident seeded is already canon as L133. Adjudicated by the kit session 2026-07-29; drains the 9-item promotion queue."}, "TL14-AN-INTRA-SESSION-INCONSISTENCY-IS-A-SELF-AUDIT-TRIGGER": {"bucket": "covered", "cited_L": "L133", "stem": "AN-INTRA-SESSION-INCONSISTENCY-IS-A-SELF-AUDIT-TRIGGER", "note": "Folded into L133-A-MECHANICAL-GATE-IS-CHECKED-NEVER-JUDGED: same 2026-07-20 rubric-none incident family; a gate graded differently on identical inputs within one session is the checked-not-judged failure, and the re-audit-every-same-session-ACCEPT rule rides under it. Adjudicated by the kit session 2026-07-29; drains the 9-item promotion queue."}, "TL15-THE-FIRST-ACCEPT-AFTER-A-REJECT-STREAK-GETS-EXTRA-SCRUTINY": {"bucket": "engagement-specific", "cited_L": null, "stem": "THE-FIRST-ACCEPT-AFTER-A-REJECT-STREAK-GETS-EXTRA-SCRUTINY", "note": "Rejected: accept-hunger scrutiny after a REVISE streak is terminus review-cadence policy for that verdict queue, not portable kit doctrine. Adjudicated by the kit session 2026-07-29; drains the 9-item promotion queue."}, "TL16-A-CALLABLE-END-TO-END-SOLVER-IN-TESTS-IS-THE-DOMINANT-SUITE-DEFECT": {"bucket": "engagement-specific", "cited_L": null, "stem": "A-CALLABLE-END-TO-END-SOLVER-IN-TESTS-IS-THE-DOMINANT-SUITE-DEFECT", "note": "Rejected: the grep-tests-for-a-callable-reference-implementation pre-gate and its Test Alignment/Coverage category are terminus task-review pipeline specifics. Adjudicated by the kit session 2026-07-29; drains the 9-item promotion queue."}, "TL17-DEFAULT-SKELETON-PWD-GUARD-IS-A-PERMITTED-REWARD-EXCEPTION": {"bucket": "engagement-specific", "cited_L": null, "stem": "DEFAULT-SKELETON-PWD-GUARD-IS-A-PERMITTED-REWARD-EXCEPTION", "note": "Rejected: the Default_Task_Skeleton.zip PWD=/ guard and its reward-less-exit exception are terminus checklist policy data, not doctrine. Adjudicated by the kit session 2026-07-29; drains the 9-item promotion queue."}, "TL18-THE-REVIEW-FORMAT-IS-FIXED-NOT-REDESIGNED-PER-TASK": {"bucket": "engagement-specific", "cited_L": null, "stem": "THE-REVIEW-FORMAT-IS-FIXED-NOT-REDESIGNED-PER-TASK", "note": "Rejected: the fixed review outline, category placement, and ./check format gate are that tool's own output conventions (parallel to TL12). Adjudicated by the kit session 2026-07-29; drains the 9-item promotion queue."}, "TL19-D-S-V-NARRATIVES-ARE-NOT-THE-REBUTTAL": {"bucket": "engagement-specific", "cited_L": null, "stem": "D-S-V-NARRATIVES-ARE-NOT-THE-REBUTTAL", "note": "Rejected: scoping the rebuttal to the literal '## Rebuttal' section is terminus submission-format procedure. Adjudicated by the kit session 2026-07-29; drains the 9-item promotion queue."}, "TL20-REWARD-COVERAGE-MUST-APPLY-THE-SKELETON-WORKDIR-EXCEPTION": {"bucket": "engagement-specific", "cited_L": null, "stem": "REWARD-COVERAGE-MUST-APPLY-THE-SKELETON-WORKDIR-EXCEPTION", "note": "Rejected: TL17's sibling — the compare-guard/verify-WORKDIR/test-ordinary-path checklist for reward-coverage blockers is the same terminus skeleton policy data. Adjudicated by the kit session 2026-07-29; drains the 9-item promotion queue."}, "TL21-ALERT-LIVENESS-REQUIRES-VERIFIED-AUDIO-STARTUP": {"bucket": "promoted", "cited_L": "L194", "stem": "ALERT-LIVENESS-REQUIRES-VERIFIED-AUDIO-STARTUP", "note": "PROMOTED as L194-AN-ALERT-IS-PROVEN-BY-ITS-EFFECT-CHANNEL-NOT-ITS-LAUNCHER; the canon entry carries 'promoted from think-like-terminus TL21 (2026-07-29)' and reproduces the AudioQueueStart -66680 stale-marker incident. Adjudicated by the kit session 2026-07-29; drains the 9-item promotion queue."}}