pi-live: operator model override + explicit parallel pin - #725
Open
clkao wants to merge 2 commits into
Open
Conversation
added 2 commits
August 17, 2026 09:35
The shared journey driver passed the auth decision's default model (openai-codex/gpt-5.6-luna:max) to child pi, ignoring the operator's SPACEDOCK_PI_LIVE_CHILD_MODEL override, which was wired only into the front-door smoke fixture. Operator-mirrored auth.json + models.json already make custom providers resolve, so route both paths through one piLiveChildModel helper. CI keeps its default: the workflow runtime-live-e2e.yml never sets the override (pinned by runtime_live_evidence_workflow_test).
The pi-live lane was the only common-journey lane omitting -parallel, inheriting GOMAXPROCS (4 on standard runners) implicitly, while claude-live and codex-live pin -parallel 3. Add the live_parallel dispatch input (default 4, the current effective concurrency), expose it as SPACEDOCK_PI_LIVE_PARALLEL, and pass it to go test after a positive-integer guard. Pins: the workflow evidence test now asserts the input default, the job env binding, the step flag, and carries a mutation control for the flag loss; the local guide's pi command carries the same pin so the contractlint run-shape reconciliation (timeout/parallel/failfast) holds.
clkao
force-pushed
the
pi-live-model-parallel-overrides
branch
from
August 17, 2026 16:55
3a827c0 to
f74a3d4
Compare
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
What
Two pi-live lane control surfaces, both previously missing or implicit:
Model override for common journeys —
SPACEDOCK_PI_LIVE_CHILD_MODEL(provider/model:thinking) was honored only by the front-door smoke fixture. The shared journey driver passed the auth decision's default (openai-codex/gpt-5.6-luna:max) to child pi, so the 17 common journeys could not be re-run against any other provider/model. Both paths now route through onepiLiveChildModelhelper. Unset, behavior is unchanged (CI derives the model from auth mode, as the workflow evidence test pins).Explicit
-parallelpin — pi-live was the only common-journey lane omitting-parallel, inheriting GOMAXPROCS (4 on standard runners) implicitly while claude-live and codex-live pin-parallel 3. The lane now takes thelive_paralleldispatch input (default"4", the current effective concurrency), exposed asSPACEDOCK_PI_LIVE_PARALLEL, validated as a positive integer in the step, and passed togo testafter-failfast.Pins
Pi parallel override lostmutation control.docs/runtime-live-ci.md): the pi common-suite command now carries the same-parallel "$PI_LIVE_PARALLEL"pin, keeping the contractlint run-shape reconciliation (timeout/parallel/failfast) green.Evidence
go test ./...andgo test ./... -racegreen on this branch (20 packages, 0 failures, incl.internal/cli167s under race).GateGuardrail,RecordedGateLifecycle,DefaultHeadlessGateStop,KeepMovingPosture) plusRejectionFlow(CI timeout) all PASS locally onmodal/Qwen/Qwen3.8-27B:xhigh, run from the stack tip withSPACEDOCK_PI_LIVE_CHILD_MODELset — the exact configuration that failed at launch (No API key found for openai-codex) before this change.