previous-draft/collectors/codex.py

Comment 1, line 1

# -*- coding: utf-8 -*-

Comment 2, line 2

previous-draft/collectors/codex.py:2

"""Codex collector: token_usage_record in ~/.codex/sessions/**/rollout-*.jsonl.

input_tokens includes cache hits. Split input = input_tokens - cached_input_tokens,
cache_read = cached_input_tokens and cache_write = cache_write_input_tokens.

Model tracking: usage records omit the model; event_msg(thread_settings_applied)
and other lines provide it. An incremental block may contain only usage lines:
1. Persist each file's latest known model in source_state (model:<path>) across blocks.
2. For unresolved events, scan the complete file by timestamp and use the latest
   model line that is no later than the event.
"""

Comment 3, line 36

previous-draft/collectors/codex.py:36

"""Scan model lines across the file in file order as (ts, model), skipping usage lines."""

Comment 4, line 58

previous-draft/collectors/codex.py:58

"""Use the latest model line with ts_model <= ts; use the first line for earlier events."""

Comment 5, line 81

previous-draft/collectors/codex.py:81

# Indices of events whose model is unresolved.