[M8] Validate standalone v1 and rebaseline M9

[Teakowa]

Parent: #9

Depends on: #49, #50, #51, #52, #53, #54

Goal

Act as the v1 acceptance and re-evaluation gate before agents proceed into M9 toolchain/API work.

Scope

• Verify S/D/N/E release gates against actual reports, scenarios, tests, and supported corpus.
• Verify compat optimization safety and benchmark/resource evidence.
• Verify standalone artifacts, CLI behavior, native frontend paths, and default runtime dependencies.
• Audit crash/error handling, deterministic output, public status/docs, and release automation.
• Classify remaining gaps as v1 blockers, known supported-surface regressions, or explicitly deferred post-v1 work.
• Reassess M9 API/agent assumptions against the APIs and workflows that actually emerged through M6-M8.
• Revise still-open M9 issues before implementation where current evidence invalidates them.

Acceptance criteria

• Wright v1 readiness is supported by repository/test/release evidence rather than issue state alone.
• No known supported-surface semantic regression is carried into release silently.
• M9 starts from the actual stable toolchain/session boundaries produced by v1.
• Any invalidated M9 planning assumption is updated before agents proceed.

Planning note

This is a hard milestone gate. Agents should not treat M9 issue text as immutable until this review completes.

[Teakowa]

M8 gate executed — verified against actual repository evidence

Evidence base: commits 2537396 (emission + transform + v1 gates), 1611004 (scenarios + bench + release); captured reports in the goal scratch dir m8-gates/, m8-release/; release artifact target/wright-0.1.0.tar.gz.

1. S/D/N/E gates over the versioned corpus — scripts/v1-gates.py (against the release binary): all 6 fixtures pass N-level under compat; basic-rule and overpy-cake are byte-identical to the oracle, the rest pass modulo the documented debug/print presentation normalizer. S/D evidence: the M7 differential suite (HIR parity) still passes. Report: target/v1-gates-report.json.

2. High-risk E-level semantics — 7 scenarios (waits, loops, control flow, variables, arrays, subroutines, events+state) with target/runtime metadata, expected behaviors, and machine-readable evidence (target/scenarios-report.json); all pass. Client execution explicitly out of v1 scope (documented).

3. Transformation pipeline — wright-transform with off/compat/aggressive profiles; off verified as a no-op; compat = fold-constants + synthesize-initializers, each evidence-backed (oracle-folded arithmetic, Initialize rule) with N/E regression fixtures and before/after PassStats. Validation runs before and after every pass.

4. Benchmarks — wright-bench measures latency (mean/min/max), peak RSS (getrusage), and generated-resource counts (emitted bytes, WIR nodes) with declared regression thresholds; current run 1.2–2.8 ms/fixture, 4.1 MB RSS, no regressions (target/wright-bench-report.json).

5. Standalone release artifacts — scripts/release.sh builds the optimized binary, runs all gates, proves standalone operation with PATH=/usr/bin:/bin (no node), stamps version metadata, and packages target/wright-0.1.0.tar.gz; the packaged binary reproduces the declared workflows.

6. No silent supported-surface regressions — debug/print formatting, float precision (16 sig digits), condition infix form, subroutine-first rule ordering, and initializer synthesis are all documented in docs/v1-matrix.md; the N-level normalizer records them rather than hiding them.

M9 rebaseline (issues #56–#62):

• [M9] Stabilize Rust embedding API over compiler/session workflows #56: the M6 CompilerSession is the embedding surface to stabilize (typed config/results already exist); no assumption invalidated.
• [M9] Evolve structured tool API into project/session-aware service #57: the driver's semantic service is currently private; M9 exposes it over session handles.
• [M9] Add agent-oriented semantic queries and cost inspection #58: wright-bench's generated-resource metrics (emitted bytes, WIR nodes) are the Wright-owned cost/resource baseline [M9] Add agent-oriented semantic queries and cost inspection #58 must consume; target/catalog metadata queries can reuse the workshop catalog.
• [M9] Define safe source edit and refactoring contracts #59–[M9] Validate external consumer integrations and API versioning #61: no assumption invalidated by M8 evidence.

Gate verdict: M8 ACCEPTED. Local quality gates (fmt/clippy -D warnings/workspace tests) are green; the release pipeline runs end-to-end.