Complete Phase 15 spoken-word module

This commit is contained in:
2026-05-12 11:56:28 +00:00
parent dbf3605712
commit a9f7fa27ff
10 changed files with 945 additions and 78 deletions

View File

@@ -36,14 +36,18 @@ Implemented today:
- Production `grammar` module implementation in `internal/modules/grammar`.
- Production `glossary` module implementation in `internal/modules/glossary`.
- Production `homophones` module implementation in `internal/modules/homophones`.
- Production `spoken_word` module implementation in `internal/modules/spoken_word`.
- Explicit runtime support for `--modules grammar` through the production runner path.
- Explicit runtime support for `--modules glossary`, including repeated stages such as `--modules glossary,glossary`.
- Explicit runtime support for `--modules homophones` through the production runner path.
- Explicit runtime support for `--modules spoken_word` through the production runner path.
Not implemented in CLI runtime path today:
- Real module execution pipeline for `spoken_word`.
- End-to-end transcript polishing with the full default module sequence.
Current reality:
- all production modules exist and are explicitly runnable by `--modules`, but default full-sequence integration remains a later phase.
Phase sequencing note:
- Phase 9 LLM infrastructure is complete (structured client, scheduler, effective config resolution, diagnostics primitives);
- Phase 10 LLM-backed validator runtime integration is complete;
@@ -51,7 +55,8 @@ Phase sequencing note:
- Phase 12 grammar module implementation and explicit runtime wiring are complete;
- Phase 13 glossary module and protected-term behavior are complete;
- Phase 14 homophones module implementation and explicit runtime wiring are complete;
- next recommended phase is Phase 15 (spoken-word module).
- Phase 15 spoken-word module implementation and explicit runtime wiring are complete;
- next recommended phase is Phase 16 (default full pipeline integration).
## Actual Go package layout
@@ -122,6 +127,10 @@ internal/modules/homophones/
module.go
prompt.go
internal/modules/spoken_word/
module.go
prompt.go
internal/framework/validators/
models.go
deterministic.go
@@ -157,17 +166,17 @@ Current runtime flow (`internal/cli/run.go`):
10. Chunk normalized transcript and compute chunk summaries.
11. Write chunking summary artifact.
12. Execute runner modules sequentially when:
- `--modules` is explicitly provided (production grammar/glossary/homophones paths); or
- `--modules` is explicitly provided (production grammar/glossary/homophones/spoken_word paths); or
- a test/injected module factory is provided.
13. Output working transcript to `--output` file or stdout.
14. Build process report (`phase` currently set to `phase14-homophones-module`).
14. Build process report (`phase` currently set to `phase15-spoken-word-module`).
15. Optionally write `--report-json`; always write run-dir `report.json`.
16. Apply work-dir retention.
Important behavior details:
- Glossary is validated and is used for explicit glossary/grammar/homophones module correction paths.
- Glossary is validated and is used for explicit glossary/grammar/homophones/spoken_word module correction paths.
- Default production CLI behavior remains deterministic normalization/chunking/reporting unless modules are explicitly selected with `--modules`.
- Explicit `--modules grammar`, `--modules glossary`, and `--modules homophones` run production module paths with LLM-backed proposal generation and validator-chain execution.
- Explicit `--modules grammar`, `--modules glossary`, `--modules homophones`, and `--modules spoken_word` run production module paths with LLM-backed proposal generation and validator-chain execution.
- Default runs (without explicit module selection) do not perform LLM calls.
- Success path is generally quiet on stderr.
- Source IDs are preserved into a canonical transcript before normalization; normalization then reassigns output IDs sequentially from `1`.
@@ -216,7 +225,7 @@ Implemented config surfaces include:
- work-dir and retention mode
Current caveat:
- LLM/module-related settings are active for explicit grammar/glossary/homophones runs; the default non-explicit path remains deterministic.
- LLM/module-related settings are active for explicit grammar/glossary/homophones/spoken_word runs; the default non-explicit path remains deterministic.
## Implemented structured LLM infrastructure
`internal/framework/contracts` now defines a typed structured-completion contract:
@@ -233,7 +242,7 @@ Current caveat:
Current runtime boundary:
- the default CLI runtime path (without explicit module selection) still does not instantiate the full production module sequence.
- LLM calls are exercised in production when `--modules grammar`, `--modules glossary`, or `--modules homophones` is explicitly requested and in tests when fake/injected clients are used.
- LLM calls are exercised in production when `--modules grammar`, `--modules glossary`, `--modules homophones`, or `--modules spoken_word` is explicitly requested and in tests when fake/injected clients are used.
`internal/framework/llm` also provides:
- a bounded `Scheduler` for controlled concurrent LLM calls with reliable permit release;
@@ -275,7 +284,7 @@ Current behavior details:
`internal/framework/contracts` provides interfaces and run-spec metadata scaffolding, including deterministic repeated module instance naming (`ResolveModuleRunSpecs`).
