Type 3Observation

Backpressure patterns for agent fan-out

proposed
voting · 3d
running
measuring
concluded

Preregistration · v1 · 31 Aug 2026 · immutable after start

An extension creates a new version rather than editing the old.

Hypothesis

Bounded-concurrency queues with per-parent budgets in front of agent fan-out reduce p95 task latency by at least 30% and cut rate-limit errors by 80% against the unbounded fan-out three patterns run today, at no loss in task completion.

Kill condition

Kill if p95 latency improves under 10% by day 10 on the replay set, or if task completion falls more than 1 point under any queue configuration.

MethodReplay 6k fan-out tasks from three patterns through four configurations: unbounded, bounded concurrency, bounded + per-parent budget, bounded + budget + priority. Measure p95 latency, 429s, completion, cost.

Expected cost$900 in replay tokens, two people for three weeks

Expected duration3 weeks

Running notes · fed from harness sessions and by the pair

manual31 Aug 2026TO Tom Okafor
Prereg filed. Predicted -40% p95, -85% 429s, confidence 0.65.
manual24 Aug 2026TO Tom Okafor
Observed in three patterns this quarter: fan-out of 40+ sub-agents hitting rate limits, retries doubling cost. r-backpressure exists at assessed tier; this is the run to test it.

Harness notes are auto-captured from Claude Code sessions: model, date, commit, session reference. Never the transcript, code or paths.