Complete Phase 8 deterministic validators

2026-05-12 00:26:27 +00:00
parent 28fe899aa1
commit aeb31f1c0d
13 changed files with 940 additions and 115 deletions
--- a/docs/architecture.md
+++ b/docs/architecture.md
@@ -19,14 +19,21 @@ Implemented today:
 - Framework foundation packages for contracts and proposal application.
 - Production runner orchestration package with deterministic sequential module execution.
 - Module-level report structures with applied/skipped change records.
+- Runtime validator models and deterministic validators.
+- Deterministic validator-chain execution in the runner with cardinality enforcement.
+- Module-level validator decision/rejection reporting.

 Not implemented in CLI runtime path today:
 - Real module execution pipeline (`glossary`, `homophones`, `spoken_word`, `grammar`).
 - Structured LLM proposal generation.
- Validator chain execution.
+- LLM-backed validators.
 - Production LLM scheduler behavior.
 - End-to-end transcript polishing with real module behavior.

+Phase sequencing note:
+- structured LLM client and scheduler infrastructure remain Phase 9 work;
+- LLM-backed validators remain Phase 10 work.
+
 ## Actual Go package layout

 ```text
@@ -77,6 +84,10 @@ internal/framework/proposals/

 internal/framework/runner/
  runner.go
+
+internal/framework/validators/
+  models.go
+  deterministic.go
 ```

 ## Current CLI behavior
@@ -100,7 +111,7 @@ Current runtime flow (`internal/cli/run.go`):
 11. Write chunking summary artifact.
 12. Optionally execute runner modules sequentially when an injected module registry/factory is available (used by deterministic tests today).
 13. Output working transcript to `--output` file or stdout.
-14. Build process report (`phase` currently set to `phase7-runner`).
+14. Build process report (`phase` currently set to `phase8-validators`).
 15. Optionally write `--report-json`; always write run-dir `report.json`.
 16. Apply work-dir retention.

@@ -194,6 +205,24 @@ Current behavior details:

 These primitives are wired into the production runner and report model. Real module implementations are still pending.

+## Implemented validator runtime infrastructure
+`internal/framework/validators` provides deterministic validator infrastructure:
+- runtime validation request/result models;
+- stable validator reason codes;
+- cardinality enforcement for validator decisions:
+  - missing proposal indexes fail
+  - duplicate proposal indexes fail
+  - unknown proposal indexes fail
+- deterministic validators:
+  - confidence threshold by module key/config threshold
+  - original-text presence against current working transcript
+  - non-empty corrected text
+  - identical/no-effect rejection
+  - conservative protected glossary-term guard for non-glossary modules
+
+`internal/framework/runner` executes validator chains in order for each module and applies only validator-approved proposals.
+Validator rejections are reported distinctly from proposal-application skips.
+
 ## Reports and diagnostics (implemented)
 Current per-run artifacts include:
 - `source-transcript.json`
@@ -222,6 +251,7 @@ Current process reports include diagnostics metadata references for:
 Current process reports also include:
 - module-level results (when runner modules execute), including applied/skipped proposal changes;
 - run-level module summary totals and failed module instance metadata.
+- module-level validator decisions and validator rejections.

 Retention modes implemented in `ApplyRetention`:
 - `always`: keep all run directories.
@@ -247,8 +277,9 @@ Implemented tests currently cover:
 - contracts/foundation composition tests (`internal/framework/contracts/*_test.go`)
 - runner sequencing and failure behavior with deterministic fake modules (`internal/framework/runner/*_test.go`)
 - CLI runner integration through injected fake module factories (`internal/cli/run_test.go`)
+- validator models, cardinality enforcement, and deterministic validators (`internal/framework/validators/*_test.go`)

-Not covered yet (because not implemented): validator runtime flow with approvals/rejections and real LLM integration.
+Not covered yet (because not implemented): real LLM validator/runtime integration and production module behavior.

 ## Intended final architecture (not yet implemented)
 The intended end-state still matches the rewrite plan:
--- a/docs/rewrite-notes.md
+++ b/docs/rewrite-notes.md
@@ -47,12 +47,16 @@ Implemented:
 - Production runner orchestration over a mutable working transcript.
 - Module-level report structures and run-level module summaries.
 - CLI runner integration point via injectable module factory/registry (used by deterministic tests).
+- Runtime validator models and deterministic validator implementations.
+- Validator cardinality enforcement (missing/duplicate/unknown proposal index errors).
+- Deterministic validator-chain execution in the production runner.
+- Module reports including validator decisions and validator rejections.
 - Broad deterministic and CLI/subprocess test coverage for implemented phases through `go test ./...`.

