emrg: usage-anchor stats — attach provider/model identity to loss/drift events (#1011)

Agent.md

@@ -119,7 +119,7 @@ Community needs voiced in HN agent-UI discussions map directly to EMRG's design:
 pkill -f "emrg.server"; rm -f ~/.emrg/emrgd.token; python -m emrg
 ```
 
-Python: `uv run pytest tests/ -v` (1106) — import check: `uv run python -c "from emrg.client.app import run_client"`
+Python: `uv run pytest tests/ -v` (1107) — import check: `uv run python -c "from emrg.client.app import run_client"`
 GUI: `cd emrg/gui && npm test` (265: 45 daemon_client + 20 conn-manager + 22 app-commands + 132 renderer smoke + 15 i18n + 8 integration + 3 commands + 8 build-config + 7 gui-state + 2 tool-group + 3 preload-api) — syntax: `node --check main.js preload.js daemon_client.js renderer/js/*.js`
 Renderer: `cd emrg/gui/renderer && npm run typecheck && npm test` (262: 5 snapshot-store + 9 utils + 3 ErrorBoundary + 2 App smoke + 11 commands + 4 copywriting + 11 i18n + 11 markdown + 15 transcript + 7 TranscriptView + 15 history + 22 composer + 14 Composer + 12 sidebar + 10 Sidebar + 9 fileTree + 9 FileTree + 16 resultPanel + 8 ResultPanel + 27 workspaceView + 8 WorkspaceView + 10 dialog + 6 Dialog + 9 ConfirmDialog + 9 RenameDialog) + `npm run build` → `renderer/dist/`
 CI: `uv run pytest` (ubuntu + **windows-2025 matrix** — Windows pytest 回归在 PR CI 即失败，v0.2.29 教训 #725) + GUI tests + **actionlint workflow lint** (`rhysd/actionlint@v1.7.12` gate, #444 — workflow 解析错误在 PR CI 即失败，如 `if:` secrets 上下文)

emrg/server/daemon.py

@@ -28,6 +28,7 @@
 from datetime import datetime
 from pathlib import Path
 from typing import Optional
+from urllib.parse import urlparse
 
 import yaml
 from websockets.asyncio.server import serve
@@ -259,7 +260,7 @@ def __init__(self, llm_config: LlmConfig) -> None:
  # (estimated_tokens, iso_ts) at anchor-loss warning time, so the
  # estimator drift of the anchor-less window can be measured when
  # the provider re-anchors (countable metric, not just a log line).
- self._missing_anchor_est: dict[str, tuple[int, str]] = {}
+ self._missing_anchor_est: dict[str, tuple[int, str, str, str]] = {}
  # Rant 2026-08-23T13:54:14: per-session dynamic-context snapshot
  # (session_id → (text, injected_ms)) so unchanged context is not
  # re-sent (context_refresh_interval_ms gating).
@@ -3023,13 +3024,23 @@ def _warn_missing_usage_anchor(
  return # already warned once for this session
  self._warned_missing_usage_anchor.add(session.session_id)
  loss_ts = datetime.now().astimezone().isoformat()
- self._missing_anchor_est[session.session_id] = (estimated, loss_ts)
+ loss_model = self.llm.config.model
+ loss_provider = _provider_slug(self.llm.config.base_url)
+ # Issue #1011 (heinrichneb): a loss window without provider identity is
+ # half a counter — after a mid-session model switch the drift file must
+ # say WHICH provider went silent. Store the loss-time identity so a
+ # later anchor_drift can attribute the window.
+ self._missing_anchor_est[session.session_id] = (
+ estimated, loss_ts, loss_model, loss_provider,
+ )
  try:
  _append_usage_anchor_event({
  "type": "anchor_loss",
  "timestamp": loss_ts,
  "session": session.session_id,
  "est": estimated,
+ "model": loss_model,
+ "provider": loss_provider,
  })
  except OSError as exc:
  logger.warning("usage-anchor stats append failed: %s", exc)
@@ -3055,7 +3066,7 @@ def _record_anchor_drift(self, session, real_pt: int) -> None:
  stored = self._missing_anchor_est.pop(session.session_id, None)
  if not stored:
  return
- est_before, loss_ts = stored
+ est_before, loss_ts, loss_model, loss_provider = stored
  try:
  _append_usage_anchor_event({
  "type": "anchor_drift",
@@ -3064,6 +3075,12 @@ def _record_anchor_drift(self, session, real_pt: int) -> None:
  "real_after": real_pt,
  "delta": real_pt - est_before,
  "loss_ts": loss_ts,
+ # Issue #1011: attribute the window — who went silent
+ # (loss identity) vs who re-anchored (current identity).
+ "loss_model": loss_model,
+ "loss_provider": loss_provider,
+ "model": self.llm.config.model,
+ "provider": _provider_slug(self.llm.config.base_url),
  })
  except OSError as exc:
  logger.warning("usage-anchor stats append failed: %s", exc)
@@ -4268,6 +4285,21 @@ def _write_exit_record(reason: str, exit_code: int, traceback_text: str | None)
 _USAGE_ANCHOR_STATS_PATH = Path.home() / ".emrg" / "logs" / "usage-anchor.jsonl"
 
 
+def _provider_slug(base_url: str) -> str:
+ """Derive a provider identity from the LLM base_url (issue #1011).
