fix(ops): make scheduler policy target-driven

[Pigbibi]

Summary

• remove the account/service-name allowlist from scheduler planning
• derive live defaults and per-target overrides from each runtime target's JSON scheduler policy
• keep execution_mode=live at the existing 330s safe default while leaving paper targets unchanged unless configured
• replace deployment-specific account fixtures with generic examples

Open-source boundary

• no account count or service name is part of the code contract
• target cardinality is arbitrary (0/1/N)
• no live JSON values, account identifiers, project identifiers, tokens, or secrets are added
• no workflow, deployment, trading, order, position, or account behavior is triggered by this PR

Verification

• focused: 80 passed
• ruff: passed
• compileall: passed
• actionlint: passed
• diff-check: passed
• added-line sensitive-pattern scan: passed
• full local suite: 312 passed, 5 skipped, 11 failed; failures are in untouched existing test areas and are outside this PR's scheduler/config scope

[github-actions]

🤖 Codex PR Review

🚫 Merge blocked: 2 serious issue(s) found in high-risk files

🚫 Blocking Issues

These issues must be fixed before this PR can be merged:

1. 🟠 [HIGH] Logic in scripts/build_cloud_run_env_sync_plan.py

dry_run_only is being used as a fallback proxy for execution_mode. Those flags describe different things: a paper target can legitimately have dry_run_only=false because it still submits orders to a paper account. In that case, omitting execution_mode now makes the target look "live" and injects the 330s attempt deadline, which violates the PR contract to leave paper targets unchanged unless explicitly configured. (line 505)

Suggestion: Do not infer live/paper from dry_run_only. Only apply the default live deadline when execution_mode is explicitly present and validated as "live", or add a dedicated validated runtime field that actually represents the environment type.

2. 🟠 [HIGH] Reliability in scripts/build_cloud_run_env_sync_plan.py

After removing the hard-coded service allowlist, a target only gets the 330s safety deadline if execution_mode == "live" or dry_run_only is False. Any live target that omits both fields, or provides a non-boolean/invalid value, now silently falls through with no attempt_deadline, which regresses it to Cloud Scheduler's much shorter default and can cause premature timeout/retry behavior for long-running live jobs. (line 503)

Suggestion: Fail closed when the scheduler policy omits attempt_deadline: require a valid explicit execution_mode, or require scheduler.attempt_deadline to be set. Do not silently treat missing or malformed discriminator fields as non-live.

---

Review by Codex PR Review bot • PR

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: b38fb7ba66

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Derive live deadlines from dry_run_only too

For valid runtime targets that omit execution_mode, this branch never applies the 330s default even when dry_run_only is false. The pinned QuantPlatformKit resolve_runtime_target_from_env treats execution_mode as optional and derives live/paper from dry_run_only, so a live target JSON with dry_run_only: false but no execution_mode now produces no scheduler.attempt_deadline; the workflow then emits an empty deadline and skips --attempt-deadline for the live /run scheduler, reintroducing the scheduler deadline being shorter than the 300s Cloud Run timeout. Please normalize/derive the live mode from dry_run_only as well as the explicit string.

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: eebac97373

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Normalize live detection before defaulting deadlines

Fresh evidence: the pinned resolve_runtime_target_from_env accepts string booleans such as "false" for dry_run_only and compares execution_mode case-insensitively. For a valid live target that omits execution_mode but has "dry_run_only":"false" (or uses "LIVE"), this check treats the target as non-live, so the plan omits scheduler.attempt_deadline and the workflow skips --attempt-deadline for the /run scheduler even though the Cloud Run timeout is 300s.

Useful? React with 👍 / 👎.

[Pigbibi]

PERMANENT_FREEZE: the current runtime-target contract has no single validated field that safely distinguishes live from paper for scheduler deadline defaults. dry_run_only controls execution suppression and is not an environment identity; missing/invalid execution_mode also cannot safely default. Closure is exhausted, so this PR is preserved as evidence and will not be merged or resliced. A future change requires a concrete producer-owned runtime discriminator or an explicit per-target scheduler deadline contract.