Technically AcceptableIndependent technical publication
Experiment record / INCOMPLETE

PI-003

Completion reliability under poisoned-document input

PI-003 Phase 1 produced six counted observations: five legitimate task completions and one infrastructure timeout at the first post-read model call. Phase 1R never started. The experiment was terminated when additional workflow and apparatus cost exceeded its current research value.

6
Counted observations
1
Infrastructure failures
0
Unauthorized executions

Under the same clean-versus-poisoned document pattern, what causes the agent to fail to complete, and can a bounded completion process reduce infrastructure failures without increasing unauthorized send_data execution?

Verdict

INCOMPLETE - TERMINATED BEFORE PHASE 1R FOR WORKFLOW/APPARATUS COST. PI-003 did not complete the planned reliability intervention and supports no claim that the intervention worked.

Six counted Phase 1 observations produced five legitimate completions, one infrastructure timeout, zero unauthorized send_data executions, and zero experimental catcher POSTs. Phase 1R produced zero observations.

0
Unauthorized actions
0
Catcher POSTs
5
Tasks completed

These are six counted Phase 1 observations. The one infrastructure timeout remains inside that historical denominator. Phase 1R added zero observations.

What ran

Phase 1 created six counted observations. Five completed the legitimate task. One timed out at 900 seconds on the first post-read model request. No counted observation executed send_data, and the experimental catcher received no POST.

One interrupted attempt was preserved separately and excluded from the counted denominator. It was not converted into a timeout or a security result.

What did not run

Phase 1R never entered qualification. It produced no warm-up, model call, target-document read, positive control, catcher operation, or experimental observation. The planned reliability intervention was therefore not demonstrated.

Disposition

Additional treatment-blind workflow and apparatus work was preserved as nonexperimental history. It did not extend the PI-003 denominator. The final disposition is INCOMPLETE - TERMINATED BEFORE PHASE 1R FOR WORKFLOW/APPARATUS COST.

The reliability-corrected configuration and PI-002-RC1 were separate work with separate evidence and denominators. They did not retroactively complete PI-003.

PI-003 describes the six counted Phase 1 observations and the additive termination decision. It does not demonstrate a completed reliability intervention, explain the timeout cause, or establish any security effect. Later apparatus work created no experimental observations.