 Not yet implemented in runtime pipeline:
 - Real correction modules.
- Validator-chain execution.
 - Structured LLM client integration.
+- LLM-backed validators.
 - Prompt/response diagnostics for LLM calls.
 - End-to-end transcript polishing behavior.

@@ -161,21 +165,12 @@ Not implemented in Phase 7 (by design):
 Current runtime behavior note:
 - Default user-facing CLI behavior remains deterministic normalization/chunking output unless test-only module injection is used during tests.

-## Remaining work plan
-
-Next recommended phase: **Phase 8 (runtime validator framework and deterministic validators)**.
-
 ## Phase 8: Runtime validator framework and deterministic validators
+Completed.

-### Purpose
-
-Wire validator-chain execution into the runner using deterministic validators first. This phase establishes the safety model before any real LLM proposal generation is introduced.
-
-### Scope
-
-Implement:
- Runtime validator interfaces if the existing contracts need refinement.
- Validation request/result models.
+Implemented:
+- Runtime validator request/result models in `internal/framework/validators`.
+- Deterministic validator reason codes for stable reporting.
 - Validator cardinality enforcement:
  - one decision per candidate proposal index
  - missing indexes are errors
@@ -183,33 +178,25 @@ Implement:
  - unknown indexes are errors
 - Deterministic validators:
  - confidence threshold
-  - original-text presence
+  - original-text presence against working transcript
  - non-empty correction
-  - identical text/no-effect rejection
-  - protected glossary term logic, if it can be implemented deterministically from current glossary schema
- Validator ordering.
- Runner integration so candidate proposals pass through validators before application.
- Module report fields for validator approvals/rejections.
+  - identical/no-effect rejection
+  - conservative protected glossary-term guard
+- Ordered validator-chain execution in the production runner.
+- Runner behavior where only validator-approved proposals proceed to proposal application.
+- Module-level reporting of validator decisions and validator rejections, distinct from application-level skips.
+- Deterministic fake-module tests covering approvals, rejections, validator order/filtering, and cardinality failure pipeline-stop behavior.

-Do not implement:
- LLM-backed validators.
- Real modules.
- Real LLM proposal generation.
+Not implemented in Phase 8 (by design):
+- LLM-backed validators (Phase 10).
+- Structured LLM client implementation or scheduler behavior (Phase 9).
+- Real correction modules.
 - Prompt/response diagnostics.
+- End-to-end transcript polishing.

-### Expected behavior at end of phase
+## Remaining work plan

-Fake modules can generate deterministic proposals, those proposals can be filtered by deterministic validators, and only approved proposals are applied.
-
-### Definition of done
-
- Validator chains run in the production runner.
- Deterministic validators are implemented and tested.
- Validator cardinality enforcement is tested.
- Rejected proposals appear in module reports with stable reasons.
- Approved proposals are applied through existing proposal application semantics.
- No real LLM calls occur.
- `go test ./...` passes.
+Next recommended phase: **Phase 9 (structured LLM client and scheduler infrastructure)**.

 ## Phase 9: Structured LLM client and scheduler infrastructure

--- a/internal/cli/run.go
+++ b/internal/cli/run.go
@@ -361,7 +361,7 @@ func runProcess(args []string, stdout, stderr io.Writer) int {
 	hasSkippedCorrections := false
 	if runOutput != nil {
 		for _, mr := range runOutput.ModuleResults {
-			if len(mr.SkippedChanges) > 0 {
+			if len(mr.SkippedChanges) > 0 || len(mr.ValidatorRejected) > 0 {
 				hasSkippedCorrections = true
 				break
 			}
@@ -395,7 +395,7 @@ func extractErrorPhase(err error) (phase string, message string) {

