docs: confirm the family_cap/max_tokens fixes work end-to-end (one further gap remains open)

[seonghobae]

Summary

Item 1 of the post-#1434-merge priority list: verify end-to-end that noema-review/opencode-review/strix actually recover now that main has both #1436's max_tokens fix and #1434's family_cap raise — not just structurally argued.

Update (softened from this PR's original "confirmed resolved" title/framing): after opening this PR, a second, independently-found, timestamped counter-example surfaced and was verified directly against its own job log — a strix run on .github's own main (job 99247611184, completed 2026-08-30T11:16:17Z, predating this PR's original claim) failed with a genuinely separate, unfixed defect: Strix's OpenAI Agents SDK client sends stream_options.include_usage=true together with tools/response_format, and whichever orchestrator/free candidate it routed to rejected that exact combination with HTTP 400 invalid_stream_options on all retries, exhausting the pool (STRIX_PROVIDER_UNAVAILABLE). This is a request-shape incompatibility, not a candidate-selection or token-budget problem — genuinely distinct from what #1434/#1436 fixed. Root-causing it has been delegated elsewhere and is not duplicated in this PR.

Accurate combined claim, checked against both pieces of evidence:

• The family_cap/max_tokens fixes are confirmed working end-to-end on ContextualWisdomLab/contextual-orchestrator#921 (unrelated to this investigation, picked because it doesn't itself hit the pull_request_target self-test trust boundary fix(zdr): cite NVIDIA's own Trial ToS for the nvidia_nim not-ZDR classification #1434/fix(strix): add missing github-token fallback for self-target status publish #1441 do): noema-review and strix both succeeded on a fresh head. noema-review's job log confirms a real, complete sidecar cycle (not vacuous): CONTEXTUAL_ORCHESTRATOR_BASE_URL was exported, which only happens after the sidecar's own preflight found a viable route and the server became healthy — the exact stage that failed 100% of the time (three independent reproductions) before these fixes landed.
• orchestrator/free is not reliable in general yet. With family_cap=8 giving more provider diversity per run, which candidate a given run draws varies, and at least one candidate family has the separate, live, unfixed stream_options gap above that can still fail Strix (though evidently not every candidate — contextual-orchestrator#921's own strix run succeeded). Two permanently-retired google/gemma-3-*-it model ids are also still admitted into the pool and will still individually fail when the alphabetical sort reaches them.

Also includes a second correction (real Devin finding, verified against the actual merged diff): updates the pingora_edge_policy.py binary-evidence entry from "two competing open fixes, unresolved" to reflect that #1435 already merged — a third, better implementation that network-verifies the real %PDF- magic prefix rather than trusting the .pdf extension alone.

Test plan

Docs-only change (plus CHANGELOG bullets). tests/test_pr_governance_audit_contract.py passes.

🤖 Generated with Claude Code

https://claude.ai/code/session_015Gs7KmNvH75nxz1sL8mKjw

[coderabbitai]

Important

• 🔍 Trigger review

This repository does not receive automatic reviews because it has fewer than 10 stars.

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro Plus

Run ID: 1e397d86-b974-4189-9f3b-ef64f37ba682

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}