Code agent: gate a patch on regression evidence

Execute a concrete boundary change

The contract is min(n, 2) for a nonnegative integer input. Safe patch implements n greater than or equal to two returning two, otherwise n. Patch breaks one test changes that boundary return value to one. The browser executes these two predefined JavaScript functions; it never evaluates user-provided source or runs a generated program. The visible candidateSource is a description of the selected function. Inputs zero, one and two have expected results zero, one and two. The final records print expected and actual results so the test verdict is auditable rather than a predetermined pass label.

Compare the same patch under both gates

With Run regression tests and Safe patch, all three cases pass and the candidate is applied in the model. Change to Patch breaks one test: input two returns one instead of two, so one of three tests fails. The gate rejects the patch and the applied outputs stay zero, one and two because the correct original remains. Change only Policy to Skip regression tests. testResults is empty and testsRun is false; the gate applies the broken candidate without test evidence. Applied outputs become zero, one and one. regressionCaught is false in this branch because the configured gate ran no test, even though the independent teaching audit reveals the error.

Keep an audit separate from gate evidence

The teachingAudit array always executes the candidate on the three shown inputs so you can inspect the consequence of a skipped gate. It is explicitly not evidence supplied to that gate. failedTests zero under Skip regression tests means zero observed test failures, not three passing tests. Under Run regression tests, inspect the length and passed fields of testResults before interpreting that count. Revisit Safe patch with tests skipped: its output happens to be correct for these cases, but its release still lacks test evidence. A good result by coincidence does not validate the release process.

Understand coverage and scope

Three cases establish behavior only on those inputs. They do not prove all nonnegative integers satisfy the contract, let alone invalid inputs, floating-point numbers or parser behavior. The function is a bounded clamp, with an explicit boundary at two. This page is not a source editor, CI service, sandbox or real code agent. It omits generated diffs, dependency installation, security review, flaky tests and deployment. A production agent needs isolated execution, a reproducible test environment, tests tied to the changed behavior and explicit release authority. The supplied readings discuss broader code-agent engineering and regression evaluation.

Inspect every stage

Play simulation reveals eight explanatory stages over sixty seconds. The timeline is reading time, not measured task runtime. Pause simulation freezes the inspection position; Resume simulation continues it. Restart returns to zero while retaining the selected policy, fixture and speed. Timeline seeks directly from zero through sixty seconds and pauses playback. Speed selects half, normal or double pace; a change settles elapsed time at the old speed first. At completion, Replay simulation starts a new inspection. Selecting either scenario input resets without autoplay. Moving to a hidden tab pauses. The final structured result appears at sixty seconds; earlier cards explain how it was computed. Waiting cards are not completed task work. Every card includes text, so meaning does not depend on color.

Save a reproducible comparison

Capture baseline stores the currently visible result and control values. Seek to sixty first for a final-outcome comparison, then change one input and inspect the difference. Replace baseline updates that snapshot; Clear baseline removes it. The notebook stores observations in this browser, and Export notes downloads them. Expand experiment and Exit full screen change the view without changing the computation. Record the policy, fixture, repeated steps or failing case, and result rather than interpreting the inspection clock as a performance benchmark. These pages contact no agent service, create no external task and modify no source file.