Incremental execution final validation continuation
Measured compiler source checkpoint: dadf2a0df3305380f8975df00ddbba13df4c1e0c.
Build and execution receipts cover public fixtures. The separate private-app
discovery receipt contains numeric observations only, with no application values
or schema names.
-
core-target-builds.json: one identical public Std.App fixture builds for web, web:app, terminal:cli and terminal:tui; actual CLI execution with EOF succeeds. -
desktop-target-builds.json: the same fixture builds for desktop and desktop:mac with the matched locked Nix toolchain. This is compilation only. -
tui-pty-execution.json: the unchanged complete terminal gate passes, including real PTY screen assertions and the terminal CLI effect/guard/exit assertions. -
std-app-execution-tests.json: six individually selected existing tests pass with exactly one passed and zero ignored per invocation, including native HTTP startup/configuration and desktop failure-to-listen handling. -
source-workload-discovery.json: the source-derived private setup executes, but its retained model stays constant despite verified database growth. This does not reproduce the large-model workload or pass its acceptance gate. -
changed-dict-red.json: an executed temporary regression confirms whole-Dict re-encoding after one value changes. Correctness checks pass before the intended performance assertion fails; no implementation fix is claimed.
The canonical compiler and generated artifacts are hash-bound in the build receipts. Cargo integration tests build their own current compiler; they are not claimed byte-identical to the canonical executable. The unchanged external memory guard and original per-test deadlines remain in effect. Initial isolated link attempts failed after dropping Nix SDK configuration; retaining the clean locked Nix context resolved them without a compiler-source change.
The small public fixture contains no indexedMap region. These results therefore do not prove view-cache activation, private application parity, physical-device IME behavior or successful desktop GUI operation. Real private-app browser validation remains a separate required gate.
dict-summary-validation-attempts.json records guard-stopped prototype
compilations. No prototype test executed locally. Hosted run 42136087088 subsequently failed
the same cache-retention invariant in three jobs; 36 of 40 jobs passed.
dict-scalar-ci-correction.json records that failure and the reviewed source
correction removing retained model pointers. The corrected checkpoint still
requires hosted full-runtime and race validation.
inference-owner-sharing.json records the executed owner-identity RED and
118 passing lower/skydb tests after sharing complete immutable inference
results. Memory savings are unmeasured. The suite's existing ignored unifier
growth test was explicitly executed separately and fails; that work remains open.