WS4 IRL re-audit: Fable orchestration economics post-v3 (~2026-07-22) #8

Open
opened 2026-07-08 15:19:08 +00:00 by jared · 1 comment
Owner

Two weeks after shipping ORCHESTRATION.md v3 (2026-07-08, commit 397145b), re-run the Fable economics mini-audit on ~5 NEW real sessions per the IRL feedback loop (eval-methodology-irl-feedback-loop):

  • Method: docs/orchestration-audit/2026-07-08-fable-econ-miniaudit.md — but compute REAL main-loop share this time: subagent transcripts persist at ~/.claude/projects///subagents/agent-*.jsonl with per-message usage blocks (vault note claude-code-subagent-transcripts-and-token-accounting).
  • Compare main-loop token share + miss shapes vs the 2026-07-08 baseline (implementation fan-out was the dominant miss; mid-session threshold decay the watch item).
  • Promote any NEW miss shapes into eval scenarios (run-set AND reserve are contaminated — new scenarios needed for any future wording tuning).
  • Baseline + shipped-wording results: vault note os-orchestration-ws4-econ-wording-results; plan: docs/plans/ws4-orchestration-economics.md.
Two weeks after shipping ORCHESTRATION.md v3 (2026-07-08, commit 397145b), re-run the Fable economics mini-audit on ~5 NEW real sessions per the IRL feedback loop ([[eval-methodology-irl-feedback-loop]]): - Method: docs/orchestration-audit/2026-07-08-fable-econ-miniaudit.md — but compute REAL main-loop share this time: subagent transcripts persist at ~/.claude/projects/<flat-cwd>/<session-id>/subagents/agent-*.jsonl with per-message usage blocks (vault note [[claude-code-subagent-transcripts-and-token-accounting]]). - Compare main-loop token share + miss shapes vs the 2026-07-08 baseline (implementation fan-out was the dominant miss; mid-session threshold decay the watch item). - Promote any NEW miss shapes into eval scenarios (run-set AND reserve are contaminated — new scenarios needed for any future wording tuning). - Baseline + shipped-wording results: vault note [[os-orchestration-ws4-econ-wording-results]]; plan: docs/plans/ws4-orchestration-economics.md.
jared added the
ready-for-agent
label 2026-07-12 22:25:52 +00:00
Author
Owner

This was generated by AI during triage.

Labeled ready-for-agent with a time gate: do not run before ~2026-07-22 (needs two weeks of NEW post-v3 sessions to audit). The run itself is agent-executable via /os-orchestration:audit-sessions per the method doc; findings review is the human step. Autonomy: afk-ready once the date gate passes.

> *This was generated by AI during triage.* Labeled ready-for-agent with a time gate: do not run before ~2026-07-22 (needs two weeks of NEW post-v3 sessions to audit). The run itself is agent-executable via /os-orchestration:audit-sessions per the method doc; findings review is the human step. Autonomy: afk-ready once the date gate passes.
Sign in to join this conversation.
No description provided.