BASEBREAK

Causal Verification Runtime for AI-Written Software Changes

PUBLIC RECEIPT: VERIFIED PROVENANCE: LIVE_NEBIUS RUN: run_3ba7d1fd-6dca-40c8-a7bb-39447bc85f1f CLOUDFLARE WORKERS EDGE

Thesis: If the patch matters, the base must break.

Basebreak proves that an AI-written patch caused the behavior it claims to change — not merely that its tests are green.

Causal Invariant: BASE = FAIL → CANDIDATE = PASS → COUNTERFACTUAL = FAIL

Causal 3-World Evidence Status

[1. Baseline World] FAIL (Exit 1)

Clean base materialization before candidate patch. Verified failure on single-use witness.

Sandbox: sbx-c9e5c0e476fa4466 | Tree: 824fa2c69a222cc2... (2.452s)

[2. Candidate World] PASS (Exit 0)

Nemotron candidate patch applied. Verified pass on single-use witness.

Sandbox: sbx-cd870e01d6d94bca | Tree: 6c4f46cbb51119d8... (2.125s)

[3. Counterfactual] FAIL (Exit 1)

Candidate delta subtracted/reverted under identical witness. Proves patch delta was causally necessary.

Sandbox: sbx-90d40f25758244d0 | Tree: 824fa2c69a222cc2... (4.617s)

Execute Bounded Causal Verification

Public live execution is disabled on this build. The Worker currently returns truthful NOT_RUN responses until trusted budget authority and an independently validated remote Python verifier are integrated.

Verified historical execution for run: run_3ba7d1fd-6dca-40c8-a7bb-39447bc85f1f (Task: TASK-RESET-VAULT-001)
Causal Invariant: BASE=FAIL (Exit 1) → CANDIDATE=PASS (Exit 0) → COUNTERFACTUAL=FAIL (Exit 1). Causal triplet verified.
Zero-Cost & Credit Drain Protection: This build fail-closes before any provider call. Client-supplied balance values do not authorize live spending, and no authentic historical receipt is exposed by default.

Cryptographic Public Receipt

Receipt SHA-256 Digest:
e2e833fd3aa9d69bdcbaad0805fd38167c680cb2f60a16209c27a80123a43708
Source Commit SHA:
80b43b84014fa61b941c7ebcd87836a56dcd7c92
Public Evidence Status:
Authentic record loaded from ledger (SHA-256 digest verified). Provenance: LIVE_NEBIUS.
Overall Verdict: VERIFIED Telemetry: Model: nvidia/Nemotron-3_5-Lightning | Tokens: 1,123 in / 693 out | Sandboxes: 3 | Duration: 12.91s
Authentic stored receipt loaded from ledger. SHA-256 digest integrity verified against canonical serialization.