IB-WP-0019-T05: state-hub token-event emission with failure isolation

Emit one record_token_event payload per completed generate run, derived
from the just-recorded usage rollup. tokens_in/out come from the
rollup, model defaults to the dominant model used (or "mixed" when
buckets disagree), agent="infospace-bench", ref_type="session", and
ref_id="<slug>/run-<run_index>". The note carries the infospace slug,
workspace, snapshot_id, and any known/estimated cost so the hub event
is self-describing.

Failure isolation: any exception from the HTTP poster (hub down,
timeout, 5xx) is caught, logged to stderr, and reported as
status=failed; the generate run still completes. INFOSPACE_BENCH_HUB_URL
overrides the default http://127.0.0.1:8000 base;
INFOSPACE_BENCH_DISABLE_HUB_TOKEN_EVENTS skips emission entirely.

Tests cover the happy path, the disable env var, poster failure, the
no-usage skip, multi-model coalescing to "mixed", and an end-to-end
run_generation against an unbindable hub port to prove the run survives
when the hub is unreachable. 116 tests pass.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

tests/test_budget_registry.py

@@ -489,6 +489,140 @@ def test_generation_report_includes_variance_line(tmp_path: Path) -> None:
     assert "calls" in report.lower()
 
 
+def test_emit_token_event_calls_poster_with_record_token_payload(tmp_path: Path) -> None:
+    from infospace_bench.budget import emit_token_event
+
+    calls: list[tuple[str, dict, float]] = []
+
+    def fake_poster(url: str, payload: dict, timeout: float) -> None:
+        calls.append((url, payload, timeout))
+
+    run_entry = {
+        "run_index": 2,
+        "snapshot_id": "abc123",
+        "rollup": {
+            "total_prompt_tokens": 1200,
+            "total_completion_tokens": 400,
+            "total_cost_usd_known": 0.0,
+            "total_cost_usd_estimated": 0.05,
+        },
+        "per_bucket": [
+            {"model": "openai/gpt-4o-mini", "total_tokens": 1600},
+        ],
+    }
+
+    result = emit_token_event(
+        run_entry,
+        infospace_slug="lefevre",
+        workspace="/tmp/workspaces/lefevre",
+        hub_url="http://hub.example",
+        poster=fake_poster,
+    )
+
+    assert result["status"] == "emitted"
+    assert len(calls) == 1
+    url, payload, timeout = calls[0]
+    assert url == "http://hub.example/state/token-events"
+    assert payload["tokens_in"] == 1200
+    assert payload["tokens_out"] == 400
+    assert payload["model"] == "openai/gpt-4o-mini"
+    assert payload["agent"] == "infospace-bench"
+    assert payload["ref_type"] == "session"
+    assert payload["ref_id"] == "lefevre/run-2"
+    assert "infospace=lefevre" in payload["note"]
+    assert "snapshot=abc123" in payload["note"]
+    assert timeout > 0
+
+
+def test_emit_token_event_respects_disable_env(monkeypatch, tmp_path: Path) -> None:
+    from infospace_bench.budget import HUB_DISABLE_ENV, emit_token_event
+
+    monkeypatch.setenv(HUB_DISABLE_ENV, "1")
+    calls: list = []
+    result = emit_token_event(
+        {"run_index": 1, "rollup": {"total_prompt_tokens": 100, "total_completion_tokens": 50}, "per_bucket": []},
+        infospace_slug="foo",
+        poster=lambda *a, **k: calls.append(a),
+    )
+
+    assert result["status"] == "disabled"
+    assert calls == []
+
+
+def test_emit_token_event_isolates_poster_failure(tmp_path: Path) -> None:
+    from infospace_bench.budget import emit_token_event
+
+    def angry_poster(url: str, payload: dict, timeout: float) -> None:
+        raise RuntimeError("hub down")
+
+    result = emit_token_event(
+        {
+            "run_index": 1,
+            "rollup": {"total_prompt_tokens": 50, "total_completion_tokens": 25},
+            "per_bucket": [{"model": "openai/gpt-4o-mini", "total_tokens": 75}],
+        },
+        infospace_slug="foo",
+        poster=angry_poster,
+    )
+
+    assert result["status"] == "failed"
+    assert "hub down" in result["reason"]
+
+
+def test_emit_token_event_skips_when_no_token_usage() -> None:
+    from infospace_bench.budget import emit_token_event
+
+    result = emit_token_event(
+        {"run_index": 1, "rollup": {"total_prompt_tokens": 0, "total_completion_tokens": 0}, "per_bucket": []},
+        infospace_slug="foo",
+        poster=lambda *a, **k: None,
+    )
+
+    assert result["status"] == "skipped"
+
+
+def test_emit_token_event_marks_multi_model_as_mixed() -> None:
+    from infospace_bench.budget import emit_token_event
+
+    captured: list[dict] = []
+
+    def fake_poster(url: str, payload: dict, timeout: float) -> None:
+        captured.append(payload)
+
+    emit_token_event(
+        {
+            "run_index": 1,
+            "rollup": {"total_prompt_tokens": 200, "total_completion_tokens": 100},
+            "per_bucket": [
+                {"model": "openai/gpt-4o-mini", "total_tokens": 150},
+                {"model": "anthropic/claude-3.5-haiku", "total_tokens": 150},
+            ],
+        },
+        infospace_slug="foo",
+        poster=fake_poster,
+    )
+
+    assert captured[0]["model"] == "mixed"
+
+
+def test_run_generation_never_fails_when_hub_is_down(tmp_path: Path, monkeypatch) -> None:
+    # Force the default hub URL to a known-bad port so the real poster fails fast.
+    from infospace_bench.budget import HUB_URL_ENV
+    from infospace_bench.generator import run_generation, status_generation
+
+    monkeypatch.setenv(HUB_URL_ENV, "http://127.0.0.1:1")  # reserved unbindable port
+    root = _build_infospace(tmp_path)
+    fixture = tmp_path / "responses.yaml"
+    _write_minimal_fixture(fixture)
+    plan_generation(root)
+
+    result = run_generation(root, fixture_responses=fixture)
+    status = status_generation(root)
+
+    assert result.status == "completed"
+    assert status["completed"] is True
+
+
 def test_plan_cli_writes_snapshot(tmp_path: Path) -> None:
     root = _build_infospace(tmp_path)
     env = os.environ.copy()