Skip to content

synthetic-auxiliary-carriers._expectations.md (SYNTHETIC)

All samples in this directory are synthetic, rather than extracted native sessions. This covers four auxiliary usage locations and an assistant without usage, following Pi session JSONL v3 at fixed pi-mono revision b45597504eeaba1f11a9920a1d1048c361ed4b8e.

Scenario and expectations

Eight lines: session (version=3, id=syn-sess-aux), model_change (syn-prov/syn-model-a), two assistants (syn-a-1 with complete usage; syn-a-2 without), standalone usage (kind=cache_warm), compaction with usage, branch_summary with usage, and toolResult with usage.

Manual calculation: input/cacheRead/cacheWrite are exclusive; totalTokens sums all four fields.

Entry Category input_total (derived) cacheRead cacheWrite output total (derived) reasoning
syn-a-1 primary 100+800+100=1000 800 100 50 1050 10
syn-a-2 primary no usage; tokens unknown, no filled zeros — — — — —
syn-u-1 auxiliary 10+90+0=100 90 0 0 100 absent
syn-c-1 auxiliary 1000 0 0 200 1200 absent
syn-b-1 auxiliary 500+300+0=800 300 0 100 900 absent
syn-t-1 auxiliary 5+0+10=15 0 10 5 20 absent
  • Six events: two primary, four auxiliary; Pi has no sub_agent category.
  • Summary (2026-01-05): call_count=6, input_total_known=2915, cache_read_known=1190, cache_write_known=110, output_total_known=355, total_tokens_known=3270: 1615 uncached +1190 cache read +110 cache write +355 output.
  • Model: syn-a-1/syn-u-1 request_field; syn-c-1/syn-b-1 use the latest preceding model_change for syn-model-a (structured_change); syn-t-1 unknown.
  • One usage_shape_deviation for syn-a-2. Other entries have all five usage fields and consistent totalTokens; no other diagnostics.