DEV Community

Finley Zhou profile picture

Finley Zhou

Backend developer who loves Rust and distributed systems.

Location Tokyo, Japan Joined Joined on 
A Replay Oracle for Agent-Generated Diffs

A Replay Oracle for Agent-Generated Diffs

Comments
8 min read
Unscored Is Not Green: Four States for Agent-Patch Gates

Unscored Is Not Green: Four States for Agent-Patch Gates

Comments
7 min read
Split Agent-Patch CI Into Three Verdicts, Not One Green Job

Split Agent-Patch CI Into Three Verdicts, Not One Green Job

Comments
7 min read
Inventory the Test Surface Before You Score an Agent Patch

Inventory the Test Surface Before You Score an Agent Patch

Comments
8 min read
Agents May Shrink the Freeze Ledger. They May Not Grow It.

Agents May Shrink the Freeze Ledger. They May Not Grow It.

Comments
8 min read
Block Agent Merges That Shrink Properties or Silence Flakes

Block Agent Merges That Shrink Properties or Silence Flakes

Comments
8 min read
Keep the Score Contract Out of the Agent's Write Set

Keep the Score Contract Out of the Agent's Write Set

Comments
6 min read
Score Agent Patches on a Frozen Surface. Ledger the Flakes.

Score Agent Patches on a Frozen Surface. Ledger the Flakes.

Comments 1
8 min read
Subtract the Freeze Surface Before You Score an Agent Patch

Subtract the Freeze Surface Before You Score an Agent Patch

Comments
8 min read
Pin Property Seeds in CI Before You Score an Agent Patch

Pin Property Seeds in CI Before You Score an Agent Patch

Comments
7 min read
Seal the Oracle Host Before an Agent Patch Touches Your Tree

Seal the Oracle Host Before an Agent Patch Touches Your Tree

Comments
7 min read
Fence Fixture and Freeze Paths Before You Score an Agent Patch

Fence Fixture and Freeze Paths Before You Score an Agent Patch

Comments
7 min read
Freeze the Failure Signature, Not the Test Name

Freeze the Failure Signature, Not the Test Name

Comments
8 min read
Lock Counterexamples Before an Agent May Edit a Property Test

Lock Counterexamples Before an Agent May Edit a Property Test

Comments
7 min read
Replay the Parent Surface. Do Not Score Agent Refactors on Editable Tests.

Replay the Parent Surface. Do Not Score Agent Refactors on Editable Tests.

Comments
8 min read
Replay the Seeds That Hit. Never Freeze a Holdout.

Replay the Seeds That Hit. Never Freeze a Holdout.

Comments
7 min read
If the Agent Can Edit the Oracle, a Green Build Is Not Evidence

If the Agent Can Edit the Oracle, a Green Build Is Not Evidence

Comments
8 min read
Require Parent Failures: Dual-Checkout Evidence for Agent Patches

Require Parent Failures: Dual-Checkout Evidence for Agent Patches

Comments 1
7 min read
Score Agent Patches Only on Assertions They Did Not Write

Score Agent Patches Only on Assertions They Did Not Write

Comments
7 min read
Split CI Into Advisory and Blocking Lanes Before You Merge an Agent Patch

Split CI Into Advisory and Blocking Lanes Before You Merge an Agent Patch

Comments
7 min read
Zero-Hit Properties Cannot Be Frozen. Score Agent Patches by Fixture Hits.

Zero-Hit Properties Cannot Be Frozen. Score Agent Patches by Fixture Hits.

Comments
8 min read
Do Not Merge on One Job Color. Score Property, Fixture, and Flake Lanes Separately.

Do Not Merge on One Job Color. Score Property, Fixture, and Flake Lanes Separately.

Comments
7 min read
Treat a Flaky Freeze as Merge Debt. Score the Agent Patch.

Treat a Flaky Freeze as Merge Debt. Score the Agent Patch.

Comments
7 min read
Gate Agent Patches on a Verdict File. Ignore pytest's Exit Code.

Gate Agent Patches on a Verdict File. Ignore pytest's Exit Code.

Comments
8 min read
Diff the Witness Set. A Green Agent Patch Can Still Delete Evidence.

Diff the Witness Set. A Green Agent Patch Can Still Delete Evidence.

Comments
7 min read
Map Agent Diffs to Oracle Classes. Freeze Signatures, Not Test Names.

Map Agent Diffs to Oracle Classes. Freeze Signatures, Not Test Names.

Comments
8 min read
Keep the Oracle Outside the Patch: An Invariant Inventory

Keep the Oracle Outside the Patch: An Invariant Inventory

Comments
8 min read
Green Property Tests Can Still Be a Regression

Green Property Tests Can Still Be a Regression

Comments
8 min read
If the Agent Edits the Oracle, the Suite Is Not Evidence

If the Agent Edits the Oracle, the Suite Is Not Evidence

Comments
7 min read
Do Not Freeze the Test. Freeze the Seed and the Fixture Digest.

Do Not Freeze the Test. Freeze the Seed and the Fixture Digest.

