PREREG-001Frozen August 27, 2026Planned condition · not a result

Qwen3.8 Flash returns to Triage Desk under revision 3

Under the same model, test, route, temperature, and High reasoning setting, can Qwen3.8 Flash complete Triage Desk when its 24,000-token visible-artifact budget is protected from a separate 16,000-token reasoning allowance?

Outcome publication committedNo provider preflight, spend approval, or model dispatch has occurred.
Protocolv0.8 · revision 3
PlanLP-EDCDDCC14715
RouteAlibaba · pinned
Issue candidate014
Launch-plan hashsha256:edcddcc147155cb6c16fab92b418df1eea79e1700c7bafca8dbc0a9564aed17c
Preregistration source hashsha256:35c3553dc29ecf2afab53d9a0f57b7b1d0e37df84c1b1082e7689d8c2ed37dcd

A promise before the answer—not a seal of correctness.

This is a timestamped promise made before seeing the answer: Proofrun fixed the condition in advance and will publish every valid model-produced outcome, including an incomplete, poor, or empty submission.

This does not certify that the test is perfect, predict that the model will perform well, or make one result representative of every use of the model.

Proofrun reserves public preregistration for consequential runs where advance commitment materially answers a fair cherry-picking concern, rather than publishing one before every routine review.

Change one consequential variable.

This condition keeps the model, test, provider route, temperature, and reasoning setting fixed against the closest earlier condition. The prospective variable is Proofrun's completion contract: reasoning and visible artifact now receive separate declared room.

The earlier conditions remain evidence under revision 1. This is a new revision-3 condition, not a repaired or replacement attempt.

The condition, before the answer.

Modelqwen/qwen3.8-flash
TestTriage Desk v1
Reasoninghigh · 16,000
Artifact budget24,000
Total ceiling40,000
Cost stop$0.03
Temperature0.7
Attempts1 initial

Fallbacks are disabled. Exact endpoint compatibility and pricing must still pass a current preflight before launch.

The result cannot choose its own visibility.

Proofrun will publish every valid model-produced terminal outcome from this condition regardless of quality or score.
  • Complete submission
  • Reviewable truncated submission
  • Invalid or non-reviewable model artifact
  • No submission after a successful model dispatch

Human evaluation and release review still apply. They cannot suppress the record because its score or artifact is embarrassing.

No result-driven retries.

Provider incident

A classified provider incident pauses the condition and may receive at most one linked retry under the exact same frozen condition. The incident remains disclosed.

Harness or protocol defect

A harness or protocol defect is disclosed and does not count as a model outcome. Any corrected attempt must remain linked to the defect record.

Nothing historical gets rewritten.

Frozen is not launched.

  1. Complete exact Alibaba endpoint and pricing preflight.
  2. Obtain separate creator approval for the paid launch.
  3. Create the immutable run and attempt records before dispatch.

Read PN-001 for the protocol change →

Inspect the commitment.

Public record content hashsha256:cfcc676f2d8e4d9d50ac643c66b94d446e81c45960bc37d2b4822a98b7b18c57