2026-06-17 · benchmark-rotation

Blocked P0 with safe P1/P2 rotation

A gated P0 lane should not stall a whole long-running goal when safe fallback work exists.

The operator sees exactly what decision is needed while the agent can keep making bounded progress elsewhere.

reproducibleuser-gatefallback
LoopX Blocked P0 with safe P concrete user todo reproducible synthetic d goal / trigger todo / gate / evidence public outcome
01

Case context

This case shows what should happen when a P0 route is blocked by a user decision: the system should neither keep forcing that lane nor stop the whole goal. The original shape was a benchmark rotation where one lane needed a large local image while other no-upload benchmark work remained safe.

The public repository does not expose raw benchmark tasks or local image names. It reproduces the control-plane behavior with a synthetic smoke. The user-facing value is seeing one concrete P0 decision while safe fallback work can continue and the gated lane does not burn automated progress budget.

02

Repository evidence

Proof

A user decision should not block all safe work.

LoopX intervention

concrete user todo, safe fallback, quota control

1concrete P0 user decision
1safe fallback lane
0gated-lane auto-progress
0private upload dependency
Synthetic fixture

`examples/showcase-0617-blocked-p0-safe-rotation-smoke.py` reproduces a P0 user gate, a P0 agent lane blocked by that gate, and a P1 no-upload fallback.

Quota contract

The smoke pins `should_run=True`, `requires_user_action=True`, `safe_bypass_allowed=True`, `safe_bypass_kind=scoped_user_gate_fallback`, and related fallback evidence.

Selected fallback

The fixture selects `terminal_bench_no_upload` while preserving the `ale_image` gate as the user-visible blocker.

Rendered evidence

The smoke checks markdown for `scoped_user_gate_fallback` and safe no-upload Terminal-Bench rotation.

03

LoopX behavior

04

What the user sees

The user sees the exact decision they need to make while the project continues safely elsewhere.

Attention load drops: the user does not need to watch repeated idle polls and does not miss the real decision.

05

Repository sources

Evidence boundary. Synthetic public fixture only; no private screenshots, raw tasks, internal links, local image names, or raw run logs.

Demopython3 examples/showcase-0617-blocked-p0-safe-rotation-smoke.py