These primitives are wired into the production runner and report model. The grammar and glossary modules are implemented; other production modules remain pending.
These primitives are wired into the production runner and report model. The grammar, glossary, homophones, and spoken_word modules are implemented.
## Implemented validator runtime infrastructure
`internal/framework/validators` provides deterministic validator infrastructure:
@@ -340,7 +349,7 @@ This helper only produces candidate proposals; validator-chain execution and pro
- diagnostics directory context
- returns explicit errors for unknown keys (`unsupported_module`) and recognized-but-unimplemented keys (`unimplemented_module`).
The `grammar`, `glossary`, and `homophones` module keys are now registered and constructible. `spoken_word` remains recognized-but-unimplemented.
The `grammar`, `glossary`, `homophones`, and `spoken_word` module keys are now registered and constructible.
## Implemented grammar production module
`internal/modules/grammar` now provides the first production module:
@@ -387,6 +396,19 @@ This vocabulary is used by deterministic validators for both glossary-stage and
- protected-term guardrails for non-glossary modules remain active and are exercised through the homophones path;
- module-level reporting and diagnostics capture through existing runner/reporting paths.
## Implemented spoken_word production module
`internal/modules/spoken_word` now provides the fourth production module:
- prompt builder aligned to Python spoken_word-module intent, constrained to conservative dysfluency cleanup;
- strong prompt guardrails preserving meaning/intent/voice/named entities/domain terms and substantive content;
- explicit guardrails against summarization, style rewriting, grammar-only cleanup, punctuation-only cleanup, invention, and meaning-changing rewrites;
- proposal generation through `internal/framework/proposal_generation` and `contracts.StructuredLLMClient`;
- scheduler-aware proposal calls through existing `contracts.LLMScheduler` hooks;
- replacement policy `require_unique` (matching Python spoken_word behavior);
- validator chain integration using existing deterministic + LLM-backed validators, including strong semantic guardrails (`spoken_word_review`, `meaning_reversal_review`);
- spoken_word confidence threshold enforcement through existing validator/config infrastructure;
- protected-term guardrails for non-glossary modules remain active and are exercised through the spoken_word path;
- module-level reporting and diagnostics capture through existing runner/reporting paths.
## Reports and diagnostics (implemented)
Current per-run artifacts include:
- `source-transcript.json`
@@ -426,7 +448,7 @@ Retention modes implemented in `ApplyRetention`:
Current runtime note:
- default non-explicit runs usually have no module-level skipped corrections, so `auto` commonly removes clean successful run directories.
- explicit grammar/glossary/homophones runs can produce validator rejections and application skips, which are reflected in reports and retention input.
- explicit grammar/glossary/homophones/spoken_word runs can produce validator rejections and application skips, which are reflected in reports and retention input.
Intentionally deferred to module/LLM phases:
- real domain proposal prompts and production module implementations remain tied to later module phases.
@@ -450,14 +472,15 @@ Implemented tests currently cover:
- production grammar module prompt constraints, proposal mapping, validator-chain behavior, confidence-threshold enforcement, diagnostics redaction, and explicit CLI/runtime integration (`internal/modules/grammar/*_test.go`, `internal/cli/run_test.go`, `internal/framework/runner/*_test.go`)
- production glossary module prompt constraints, proposal mapping, validator-chain behavior, confidence-threshold enforcement, diagnostics redaction, repeated-stage behavior, and explicit CLI/runtime integration (`internal/modules/glossary/*_test.go`, `internal/cli/run_test.go`, `internal/framework/runner/*_test.go`)
- production homophones module prompt constraints, proposal mapping, validator-chain behavior, confidence-threshold enforcement, diagnostics redaction, protected-term behavior, and explicit CLI/runtime integration (`internal/modules/homophones/*_test.go`, `internal/cli/run_test.go`, `internal/framework/runner/*_test.go`)
- production spoken_word module prompt constraints, proposal mapping, validator-chain behavior, semantic guardrail behavior, confidence-threshold enforcement, diagnostics redaction, protected-term behavior, and explicit CLI/runtime integration (`internal/modules/spoken_word/*_test.go`, `internal/cli/run_test.go`, `internal/framework/runner/*_test.go`)
- glossary-derived protected-term extraction and stable behavior (`internal/framework/validators/protected_terms_test.go`)
Not covered yet (because not implemented): production `spoken_word` module plus full default-sequence transcript-polishing runtime behavior.
Not covered yet (because not implemented): full default-sequence transcript-polishing runtime behavior as a single default path.
## Intended final architecture (not yet implemented)
The intended end-state still matches the rewrite plan:
- sequential module pipeline over a mutable working transcript
- real module implementation (`spoken_word`)
- default full module-sequence integration in the standard runtime path
- structured LLM proposal generation
- deterministic and LLM validators
- validator cardinality enforcement in pipeline execution