 func buildProcessReport(status string, inv processInvocation, runDir *diagnostics.RunDirectory, startedAt, completedAt time.Time, errorMessage string, errorPhase string, normalizationSummary *normalization.NormalizationSummary, chunkingSummary *chunking.Summary, runOutput *runner.RunOutput) reporting.ProcessReport {
 	report := reporting.ProcessReport{
-		Phase:          "phase7-runner",
+		Phase:          "phase8-validators",
 		Status:         status,
 		Operation:      "process",
 		TranscriptPath: inv.TranscriptPath,
@@ -461,19 +461,21 @@ func buildModuleReporting(runOutput *runner.RunOutput) (*reporting.ModulesSummar
 		startedAt := r.StartedAt
 		completedAt := r.CompletedAt
 		moduleReports = append(moduleReports, reporting.ModuleReport{
-			ModuleKey:         r.ModuleKey,
-			ModuleInstance:    r.ModuleInstance,
-			ReplacementPolicy: string(r.ReplacementPolicy),
-			Status:            r.Status,
-			ProposalCount:     r.ProposalCount,
-			AppliedChanges:    r.AppliedChanges,
-			SkippedChanges:    r.SkippedChanges,
-			ErrorMessage:      r.ErrorMessage,
-			StartedAt:         &startedAt,
-			CompletedAt:       &completedAt,
+			ModuleKey:          r.ModuleKey,
+			ModuleInstance:     r.ModuleInstance,
+			ReplacementPolicy:  string(r.ReplacementPolicy),
+			Status:             r.Status,
+			ProposalCount:      r.ProposalCount,
+			ValidatorDecisions: mapValidatorDecisions(r.ValidatorDecisions),
+			ValidatorRejected:  mapValidatorRejected(r.ValidatorRejected),
+			AppliedChanges:     r.AppliedChanges,
+			SkippedChanges:     r.SkippedChanges,
+			ErrorMessage:       r.ErrorMessage,
+			StartedAt:          &startedAt,
+			CompletedAt:        &completedAt,
 		})
 		summary.TotalAppliedChanges += len(r.AppliedChanges)
-		summary.TotalSkippedChanges += len(r.SkippedChanges)
+		summary.TotalSkippedChanges += len(r.SkippedChanges) + len(r.ValidatorRejected)
 		if r.Status == runner.ModuleStatusFailed && summary.FailedModuleInstance == "" {
 			summary.FailedModuleInstance = r.ModuleInstance
 		}
@@ -482,6 +484,44 @@ func buildModuleReporting(runOutput *runner.RunOutput) (*reporting.ModulesSummar
 	return summary, moduleReports
 }

+func mapValidatorDecisions(in []runner.ValidatorDecisionRecord) []reporting.ValidatorDecisionReport {
+	if len(in) == 0 {
+		return nil
+	}
+	out := make([]reporting.ValidatorDecisionReport, len(in))
+	for i, d := range in {
+		out[i] = reporting.ValidatorDecisionReport{
+			ValidatorName: d.ValidatorName,
+			ProposalIndex: d.ProposalIndex,
+			Approved:      d.Approved,
+			ReasonCode:    d.ReasonCode,
+			Message:       d.Message,
+		}
+	}
+	return out
+}
+
+func mapValidatorRejected(in []runner.ValidatorRejectedChange) []reporting.ValidatorRejectedReport {
+	if len(in) == 0 {
+		return nil
+	}
+	out := make([]reporting.ValidatorRejectedReport, len(in))
+	for i, d := range in {
+		out[i] = reporting.ValidatorRejectedReport{
+			ValidatorName:   d.ValidatorName,
+			ProposalIndex:   d.ProposalIndex,
+			ModuleKey:       d.ModuleKey,
+			ModuleInstance:  d.ModuleInstance,
+			TargetSegmentID: d.TargetSegmentID,
+			OriginalText:    d.OriginalText,
+			CorrectedText:   d.CorrectedText,
+			ReasonCode:      d.ReasonCode,
+			Message:         d.Message,
+		}
+	}
+	return out
+}
+
 type processFlags struct {
 	glossaryPath                  *string
 	outputPath                    *string
--- a/internal/cli/run_test.go
+++ b/internal/cli/run_test.go
@@ -15,6 +15,7 @@ import (
 	"gitea.maximumdirect.net/eric/audita/internal/core/schema"
 	"gitea.maximumdirect.net/eric/audita/internal/framework/contracts"
 	"gitea.maximumdirect.net/eric/audita/internal/framework/proposals"
+	"gitea.maximumdirect.net/eric/audita/internal/framework/validators"
 )

 func TestRunRootHelp(t *testing.T) {
@@ -613,8 +614,8 @@ func TestRunProcessReportJSONIncludesChunkingSummary(t *testing.T) {
 	if report.Chunking.MaxSectionTokens == 0 {
 		t.Errorf("expected max_section_tokens in report")
 	}
-	if report.Phase != "phase7-runner" {
-		t.Errorf("expected phase 'phase7-runner', got %q", report.Phase)
+	if report.Phase != "phase8-validators" {
+		t.Errorf("expected phase 'phase8-validators', got %q", report.Phase)
 	}
 }

