📖About

Both Forge builds and Albus diagnoses die on a 300k-token cap after ~5-15 rounds. Root cause: the shared LLM loop re-sends the whole growing transcript each round and the cap counts cumulative re-billed prompt tokens (though its own comment calls it a fresh-input ceiling), so the same fat starting brief is re-counted every round. Fix: (A) count only genuinely-new input tighten per-round compaction so requests stay lean (helps the lane overload that caused Albus to error out). (B) slim the starting briefs (Albus diagnosis context: fewer decisions/family diagnoses/shorter description; Forge build brief). Keep billed-token cost accounting unchanged.

⚖️Decisions

• 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)
• [status-router] planned -> executing | event=bypass_executing | bypass transition (or-bx-01)
• [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 shared LLM tool loop no longer aborts a run with SCOPE_TOO_BIG just because the same brief was re-sent each round - the input ceiling counts genuinely new content pulled in, not re-billed context, matching its stated fresh-input intent. (must_have)
• Each round sends a leaner request - old tool results are compacted to a short takeaway and only the recent rounds stay verbatim, so single requests stay well below the level that overloads the model lane. (must_have)
• Albus troubleshooter and Forge build starting briefs are slimmed so they do not re-send tens of thousands of tokens of history every round. (must_have)
• A multi-round investigation that re-reads the same brief completes instead of dying at the cap, while a run that genuinely pulls in too much new data still stops honestly. (must_have)
• The shared LLM loop no longer aborts with SCOPE_TOO_BIG merely because the same brief is re-sent each round: the input ceiling counts genuinely-new content, not re-billed context. (must_have)
• Each round sends a leaner request (old tool results compacted to a short takeaway, recent rounds verbatim), keeping single requests well below lane-overload size. (must_have)
• Albus diagnosis and Forge build starting briefs are slimmed so they do not re-send tens of thousands of tokens of history each round. (must_have)
• A multi-round run that re-reads the same brief completes; a run pulling in genuinely too much NEW data still stops honestly. (must_have)