[Birfy]

Required human contributor answers

Per the repo owner's standing instruction (Danyang, see #71 discussion), the intent items below are filled from issue #93 rather than waiting for a human answer.

1. What outcome are you trying to achieve?

   Answer: (from [Deps] Pin Jev on the direct TypeSafe path (it sends jev-latest) and re-run benchmarks on jev-1.14.0 #93) Saved runs and published numbers say which Jev answered them, and two runs a week apart don't differ just because jev-latest moved.

2. What is the highest-leverage variable for that outcome?

   Answer: The model id the direct TypeSafe path sends by default (jev-latest, a moving alias, with no override).

3. How will this work move that variable rather than create busywork?

   Answer: It pins the default to jev-1.13.0 (the version every recorded number used), lets TYPESAFE_MODEL choose another version on that path, and records the requested id in each step.

4. Who is the customer or user, and what do they actually need?

   Answer: Users with a direct TYPESAFE_API_KEY and anyone comparing benchmark runs; they need reproducible runs and to know which version answered.

After answering the four questions: is a PR needed at all to achieve this outcome?

Human decision and rationale (or explicit source reference): Yes, the default is a code constant in s1a/decision_models/wire.py (#93 scope items 1, 2 and 4).

For a bug fix: link actual observed failure evidence.

Evidence: Not a crash. Before this PR client_from_env("typesafe", ...) always used TYPESAFE_DEFAULT_MODEL = "jev-latest" and ignored TYPESAFE_MODEL (see the table in #93 and the old wire.py). TypeSafe's 1.14 post says jev-latest moves to jev-1.14.0 a week after 2026-10-14.

Why

Closes #93 (items 1, 2 and 4; item 3, the 1.13.0 vs 1.14.0 benchmark re-run, needs live TypeSafe keys and is left for a follow-up, see below).

The direct TypeSafe path sent the moving alias jev-latest, while the OpenRouter path and every published number use 1.13. TYPESAFE_MODEL was ignored on the direct path.

How

• s1a/decision_models/wire.py: TYPESAFE_DEFAULT_MODEL = "jev-1.13.0"; the typesafe backend reads TYPESAFE_MODEL and falls back to the pin. An OpenRouter id (typesafe/...) there is a MODEL_SERVICE_CONFIG_ERROR that says to use a TypeSafe name or set TYPESAFE_API_URL, instead of a request TypeSafe would reject.
• s1a/decision_models/jev.py + types.py: JevModel keeps requested_model in the reply's raw payload, and it is a provenance key, so tool-front ticks get it through decision.provenance. model stays the id the response reports, else the one sent.
• s1a/browser/decision_model.py: browser ticks now record model and (for Jev) requested_model.

What

• Default direct-path model: jev-latest changes to jev-1.13.0. Set TYPESAFE_MODEL=jev-1.14.0 to try 1.14.
• New step-record fields: requested_model (tool and browser ticks), model (browser ticks).
• Docs: docs/configuration.md (both defaults, override, 60-day deprecation notice for 1.13.0), .env.example, docs/why.md, docs/browser-front.md, a note in docs/benchmarks.md that the 2026-09-19 jev-latest run was on 1.13; CHANGELOG under Changed.
• Kept 1.13.0 as the default for now: [Deps] Pin Jev on the direct TypeSafe path (it sends jev-latest) and re-run benchmarks on jev-1.14.0 #93 says to move to 1.14.0 only after the re-run, which this PR does not do.

Verification

• uv run ruff format --check . && uv run ruff check . && uv run ty check: all pass
• uv run pytest -q (996 passed, 127 skipped) and scripts/smoke.sh (exit 0)
• CHANGELOG.md and the docs say what the code does now

New tests: no default is jev-latest/jev-preview; TYPESAFE_MODEL overrides the pin on the direct path and the empty value falls back; an OpenRouter id on the direct path is a config error (in tests/test_decision_models_jev.py and tests/test_browser_policy.py); requested_model is recorded next to a response-reported model; browser ticks carry both.

Demo / evidence

• Video link, or exemption/blocker with follow-up: Exempt (config default and record fields, offline-tested; no behaviour change in an agent run beyond which model id is sent).
• Application/task → agent actions → independently checked result: N/A
• System1-Agents / System1-Omni commits, checkpoint and worker/frontend provenance: base main 5693f51. Follow-up for [Deps] Pin Jev on the direct TypeSafe path (it sends jev-latest) and re-run benchmarks on jev-1.14.0 #93 item 3: re-run public231, s1a probe and the --order reverse flip rate on jev-1.13.0 and jev-1.14.0 with a TypeSafe key, then decide whether to move the default.

[hsliuustc0106]

Maintainer review (light-impact + self-review)

Tip: 62ebc820da90b903c1456b467df8d225d40cb55f · MERGEABLE · required checks green on this head (core 3.11/3.13, full, benchmarks).

Impact

• changed: TYPESAFE_DEFAULT_MODEL pin, client_from_env honouring TYPESAFE_MODEL, requested_model provenance
• callers/dependents (first hop): JevModel.from_env, browser tick recording via Decision.provenance
• related tests: tests/test_decision_models_jev.py, tests/test_browser_policy.py
• not expanded: OpenRouter proxy path beyond existing override behaviour

Risk: low · Missing tests: none obvious · Regression: default id change for direct TypeSafe callers who relied on moving jev-latest (intentional; override documented).

Self-review readiness: Template answers traceable to #93; demo N/A appropriate for config/provenance chore. No blocking gaps.

Verdict: APPROVE — correct, scoped, CI green on tip.