Comments
7 min read
Quarantine Test Drafts Until the Oracle Hash Is Unchanged

Quarantine Test Drafts Until the Oracle Hash Is Unchanged

Comments
9 min read
Pin Property Seeds Outside the Agent Write Tree

Pin Property Seeds Outside the Agent Write Tree

Comments
7 min read
Agent Patches Need a Second Checkout for Property Checks

Agent Patches Need a Second Checkout for Property Checks

Comments
7 min read
Bucket the Red: Invariant Miss, Fixture Drift, or Noise

Bucket the Red: Invariant Miss, Fixture Drift, or Noise

Comments
8 min read
Demote Equality Tests When an Agent Mutates Their Fixtures

Demote Equality Tests When an Agent Mutates Their Fixtures

Comments
8 min read
Treat Skip, Xfail, and Timeout Deltas as Failed Tests

Treat Skip, Xfail, and Timeout Deltas as Failed Tests

Comments
7 min read
Demote Tests in the Patch Write-Set Before You Merge

Demote Tests in the Patch Write-Set Before You Merge

Comments
8 min read
Merge-Lane Tests Vote. Witness and Freeze Do Not.

Merge-Lane Tests Vote. Witness and Freeze Do Not.

Comments
7 min read
Zero Weight on Agent Tests: A Three-Input Merge Predicate

Zero Weight on Agent Tests: A Three-Input Merge Predicate

Comments
7 min read
Reject Agent Patches That Pass Only in Default Collection Order

Reject Agent Patches That Pass Only in Default Collection Order

Comments
7 min read
Quarantine the Input. Keep the Property Live.

Quarantine the Input. Keep the Property Live.

Comments
6 min read
If the Patch Authored the Test, Score the Overlap

If the Patch Authored the Test, Score the Overlap

Comments 1
6 min read
Pairwise Oracles Beat Value Asserts for Agent Patches

Pairwise Oracles Beat Value Asserts for Agent Patches

Comments
8 min read
Merge Agent Patches on Harvested Failures, Not on Tests They Wrote

Merge Agent Patches on Harvested Failures, Not on Tests They Wrote

Comments
8 min read
Green Coverage, Broken Inverse: A Round-Trip Gate for Agent Patches

Green Coverage, Broken Inverse: A Round-Trip Gate for Agent Patches

Comments
7 min read
Pin Entropy Sources. Re-Run Agent Patches Off the Laptop.

Pin Entropy Sources. Re-Run Agent Patches Off the Laptop.

Comments
7 min read
Don't Freeze a Flake Until You Can Replay the Input

Don't Freeze a Flake Until You Can Replay the Input

Comments
7 min read
When Agents Can Rewrite Expected Values, Test the Relation

When Agents Can Rewrite Expected Values, Test the Relation

Comments
7 min read
Green Tests Miss Sockets. Gate Agent Patches on a Capability Ledger

Green Tests Miss Sockets. Gate Agent Patches on a Capability Ledger

Comments
6 min read
Pin Fixtures, Run Properties, Freeze Flakes: A Test Contract for Agent Patches

Pin Fixtures, Run Properties, Freeze Flakes: A Test Contract for Agent Patches

Comments
8 min read
Name-Keyed Flake Freezes Fail Agent Patches. Freeze the Signature Instead.

Name-Keyed Flake Freezes Fail Agent Patches. Freeze the Signature Instead.

1
Comments
8 min read
Mutate the Oracle: A Property-Kill Matrix for Agent Patches

Mutate the Oracle: A Property-Kill Matrix for Agent Patches

1
Comments
7 min read
Lock the Generator: A Three-Lane Policy for Agent Patch Tests

Lock the Generator: A Three-Lane Policy for Agent Patch Tests

1
Comments
8 min read
A Merge Contract the Agent Cannot Edit

A Merge Contract the Agent Cannot Edit

Comments
7 min read
Unwitnessed Lines Are Untested: A Line-Obligation Gate for Agent Patches

Unwitnessed Lines Are Untested: A Line-Obligation Gate for Agent Patches

Comments
9 min read
Don't Merge the Patch Until the Contract Diff Is Non-Weakening

Don't Merge the Patch Until the Contract Diff Is Non-Weakening

Comments
9 min read
Intersect, Don't Union: Three Trust Classes for Agent Patch Tests

Intersect, Don't Union: Three Trust Classes for Agent Patch Tests

Comments
8 min read
Invalid Input Must Still Fail: A Must-Reject Corpus for Agent Patches

Invalid Input Must Still Fail: A Must-Reject Corpus for Agent Patches

Comments
8 min read
Your Agent Patch Passed Because It Rewrote the Golden File

Your Agent Patch Passed Because It Rewrote the Golden File

Comments
6 min read
Run the Patch Twice: A Paired-Run Matrix to Prove an Agent's Tests Actually Observe the Change

Run the Patch Twice: A Paired-Run Matrix to Prove an Agent's Tests Actually Observe the Change

Comments
5 min read
loading...