Skip to content
AI usage dashboard

synthetic-step-finish expectations (all synthetic)

Scenario: two sessions (main and parent_id-linked child), two assistant messages, three step-finish parts and one text part, which has no usage and is excluded. message.data.tokens aggregates a turn’s message parts: do not read or add it. Under projector applyUsage, five session.tokens_* columns sum current parts for comparison.

Manually calculated expectations:

  • Three usage_events (model_call; daily call_count=3 on 2026-06-13):
    • opencode:part:part_syn_1: primary, model_raw=claude-sonnet-4-6, provider=anthropic, time_basis=observed_at, source_revision=1781337600500. input_uncached=1000, input_total=9500 (derived: 1000+8000+500), output_total=250 (200+50), output_reasoning=50, cache_read=8000, cache_write=500, total_tokens=9750, source_total=9750 (matches, no diagnostic), cost=12000 micro-USD (estimated).
    • opencode:part:part_syn_2: primary, same model; input_total=9500 (500+9000+0), output_total=80 (80+0), total_tokens=9580 (derived), source_total=NULL (no total field).
    • opencode:part:part_syn_3: sub_agent with nonempty session.parent_id, model_raw=gpt-5.2, provider=openai. input_total=1300 (200+1000+100), output_total=50 (40+10), total_tokens=1350, source_total=1350 (matches).
  • Text part part_syn_text creates no event and is not counted in records.
  • Daily summary (2026-06-13 UTC): call_count=3, input_total=20300, cache_read=18000, cache_write=600, output=380, total_tokens=20680.
  • Two matched reconciliations:
    • ses_syn_main: part sum 9750+9580=19330 matches five columns 1500+280+50+17000+500=19330.
    • ses_syn_sub: 1350=200+40+10+1000+100.
  • Exactly one latest_fallback diagnostic: registry empty; implementation based on documentation/source.
  • Repeat scanning leaves call_count=3: same key/content is unchanged.