View File

@@ -68,6 +68,10 @@ Implemented:
- Production homophones module package with Python-aligned prompt intent and guardrails.
- Explicit `--modules homophones` runtime path through runner, shared proposal generation, validators, application, reporting, and diagnostics.
- Focused multi-module runtime tests for already-implemented interoperability (for example `glossary,homophones`) without claiming full default-pipeline completion.
- Production spoken_word module package with Python-aligned prompt intent and guardrails.
- Explicit `--modules spoken_word` runtime path through runner, shared proposal generation, validators, application, reporting, and diagnostics.
- Focused multi-module runtime tests for already-implemented interoperability (for example `spoken_word,grammar`) without claiming full default-pipeline completion.
- All production modules now exist (`glossary`, `homophones`, `spoken_word`, `grammar`), but default full-sequence integration remains Phase 16 work.
- Broad deterministic and CLI/subprocess test coverage for implemented phases through `go test ./...`.
- Internal typed structured LLM contract (`StructuredLLMClient.CompleteStructured(ctx, req, out)`).
- `internal/framework/llm` instructor-go-backed adapter with:
@@ -83,8 +87,6 @@ Implemented:
- Generic JSON diagnostics primitives for LLM interactions (request metadata, request payload, response payload, optional error payload) with secret redaction.
Not yet implemented in runtime pipeline:
- Real correction module for `spoken_word`.
- Domain proposal prompts for remaining real module.
- End-to-end transcript polishing behavior with the full default module sequence.
## Completed phases
@@ -223,7 +225,7 @@ Not implemented in Phase 8 (by design):
## Remaining work plan
Next recommended phase: **Phase 15 (spoken-word module)**.
Next recommended phase: **Phase 16 (default full pipeline integration)**.
## Phase 9: Structured LLM client and scheduler infrastructure
@@ -418,41 +420,29 @@ Not implemented in Phase 14 (by design):
## Phase 15: Spoken-word module
### Purpose
Completed.
Implement conservative dysfluency cleanup while preserving substantive meaning.
Implemented:
- Production spoken_word module package in `internal/modules/spoken_word`.
- Spoken_word prompt builder aligned to Python intent and constrained to conservative dysfluency cleanup.
- Prompt context using glossary/protected-term information (names, aliases, plurals where present) to avoid damaging known domain terms.
- Strong prompt guardrails preserving meaning, intent, speaker voice, named entities, game/domain terms, and substantive content.
- Explicit prompt guardrails against summarization, style rewriting, grammar-only cleanup, punctuation-only cleanup, invention, event reordering, and certainty inflation.
- Spoken_word proposal generation through shared `internal/framework/proposal_generation` using `contracts.StructuredLLMClient`.
- Scheduler-aware spoken_word proposal generation through existing scheduler hooks.
- Spoken_word replacement policy `require_unique` (matching Python behavior).
- Spoken_word validator chain using existing deterministic and LLM-backed validators.
- Strong semantic guardrails in runtime validator chain through existing LLM-backed validators (`spoken_word_review`, `meaning_reversal_review`).
- Spoken-word confidence threshold enforcement through existing config + confidence-threshold validator behavior.
- Protected-term guardrails remaining active for spoken_word via existing deterministic validators.
- Explicit runtime support for `--modules spoken_word` through normalization, chunking, runner, proposal generation, validation, application, and reporting.
- Prompt/response diagnostics artifacts for spoken_word proposal + validator interactions with secret redaction.
- Module-level reports for spoken_word including generated proposals, validator decisions/rejections, applied changes, and application skips.
- CLI/runtime fake-client tests for approved cleanup, validator rejection, meaning-changing rejection, application skips, diagnostics, protected-term rejection behavior, failure/error.log behavior, and report outputs (`--report-json` and run-dir `report.json`).
- Focused interoperability tests for already-implemented module combinations (for example `spoken_word,grammar`) to verify working-transcript handoff and guardrails without claiming full default-sequence parity.
### Scope
Implement:
- `spoken_word` module package.
- Spoken-word prompt builder ported from Python.
- Spoken-word structured response model.
- Spoken-word replacement policy.
- Spoken-word confidence threshold handling.
- Spoken-word validator chain.
- Strong semantic guardrails using LLM-backed validators.
- Prompt/response diagnostics.
- CLI support for `--modules spoken_word`.
- Fake LLM tests.
- Tests for rejection of meaning-changing cleanup.
Do not implement:
- Prompt redesign beyond faithful porting.
- New stylistic rewriting behavior not present in the Python implementation.
### Expected behavior at end of phase
Running `audita process ... --modules spoken_word` should perform real conservative dysfluency cleanup, with guardrails against semantic changes.
### Definition of done
- Spoken-word module runs in the production runner.
- Spoken-word proposals are generated through structured LLM calls.
- Semantic validators reject meaning-changing proposals.
- Applied/skipped spoken-word changes appear in reports.
- Prompt/response diagnostics are written.
- `go test ./...` passes without requiring external LLM credentials.
Not implemented in Phase 15 (by design):
- Full default module sequence execution as a feature-complete claim.
## Phase 16: Default full pipeline integration