📖About

Two defects observed live 2026-07-13: (1) Albus RSI diagnoses (2558-03, 2558-04 meta, 2633-02) repeatedly end with diagnosis+recommendations recorded but WITHOUT the mandatory Albus-RSI-bypass-troubleshooting-decision YES/NO line, wedging the one-at-a-time lane until the 2-hour staleness default (CAROL-INI-2694) releases it — 2 hours of dead lane time per diagnosis. (2) The RSI loop logs an unattributed LLM-call rejection (no agent_id) EVERY tick, source unknown — all visible call sites pass attribution; likely the same failure that kills Albus decision-line writing. Scope: instrument the LLM gate rejection log with caller identification, catch the culprit live, fix the missing attribution, and make the decision-line write robust (record a deterministic decision line derived from the diagnosis verdict when the LLM lane dies mid-flow, so the lane never waits on a formality). Pipeline-only.

⚖️Decisions

• Auto-detected remediation target INI-999901307 from title/description scan (matched CAROL-INI-2694 -> row id 999901307 (CAROL-INI-2694-00: RSI diagnosis loop survives a missing bypass-decision row — s)); override by setting remediates_initiative_id explicitly at bypass_start. (system-auto-detect)
• Elrond's bypass methodology checklist (a reminder, not a gate -- you've got this): 0. File it requested_mode='bypass' (planner-vs-bypass is a deliberate choice). bypass_start REFUSES a non-bypass initiative (CAROL-INI-1846), and the dispatcher only skips the bypass lane when the mode says bypass -- a 'planner' mistag lets Merlin's pipeline grab the placeholder step and block your finished work. 1. Filed as planned status -- let the bypass claim/activate it; never file active. 2. Open the bypass (bypass_start) with your droid id + the remediation answer (remediates_initiative_id=NNN, or remediates_nothing=True). 3. Work the blocks for your work-type: template -> design -> code -> test -> review. Do the real work; record decisions on the initiative as you make them. 4. Reality is recorded for you at close -- code (files changed), each decision, and the twin-review verdict become real activities tied to this initiative and show in the Activity Tracker like a planner run (CAROL-INI-1840). No dummy rows. 5. Keep the initiative status moving; it parks in 'reviewing' and is tagged uat-pending for you at close (CAROL-INI-1836), so the stuck-watchdog leaves it alone until UAT. 6. Close runs the gates (design/architecture compliance + caller-audit). If a gate flags something pre-existing or unrelated to your change, waive it with a clear written rationale -- audit, don't skip. 7. Bypass skips the planner's auto-orchestration, NOT the standards. Same template checklist, same review, same observability as a planner run. (elrond)
• Current state at filing (Elrond validity check): The RSI diagnosis lane still experiences unattributed LLM-call rejections every tick and may still wedge for 2 hours when an Albus diagnosis dies before recording the bypass-decision line. No completed initiatives have resolved either defect. (elrond)
• [status-router] planned -> executing | event=bypass_executing | bypass transition (or-bx-01)
• Remediates INI-999901341: Orion bypass opened to remediate this blocked/closed parent. — CAROL-INI-696 linkage: this bypass is the operator-driven remediation of a previously-blocked-or-closed initiative. target_initiative_id set on this row; canonical Orion remediated: row will land on the parent at bypass_end. (shared.bypass.bypass_start)
• Root cause 1 (unattributed LLM call, every tick): dispatch_preflight.relevance_status had agent_id default None and its only caller (autoclose_if_stale, reached from the RSI loop queue reconciler via the dispatcher enqueue) passed nothing — the LLM attribution gate rejected the relevance check every dispatch tick, so premise re-validation silently never ran. Fix: default attribution ds-s1 (the Dispatcher, the actor everywhere else in that module). Found by instrumenting the gate rejection log with the caller stack (kept — rejections are now self-identifying). (orion)
• Root cause 2 (missing Albus decision line): the diagnosis harness expects the LLM RUN to write the Albus-RSI-bypass-troubleshooting-decision row itself via tools; the fleet provider skips tool instructions ~50%, nothing else wrote it, and the one-at-a-time lane wedged 2h per diagnosis until the 2694 staleness default (2558-03, 2558-04 meta, 2633-02 all hit this today). Fix: after the diagnosis commit, the harness checks for the row and derives a deterministic NO (diagnose-only) decision when absent — same semantics as the 2694 default, but immediate. Trust-the-LLM stays for the verdict content; bookkeeping is now harness-owned. (orion)
• Test: shared/test_ini2741.py (7 checks) registered in the persistent runner. (orion)
• Cookbook 457 reconciled with live semantics (sync-sst-on-discrepancy): decision marker corrected to the code wording (troubleshooting decision), NO now documented as diagnose-only-still-retriggers per the CAROL-INI-2423 ruling (old escalate-back-to-blocked line was stale), and the 2694 staleness backstop + 2741 deterministic harness fallback added as the DECISION-LINE RELIABILITY section. (orion)
• Criterion 1 VERIFIED LIVE: zero unattributed-LLM rejections across a ~7-tick watch window (was 1/minute for hours); the relevance check now passes the gate (provider=deepseek) and actually runs. Criterion 2: deterministic fallback shipped + unit-proven (7 checks); first live confirmation lands with the next fresh diagnosis — noted transparently for UAT. (orion)
• [status-router] executing -> reviewing | event=bypass_reviewing | bypass transition (or-bx-01)
• [status-router] reviewing -> closed | event=operator_signoff | Auto-accepted (CAROL-INI-1859): Orion-initiated, >2 days in reviewing with no objection. (el-srac-01)

✅Success criteria

• The every-minute unattributed LLM-call rejection in the diagnosis loop log stops (identified caller carries proper attribution), verified over at least 5 consecutive loop ticks. (must_have)
• An Albus diagnosis that dies before recording its YES/NO decision line no longer wedges the lane for 2 hours — the decision line is recorded deterministically from what the diagnosis already concluded, verified on the live lane. (must_have)
• A regression test covers both behaviors in the persistent runner. (must_have)