Skip to content

feat(lanes): v2 — Fable planner + Sonnet Lite + escalation protocol (#385)#401

Merged
BaseInfinity merged 2 commits into
mainfrom
feat/lanes-v2-fable-sonnet
Jun 11, 2026
Merged

feat(lanes): v2 — Fable planner + Sonnet Lite + escalation protocol (#385)#401
BaseInfinity merged 2 commits into
mainfrom
feat/lanes-v2-fable-sonnet

Conversation

@BaseInfinity

Copy link
Copy Markdown
Owner

Summary

Lane restructure based on Fable 5 research (2 review tests, 8 unique findings vs Codex, cross-model verified):

  • Setup A: Fable 5 high planner, Opus 4.6 max driver, GPT-5.5 xhigh reviewer + Fable escalation for high-risk
  • Setup B: unchanged (OpusPlan + Sonnet + Codex)
  • Setup C: Sonnet replaces Haiku — same model as B driver, one less model to manage

Why this structure (both Fable and Codex agree)

  • Fable plans best (proven: found symlinks, correct file counts, deeper architecture analysis)
  • GPT-5.5 reviews (cross-family catches plan-level blind spots — proven: caught 2 things Fable missed)
  • Fable reviewing its own plans would miss plan-level flaws (both reviewers independently flagged this)
  • Escalation is rare (~5% of PRs): security, releases, wide-blast architecture only

Test plan

  • doc-consistency: 42/42
  • release-dry-run: 25/25

Setup A: Fable 5 high plans, Opus 4.6 max drives, GPT-5.5 xhigh reviews
  + Fable escalation for security/release/architecture (~5% of PRs)
Setup B: Opus plan mode + Sonnet driver + GPT-5.5 xhigh (unchanged)
Setup C: Sonnet driver (was Haiku) — same model as B, one less to manage
… + AGENTS.md

P1: billing section still described Setup A as Opus planner+driver
P1: credit warning still cited Haiku pricing for Setup C
P2: AGENTS.md still had old lane descriptions
@BaseInfinity BaseInfinity merged commit 54d06e6 into main Jun 11, 2026
4 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant