A Fast Default Gate and an Opt-In Display Tier

This sprint was asked to do the classification work that made the rest of the job possible. A fresh agent could hold it because the boundary was per-test behavior: display recording, terminal automation, media validation, and real-time replay left the default tier, while cheap neighboring checks stayed put. The run facts show no repair on this slice, so the risk was carried by the audit rather than by later recovery.

The tier needed a command boundary so selecting expensive tests did not depend on memory of cargo flags. This sprint was scoped to the runner and its checks: set the opt-in environment, select only the intended ignored tests, and distinguish missing local display tools from a real failure. It stayed green, which left the later historical plan-impact note as a structural warning rather than a repair.

This was the milestone's thin closing piece: update the places that tell contributors what verification means. Its boundary was documentation alignment with the tier and runner already built, so the agent did not need to reason about test selection internals. The sprint finished green, and the milestone review then supplied the discovery constraint that affected later work.

This sprint isolated the fake replay fixture before broader polling changes could mask it. The brief drew a clean line between tests that merely need a replay process to move through states and tests where elapsed playback time is the assertion. The green result meant later wait cleanup started from a faster fixture rather than compensating around a slow fake process.

This was the stress point in the timing milestone. The sprint had to change shell scripts and integration tests from fixed polling to condition-based readiness while preserving bounded failure behavior, and the gate sent it back once after review. That repair is evidence that the hard part was not finding sleeps, but proving the shortened waits still failed correctly when readiness never arrived.

The final timing sprint applied the same discipline to in-module and unit-test waits. Its scope was broad, so the brief constrained the work to pure polling around channels, files, processes, and state flags, while leaving tests that prove timing semantics alone. The run stayed green here, which fits a conservative sweep over local wait helpers rather than a semantic rewrite.

Before separate integration files could become one threaded test binary, tests that mutate process state needed an explicit guard. This sprint was scoped to helpers and fixtures rather than movement, which kept concurrency risk out of the later relocation work. It stayed green and gave the consolidation agents a support layer to reuse instead of rediscovering isolation issues in moved test bodies.

This sprint carried the main file-layout reduction for the default tier. Its charge was structural but risky: preserve names, assertions, and cargo name filtering while moving default tests into one tiered integration binary. The ledger note says mixed display files had to be split by hand during recovery, so the difficult boundary was separating default tests from display tests without leaving any top-level leftovers.

This sprint addressed the review finding from the first milestone directly. Once display tests moved into the consolidated structure, the runner and its gate coverage had to change with them or the command would keep selecting the old shape. The sprint stayed green, which indicates the earlier historical plan-impact note was carried into the right boundary before the file move landed.

The final sprint was intentionally layered over the completed layout: readable output could only reflect the finished default and display tiers. It came under run stress, with failed gates and investigation recorded before the ledger marked the code green by manual confirmation. That boundary matters because the brief required optional tooling only; the ordinary cargo path could not become dependent on nextest.