Files
ARCADIA-Studio/harness/experiments/analysis-mixed-index-20260924/README.md
T

65 lines
3.9 KiB
Markdown

# Private mixed typed-index checkpoint
Source **93eb6117f95353d25d594d59780018f197c51672**, published separately on
`codex/analysis-mixed-index-956atf-20260924`. This implements the consumed
complete-Project owner and shared typed-site verifier from the
[accepted contract](../../analysis-mixed-index-contract-20260924.md). It retains
original bundle/file authority, complete global and scoped parameter names,
qualified occurrence/expression paths and aggregate storage reservations.
There is no conversion into the legacy index/updater and no public source8
activation. The Windows prerelease source is unaffected.
On Windows 11, rustc/Cargo 1.98.1, `x86_64-pc-windows-gnullvm`, strict core
all-target Clippy passes and all seven new tests pass. The actual inventory has
216 names. The unfiltered core library remains **FAILED: 200 passed, 16 failed**;
all 209 prior named outcomes are unchanged, plus seven new passes. Every failure
retains its unsupported-public-source8 diagnostic. Full formatting remains
**FAILED** with the byte-identical eight inherited blocks in five files.
Neither failure is reclassified as success. Linux checks were not run.
The [independent runtime/oracle review](RUNTIME-REVIEW.md) checked all 545 captured
source inputs against immutable Git and disk bytes, exact raw named outcomes,
the baseline failures, formatter output and seven authored test oracles.
Tests cover OA/SA and Unicode/CRLF complete maps, qualified call-label damage,
physical token uniqueness, retained-plus-new storage boundaries, copied text,
original authority and bounded-stack success/failure cleanup. The earlier
[source review](../../analysis-mixed-index-20260924/source-review.md) found three
diagnostic attribution/fallback defects, repaired before compilation. Those
reviewed production hashes remain unchanged in this checkpoint.
Two original test failures are preserved: `a18b7966` strict lint rejected a
manual range pattern; `1d02206c` focused runtime passed six and failed one because
the test incorrectly held the parser's three-frame peak fixed. The contract
permits structural depth verification to raise that peak. The exact authored
root/Left/Argument/Argument path requires four, and `93eb6117` compares against
that peak while preserving every other usage check. Both correction diffs are
archived. No production behavior was weakened to obtain a pass.
`evidence.zip` contains 47 pinned members: nine raw attempt receipts/logs/input
maps, captured launchers, three baseline gate records, source/oracle/runtime
reviews, the two diffs and a separately captured current toolchain report.
[manifest.json](manifest.json) records every member's length and SHA-256 plus
the ZIP identity. Raw launchers are historical execution records: their local
path assumptions must be adapted deliberately for another checkout, and reading
or hashing them does not run archived code. Receipts bind a declared source
input set and selected environment fields; they are not authenticated build
attestations or exhaustive environmental capture.
To run the meaningful subject gates from the identified source, use the project
Windows toolchain setup, one compiler job, `CARGO_INCREMENTAL=0`, and no
`ARCADIA_CODEC_LIFECYCLE_PHASE` or `RUST_MIN_STACK` override:
```text
cargo test --locked --offline -p oa-core --lib -- --list
cargo test --locked --offline -p oa-core --lib language::analysis_source::mixed::project::tests::index_tests:: -- --test-threads=1
cargo test --locked --offline -p oa-core --lib -- --test-threads=1
cargo clippy --locked --offline -p oa-core --all-targets -- -D warnings
cargo fmt --all -- --check
```
The final two failure dispositions above remain expected limitations of this
private checkpoint, not completion criteria for the application. Ordinary
occurrence indexing, original-coordinate edit composition, complete output
admission, public source8, checked persistence and graphical/API parity remain
required work, coordinated in issue 7.