+
+ Deterministic hostname extraction — 'https://api.openai.com/v1' →
+ 'api.openai.com'; a local endpoint ('http://localhost:11434/v1') →
+ 'localhost'. Empty/unparsable input falls back to the raw string so the
+ event still carries something attributable. No heuristics, no DNS.
+ """
+ try:
+ host = urlparse(base_url).hostname
+ return host or base_url
+ except Exception:
+ return base_url
+
+
 def _append_usage_anchor_event(record: dict, path: Path | None = None) -> int:
  """Append a countable usage-anchor event (one JSON line) and return the
  cumulative total (community feedback 2026-08-26T07:18:29 — heinrichneb:

tests/test_daemon.py

@@ -710,6 +710,9 @@ def test_usage_anchor_loss_event_is_countable(tmp_path, monkeypatch):
  e1, e2 = json.loads(lines[0]), json.loads(lines[1])
  assert e1["type"] == "anchor_loss" and e1["session"] == "s1"
  assert e1["total"] == 1 and "timestamp" in e1
+ # Issue #1011: the loss event must say WHICH provider went silent
+ assert e1["model"] == "gpt-4o-mini" # default LlmConfig model in _make_server
+ assert e1["provider"] == "localhost" # base_url host of _make_server
  assert e2["session"] == "s2" and e2["est"] == 60_000 and e2["total"] == 2
 
 
@@ -759,12 +762,66 @@ def test_usage_anchor_drift_measured_on_reanchor(tmp_path, monkeypatch):
  assert drift["real_after"] == 180_000
  assert drift["delta"] == 80_000
  assert drift["loss_ts"] == loss["timestamp"]
+ # Issue #1011: drift window is attributable — loss identity (who went
+ # silent) + re-anchor identity (who finally reported real tokens)
+ assert drift["loss_model"] == loss["model"] == "gpt-4o-mini"
+ assert drift["loss_provider"] == loss["provider"] == "localhost"
+ assert drift["model"] == "gpt-4o-mini"
+ assert drift["provider"] == "localhost"
 
  # Re-anchor measured once — a second call (no pending loss) is a no-op
  server._record_anchor_drift(_SidSession(sid), 200_000)
  assert len(stats.read_text(encoding="utf-8").strip().splitlines()) == 2
 
 
+def test_usage_anchor_cross_provider_window_attributable(tmp_path, monkeypatch):
+ """Issue #1011 — heinrichneb: 'a counter that can't say WHICH provider
+ went silent is half a counter'. The loss window's identity must survive
+ from anchor_loss to anchor_drift: a mid-session model switch produces a
+ drift window whose loss side names the silent provider and whose re-anchor
+ side names the new provider."""
+ stats = tmp_path / "usage-anchor.jsonl"
+ monkeypatch.setattr(daemon_mod, "_USAGE_ANCHOR_STATS_PATH", stats)
+ server = _make_server()
+ sid = "x-provider"
+ messages = [
+ {"role": "assistant", "content": "x"},
+ {"role": "user", "content": "y"},
+ ]
+ # Loss detected under provider A (deepseek)
+ server.llm.config.model = "deepseek-chat"
+ server.llm.config.base_url = "https://api.deepseek.com/v1"
+ server._warn_missing_usage_anchor(_SidSession(sid), messages, 50_000)
+
+ # Model switch to provider B (openai) — #1003 invalidates anchors; the
+ # pending drift window is dropped, so no cross-provider drift is measured.
+ # But the anchor_loss event itself must still name provider A.
+ server.llm.config.model = "gpt-4o"
+ server.llm.config.base_url = "https://api.openai.com/v1"
+ server._invalidate_usage_anchors_on_switch()
+
+ lines = stats.read_text(encoding="utf-8").strip().splitlines()
+ assert len(lines) == 1
+ loss = json.loads(lines[0])
+ assert loss["type"] == "anchor_loss"
+ assert loss["model"] == "deepseek-chat"
+ assert loss["provider"] == "api.deepseek.com"
+
+ # Same-provider window: loss + re-anchor both under provider B (fresh
+ # session — warn-once is per session lifetime, so a new window needs a
+ # new session id to emit a second anchor_loss)
+ server._warn_missing_usage_anchor(_SidSession("x-provider-2"), messages, 60_000)
+ server._record_anchor_drift(_SidSession("x-provider-2"), 150_000)
+ lines = stats.read_text(encoding="utf-8").strip().splitlines()
+ assert len(lines) == 3
+ loss_b, drift_b = json.loads(lines[1]), json.loads(lines[2])
+ assert loss_b["provider"] == "api.openai.com"
+ assert drift_b["loss_model"] == "gpt-4o"
+ assert drift_b["loss_provider"] == "api.openai.com"
+ assert drift_b["model"] == "gpt-4o"
+ assert drift_b["provider"] == "api.openai.com"
+
+
 def test_context_message_injection_format(tmp_path):
  """Context message carries tz-aware time and lands before the user prompt.