UTC
FORTUNE-OS / ENGINEERING STORIES

Humans & machines.

The work we share, with the evidence behind it.

Some source details are incomplete. View coverage.
LATITUDE / LAST 7 DAYS

Beyond the commit trail.

Explore model telemetry
Engineering project · Captured
Captured traces1,757Engineering trace cohort
Reported tokens11.5BTrace usage, separate from Entire
Estimated model cost$10.55KUSD · telemetry estimate
56.3K tool calls captured
MODEL SPANSUnreported model60.8Kclaude-opus-5-556.3Kclaude-sonnet-5-55,982

Whole engineering project; district filters do not narrow these counts. Business agents are separate in the Observatory. Trace tokens, checkpoint reports and tool cohorts are not added together; estimated cost is not a provider invoice.

Autopilot participation
26Accepted PRs with known Fortune identity
AI-assisted PRs
10Entire or explicit coauthor evidence
Checkpoints available
79/87Unique checkpoints in selected commits
Reported output tokens
1.9MEntire checkpoint reports; not a bill
A SHARED MANIFEST

Who helped this ship?

97%

of accepted, non-dependency PRs have known agent involvement.

Autopilot26
AI-assisted10
No known attribution1
Dependency automation0

Missing attribution is unknown. It does not mean “written entirely by a human”.

CAPTURED EVIDENCE

Inside the machine room

Input tokens7.1M
Cache read tokens648.4M
Cache write tokens5.1M
Line attribution samples70
Reported agent lines10.6K
Estimated model costView Latitude
gpt-6-astraclaude-opus-5-5claude-sonnet-5-5claude-opus-5

PR involvement, token usage and line attribution measure different things. Coverage and metric versions constrain comparisons.

The human + machine league

#CREW MEMBERMERGEDREVIEWSLINES −AI-LINKED PRs
01Anton Vinogradov@awinogradov10331,77429
02fortune-reviewer@fortune-reviewer0710—
02Arthur Kozulia@rovnyart13290—
FROM THE ACTUAL RUN LEDGER

Agent operations

22runs started
22runs with a PR
0human decision gates
31.0mmedian run duration
completed20
canceled2

6243 metadata-only runs, joined to delivered PRs. Separate column-scoped read-only database role; prompts and run specifications excluded. A run, a PR and an accepted delivery remain separate counts.