Scoped normalization debug measurements

All 27 retained fixtures execute to 42 with no imports or degradations. Source, normalized IR and generated machine-code hashes match the 319 baseline and all three 320 candidates for every fixture.

The baseline is an immutable prepublication 0.9.319 r5 debug snapshot, and the 0.9.320 r2/r3/r5 builds are unpublished candidates. Exact artifact identities are retained in the data.

Three normalization samples per fixture were collected using immutable debug libraries. The r3 candidate includes the generic-argument pass; r5 includes source-r18 correctness fixes. These snapshots precede the later exact import-alias correction. Concurrent system load was uncontrolled; shallow-case noise and additional work mean these results do not establish uniform speedup. They are not release compilation or full-selfhost measurements.

ShapeDepth319 median ms320 r2 median ms320 candidate r3 median ms320 candidate r5 median ms
nested-let01.4010.4881.7380.845
nested-let45.8101.6131.9413.080
nested-let8187.9084.9023.6183.728
typed-call00.4570.6021.3281.680
typed-call50.9580.9711.8170.925
conditional00.5700.5320.5821.088
conditional527.8442.1223.9321.785
bool-match00.5500.4380.7820.583
bool-match528.5123.1243.2754.260

Nested-let depths 0–8 and typed-call, conditional and boolean-match depths 0–5 are the complete measured scope. See the measurement data for all 324 raw samples, receipts, source-r18 and historical source-r11 manifests, fixture and probe source, hashes and library/binary identities. No claim is made about asymptotic complexity or optimized release performance.

The 319 baseline was built without --release and copied from target/debug; both standalone probes use optimization level 2. The data includes build evidence correcting the nested-let r5 raw receipt’s mistaken release label.