Autonomous-PR outcomes across the fleet. Cost is API-equivalent (notional on a Pro subscription). Quarantines are a safety signal — afk refused to ship — not failures.
Fleet state · not yet wired up
Live gate/promote/merge and dependency-frontier state isn't ingested
yet — this panel is a placeholder for a future feature, not a
fabricated number.
98%
landed rate
n = 731 gate-passed
98%
landed rate (executor repos)
n = 8 exec repos
$3.88
API-equiv $ / merged PR
n = 719 merged
fully loaded: includes failed runs, quarantines, and scout overhead
33%
quarantine (safety) rate
n = 811 issues
18%
escalation rate
n = 988 runs
0%
rejected rate
n = 1,395 proposals
Pipeline
Stage funnel
By repository
Outcomes per repo
landed
quarantined
other
Orbit-Wars
77 · 87%
afk-agent-system
430 · 89%
afk-cockpit
106 · 94%
graphmark
14 · 93%
household-bms
48 · 96%
household-command-center
1 · 100%
housing-commute-analysis
15 · 93%
the-workshop
120 · 79%
Producers
Proposal quality
Safety axis
Quarantine by category
Retries & recovery
Runs per issue
12345
Retries & recovery
Quarantine → retry rescue
retry rescues 60% of first-run quarantines · 105% of those actually retried
74%
first-try pass rate
16%
final-state quarantine (issues)
27%
quarantine rate (runs)
Cost · API-equivalent
fleet telemetry · per attempt · Jun 7 – Aug 2, 2026
Spend by model tier
$2,472.78 total across the window
⚠ 27% of attempt spend ($670.16 · 397 attempts) has no recorded model_tier — telemetry gap, not a model
Overhead · orchestrator + scout ops
$317.55
285 cost events · additive to the tier spend above · not a model tier
Claude Code transcripts, afk worktrees only · per API call · Jul 3 – Aug 2, 2026
$0.03
median / call
$0.08
p90 / call
26,363
calls logged
whole machine, all projects: $9,040.72 across 64,797 calls
Cost dial
Per-API-call cost distribution
<$.05.05–.10.10–.15.15–.20.20–.30.30–.50$.50+
Model mix
Opus share by day
Jul 3
range: 0%–62%
Aug 2
Throughput
Slices per day
06-0706-1306-1906-2707-0307-1407-2307-30
Context cost
Tokens by type, per day
1.94B tokens · 95.8% cache reads — the context is overwhelmingly cached
cache read
cache write
input
output