chore(orchestration): rule 57 within-WI fan-out + ANDROID_SERIAL routing + android lease - #2015
Open
lllyys wants to merge 2 commits into
Open
chore(orchestration): rule 57 within-WI fan-out + ANDROID_SERIAL routing + android lease#2015lllyys wants to merge 2 commits into
lllyys wants to merge 2 commits into
Conversation
…ing + android lease Post-#138 orchestration follow-ups (the 'why so few subagents' analysis, grilled): - NEW rule 57 (ultracode within-WI analytical fan-out): parallel context sweep + pre-write adversarial 'false-green' brainstorm (cost-asymmetry-justified, instrumented hypothesis) + deep+broad post-write audit in parallel; write + reconcile stay solo (reconcile now requires written rationale per accepted finding - the step that leaked WI-6's targets-vs-ceilings defect); named follow-ups: mechanize logged-vs-asserted check, lens falsification discipline. - rule 48: analytical fan-out prices differently than worktree lanes (no setup/ conflict/resource when overlapping or pre-write; net-positive IN EXPECTATION at N=1 under ultracode - honest ledger incl. reconcile tax + infra failure). - ANDROID_SERIAL routing in run-android-tests.sh (validate + export + all-device ambiguity guard incl. physical) + sim-lease.sh 'acquire android' emulator- serial lease (capacity = online count, race-safe under the select mutex, no silent set-e exit) -> a 2nd AVD unlocks width-2 Android dispatch. Self-tests extended: 10 routing cases + 6 android-lease cases, ALL PASS. - rules 52/55 + dispatch/feature-workflow skills synced end-to-end (lease-by- platform, ANDROID_SERIAL in briefs/test gates/re-runs/cleanup, ledger device column) - 3-round Codex loop chased the propagated-consistency findings. Audits: Codex gpt-5.5/high 3 rounds (R1 block: skill wiring + physical-device ambiguity + set-e silent exit -> fixed; R2: rule 55 brief template -> fixed; R3: lease-release wording + platform-routing summary -> fixed; 3-round cap reached, remaining escalation = the PR review itself) + a 3-lens adversarial workflow (fix-before-merge -> fixed) + a full 5-angle reason-grill of the rationale (HOLDS-IF-NARROWED -> over-claims corrected in rule text: hindsight counterfactual -> instrumented hypothesis; zero-wall-clock -> in-expectation; 'ONLY real gaps' -> 'gaps this session exposed').
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Orchestration follow-ups from the "why so few subagents?" analysis (post-#138)
Three additive changes — no redesign of
/dispatch(the grilled diagnosis: lane-level low fan-out was structurally forced by the N=1 degrade + single-emulator serialization; the movable margin is within-WI analytical work).1. NEW rule 57 — ultracode within-WI analytical fan-out
Parallel context sweep + pre-write adversarial "how could this false-green?" brainstorm + deep+broad post-write audit in parallel. Write + reconcile stay solo; the reconcile now requires a written rationale per accepted/deflected finding (the step that leaked WI-6's targets-vs-ceilings defect). Pre-write discoverability is framed as an instrumented hypothesis justified on cost asymmetry — not a claimed counterfactual save. Named follow-ups: mechanize the logged-vs-asserted check; per-lens falsification discipline.
2. rule 48 — honest pricing for analytical fan-out
Read-only analytical spawns price differently than worktree lanes (no setup/conflict/resource when overlapping the serial critical section or preceding the write) → net-positive in expectation at N=1 under ultracode, with the residual costs named (reconcile tax, infra failure, context dilution).
3. ANDROID_SERIAL routing +
androidemulator leaserun-android-tests.sh: validate + exportANDROID_SERIAL, ambiguity guard counting ALL online devices (emulator + physical).sim-lease.sh acquire android: leases an online emulator serial (capacity = online count, race-safe under the existing select mutex, no silentset -eexit). Rules 52/55 + dispatch/feature-workflow skills synced end-to-end (lease-by-platform, serial in briefs/test-gates/re-runs/cleanup). A 2nd AVD now unlocks width-2 Android dispatch. Self-tests extended (10 routing + 6 lease cases) — ALL PASS.Audit trail
Meta-process change (no bug/feature row). Version bumped 3.67.7 → 3.67.8.
Do not merge yet — held for review per operator instruction.