Add Phase 2 fixtures and compatibility tests

docs/rewrite-notes.md

@@ -551,20 +551,64 @@ Before accepting a phase implementation, check:
 - Does `go test ./...` pass?
 - Is the code idiomatic Go rather than Python-shaped Go?
 
-## Red flags during the rewrite
+## Phase 2 Status (Go Implementation)
 
-Watch for:
+The Go rewrite has completed Phase 2 with the following capabilities:
 
-- module stages running concurrently and changing pipeline semantics;
-- prompt wording changes mixed into infrastructure commits;
-- validators silently ignoring malformed LLM output;
-- proposal application mutating text without report entries;
-- report data printed to stdout unexpectedly;
-- API keys appearing in diagnostics;
-- global mutable LLM clients that make tests order-dependent;
-- unbounded goroutine creation;
-- filesystem paths hard-coded outside config defaults;
-- tests that require a real LLM when a fake LLM would be better.
+### Implemented Features:
+
+1. **Typed Schema Validation**:
+   - Transcript validation supports both bare segment arrays and objects with `segments` field
+   - Glossary validation matches the existing YAML format
+   - Comprehensive field validation and error reporting
+
+2. **Deterministic Normalization**:
+   - Chronological sorting of segments
+   - Same-speaker adjacent segment merging with configurable gap thresholds
+   - Ellipsis insertion for gaps above the ellipsis threshold
+   - Configurable maximum segment duration and token limits
+   - Sequential ID reassignment starting at 1
+   - Category preservation during merging
+
+3. **Process Output**:
+   - Output is normalized transcript JSON (not raw passthrough)
+   - Canonical transcript shape as specified in architecture docs
+   - Proper stdout/stderr discipline preserved from Phase 1
+
+4. **Enhanced Reporting**:
+   - Success reports include normalization statistics
+   - Failure reports include error phase and concise messages
+   - No secrets leaked to reports
+
+5. **Minimal Diagnostics**:
+   - Per-run work directories with unique timestamps
+   - Source transcript artifacts (raw and parsed)
+   - Normalized transcript artifacts
+   - Normalization summary artifacts
+   - Authoritative report.json in run directory
+   - Error logs on failure
+   - Basic retention policy implementation
+
+### Not Yet Implemented (Future Phases):
+
+- Chunking and token-bounded section building
+- LLM proposal generation and structured output
+- Validator chains and decision logging
+- Module execution pipeline
+- Full diagnostics retention policy
+- Skipped correction detection for auto retention
+- Real module implementations (glossary, homophones, spoken_word, grammar)
+
+### Compatibility Notes:
+
+The Go implementation maintains the public CLI contract from the Python implementation:
+- `audita process transcript.json --glossary glossary.yaml --output corrected.json`
+- stdout contains transcript JSON when --output is omitted
+- stderr contains logs and errors only
+- --report-json writes machine-readable reports
+- Nonzero exit codes on failure
+
+The normalization behavior is deterministic and matches the architectural specifications, but may differ from Python-era behavior in edge cases until full module implementation is complete.
 
 ## When to consider an HTTP API