fix(telemetry): a counter says what was measured

[blafourcade]

What

Two tools declared a limit that was true when it was written and is not any more. Both are now measured, and the captures that settle them ship as fixtures.

Copilot — input against cache_read

copilot.ts said the exclusivity was unconfirmed, because the only capture held cache_read: 0, where an inclusive and an exclusive input produce the same number. The comment named what would close it: "one capture of a session with a non-zero cache_read".

Captured on @github/copilot@1.0.82:

field	value
tokenDetails.input	9
tokenDetails.cache_read	42038
tokenDetails.cache_write	21404
tokenDetails.output	408
modelMetrics.<model>.usage.inputTokens	63451

9 + 42038 + 21404 = 63451, and Copilot's own terminal line for that run read ↑ 63.5k (42.0k cached, 21.4k written).

An input that already counted the cached prompt would read 63451, not 9. The four counters are disjoint, so the report is right to add them — it would otherwise have over-counted that one session by 42038 of 63451.

OpenCode — a non-Anthropic provider whose cache is exercised

opencode-export.ts said no capture ever put a large cache.read beside input for a non-Anthropic provider, "which is the one comparison that would show input failing to shrink if it already counted the cached tokens". The earlier probe of a second provider never exercised its cache.

Captured from opencode export --sanitize, opencode 1.14.20, providerID opencode, modelID ling-3.0-flash-fin-free, three billed turns of one session:

turn	input	cache.read	total
1	28242	640	29089
2	269	28928	29356
3	196	29184	29438

input falls as cache.read climbs, and total == input + output + reasoning + cache.read + cache.write holds on all three turns — which it could not if the counters overlapped.

What each tool still cannot say

Narrower than before, and both limitation strings now say that instead of the old sentence. For OpenCode: a provider reporting prompt tokens inclusive of the cached ones has still never been captured here.

Those strings are user-visible — they are the reason a by_tool row carries — so cli/tests/fixtures/cli-owns-read/expected-envelope.json moves with them.

The guards bite

Mutating copilot-events.ts to fold cache_read into input turns exactly the new case red, and only it:

→ expected 42047 to be 9 // Object.is equality
Tests  1 failed | 10 passed (11)

Fixtures

Both are real captures, sanitized: session ids replaced, project id zeroed, directory and title already redacted by --sanitize, the modified-file path replaced. Neither carries a machine path or a personal name.

• cli/tests/fixtures/local-cost/.copilot/session-state/55555555-…/events.jsonl
• cli/tests/fixtures/telemetry-sink/opencode-export-non-anthropic-cache.json

Gates

• cli vitest — 3504/3504
• node --test "scripts/__tests__/**/*.test.js" — 373/373
• biome check src tests — the only two warnings are pre-existing, in a file this branch does not touch (telemetry-evidence-adapter.integration.test.ts)

🤖 Generated with Claude Code

https://claude.ai/code/session_01VWNxk63AGKkqE8HRqHLjGp