FIELD NOTE / TIKTOK
The terminal says zero. The suite did not pass.
The short film, the complete written thought, and the evidence behind it.
The terminal says zero. The suite did not pass.
Short video caption
An exit code of zero can belong to a wrapper that hid a failing child test. Parse the runner's structured report, require nonzero relevant tests, bind it to the revision, and inject a deliberate fault. Simulated build. #EricFieldNotes
Full written post / accessible read
A coding agent runs tests. The command exits zero, so it writes 'all good.' But the wrapper continued after a failed child process and returned the status of its final logging command. The acceptance tests were red.
In this simulated build, the wrapper printed the failure above a wall of output. The post-tool summary only captured the final exit status. Nobody compared a structured test report with the exact revision that would ship.
A trusted job calls the suite directly, parses its structured report, asserts nonzero relevant cases, and binds the result to the current revision. Then a deliberate broken implementation must make that job fail. That negative control shows the gate can actually detect a fault.
Put the receipt check outside the agent's editable scripts. Block release when the report is missing, stale, empty or insensitive to the injected fault. Do this because process exit alone is too weak a claim for a consequential change.
#EricFieldNotes
Evidence and boundary
On-screen boundary: SIMULATED TEST WRAPPER. The sources below support documented mechanisms and specifications; illustrative scenarios are not presented as measured incidents.