fix(observability): keep the per-turn usage's model spelling the golden corpus defines (T3-4)

calculator.usage by_model is part of the shared golden corpus, which keeps
the recorded spelling (openai:gpt-5.1); faeb7022 changed it and three golden
screen-number cases failed. The GUI never shows that map; the session usage
table (by_model_parts), the ledger grouping and the turn lists still read one
name per model.
This commit is contained in:
Devika Verma
2026-11-05 16:43:29 +05:30
parent 7703132a17
commit c1dd0ed2ba
2 changed files with 3 additions and 4 deletions
+1 -3
View File
@@ -10,8 +10,6 @@ from __future__ import annotations
from typing import Any, Iterable, Optional
from .model_names import canonical_model
BUCKET_ORDER = ["waiting", "tools", "model", "auto_approver", "summary", "sandbox"]
@@ -62,7 +60,7 @@ def usage(recording: dict[str, Any], session_id: Optional[str] = None,
total += t
p = attrs.get("purpose") or "unattributed" # the format's word for a call with no purpose (V3-19)
by_purpose[p] = by_purpose.get(p, 0) + t
key = canonical_model(attrs.get("model") or "?") + ("" if p == "main" else f" ({p})") # T3-4
key = attrs.get("model", "?") + ("" if p == "main" else f" ({p})")
by_model[key] = by_model.get(key, 0) + t
u = attrs["usage"]
if u.get("cache_read") is not None: # a call with no cache count stays out of the share
@@ -243,6 +243,7 @@ def test_one_model_is_read_and_grouped_under_one_name(tmp_path):
heavy = usage_api.heaviest(store, start=span[0], end=span[1])["turns"]
assert sorted(m for t in heavy for m in t["model_ids"]) == ["anthropic:model-y", "model-x", "model-x"]
rec = store.load_recording("s1")
assert calculator.usage(rec)["by_model"] == {"model-x": 220, "anthropic:model-y": 110}
# the per-turn usage keeps the recorded spelling: the golden corpus defines it
assert calculator.usage(rec)["by_model"] == {"openai:model-x": 110, "model-x": 110, "anthropic:model-y": 110}
assert set(read_api.usage(rec, "s1")["by_model_parts"]) == {"model-x", "anthropic:model-y"}
store.close()