Phase 4: classify VS Code large-target parse errors

[jununfly]

This was generated by AI during triage.

What to build

Classify the VS Code large-target parse errors from #81 before the Phase 4 decision document is finalized.

The #81 profile succeeded, but the Rust run reported 46 parse errors. Phase 4 should not require zero parse errors on a real large repository, because some files are intentionally malformed fixtures or prompt-heavy test inputs. It does need a taxonomy so rollout decisions do not hide real JS/TS syntax gaps.

Scope

Use the raw artifact from #81:

• docs/benchmarks/2026-06-13-rust-indexing-core-phase-4-vscode-profile.raw.json

Classify each parse error into one of:

• Intentional invalid fixture / malformed test input.
• Generated or prompt-heavy source not meant as normal app code.
• Real supported JS/TS syntax gap.
• Unknown.

Acceptance criteria

• Every parse error in the Phase 4: validate Rust indexing readiness on VS Code or a same-class large JS/TS repo #81 VS Code profile artifact is classified.
• The taxonomy records counts and representative paths for each category.
• Any real supported JS/TS syntax gap has a follow-up issue.
• Any unknown parse error has either a follow-up issue or a documented reason it remains unknown.
• The Phase 4 decision document treats high unknown share as a blocker for default rollout.

Agent notes

Do not turn this into a requirement for 0 parse errors. The goal is decision-quality classification.

[jununfly]

Completed in branch codex/issue-87-reference-resolution-profile, commit 5400ec4.

Evidence:

• docs/benchmarks/2026-06-13-rust-indexing-core-phase-4-vscode-parse-error-taxonomy.md

All 46 VS Code parse errors from the Phase 4 large-target profile were classified. Counts: 15 malformed fixture/test input, 15 prompt/generated or non-normal app source, 16 real supported JS/TS syntax gaps, 0 unknown.

Follow-up created and left open: #88 for the 16 real supported JS/TS syntax-gap paths. Phase 4 does not require zero parse errors, but the real syntax-gap subset remains a default-rollout blocker.