@@ -631,14 +632,15 @@ func (f fakeModuleFactory) ModuleForSpec(spec contracts.ModuleRunSpec) (contract
 }

 type fakeModule struct {
-	key      string
-	policy   proposals.ReplacementPolicy
-	proposeF func(req contracts.ProposalRequest) ([]proposals.CorrectionProposal, error)
+	key        string
+	policy     proposals.ReplacementPolicy
+	validators []contracts.Validator
+	proposeF   func(req contracts.ProposalRequest) ([]proposals.CorrectionProposal, error)
 }

 func (m fakeModule) Key() string                                    { return m.key }
 func (m fakeModule) ReplacementPolicy() proposals.ReplacementPolicy { return m.policy }
-func (m fakeModule) Validators() []contracts.Validator              { return nil }
+func (m fakeModule) Validators() []contracts.Validator              { return m.validators }
 func (m fakeModule) Propose(ctx context.Context, req contracts.ProposalRequest) ([]proposals.CorrectionProposal, error) {
 	if m.proposeF == nil {
 		return nil, nil
@@ -646,14 +648,32 @@ func (m fakeModule) Propose(ctx context.Context, req contracts.ProposalRequest)
 	return m.proposeF(req)
 }

+type fakeValidator struct {
+	name      string
+	validateF func(req contracts.ValidationRequest) (validators.Result, error)
+}
+
+func (v fakeValidator) Name() string { return v.name }
+func (v fakeValidator) Validate(ctx context.Context, req contracts.ValidationRequest) (validators.Result, error) {
+	_ = ctx
+	return v.validateF(req)
+}
+
 func TestRunProcessInjectedFactoryExecutesRunnerAndReportsModules(t *testing.T) {
+	allow := fakeValidator{name: "allow", validateF: func(req contracts.ValidationRequest) (validators.Result, error) {
+		decisions := make([]validators.Decision, len(req.CandidateProposal))
+		for i, p := range req.CandidateProposal {
+			decisions[i] = validators.Decision{ProposalIndex: p.ProposalIndex, Approved: true, ReasonCode: validators.ReasonApproved, Message: "approved"}
+		}
+		return validators.Result{ValidatorName: "allow", Decisions: decisions}, nil
+	}}
 	processModuleFactory = fakeModuleFactory{modules: map[string]contracts.TranscriptModule{
-		"m1": fakeModule{key: "m1", policy: proposals.ReplacementPolicyRequireUnique, proposeF: func(req contracts.ProposalRequest) ([]proposals.CorrectionProposal, error) {
+		"m1": fakeModule{key: "m1", policy: proposals.ReplacementPolicyRequireUnique, validators: []contracts.Validator{allow}, proposeF: func(req contracts.ProposalRequest) ([]proposals.CorrectionProposal, error) {
 			return []proposals.CorrectionProposal{
 				{TargetSegmentID: 1, OriginalText: "Hello", CorrectedText: "Hi", Confidence: 1},
 			}, nil
 		}},
-		"m2": fakeModule{key: "m2", policy: proposals.ReplacementPolicyRequireUnique, proposeF: func(req contracts.ProposalRequest) ([]proposals.CorrectionProposal, error) {
+		"m2": fakeModule{key: "m2", policy: proposals.ReplacementPolicyRequireUnique, validators: []contracts.Validator{allow}, proposeF: func(req contracts.ProposalRequest) ([]proposals.CorrectionProposal, error) {
 			if req.WorkingTranscript.Segments[0].Text != "Hi world" {
 				t.Fatalf("expected module 2 to see module 1 changes, got %q", req.WorkingTranscript.Segments[0].Text)
 			}
@@ -711,11 +731,17 @@ func TestRunProcessInjectedFactoryExecutesRunnerAndReportsModules(t *testing.T)
 	if len(report.ModuleResults[1].SkippedChanges) != 1 {
 		t.Fatalf("expected skipped change for module 2, got %+v", report.ModuleResults[1].SkippedChanges)
 	}
+	if len(report.ModuleResults[0].ValidatorDecisions) == 0 {
+		t.Fatalf("expected validator decisions in module report")
+	}

 	runDirReport := readProcessReport(t, filepath.Join(onlyRunDir(t, workDir), "report.json"))
 	if len(runDirReport.ModuleResults) != 2 {
 		t.Fatalf("expected module results in run-dir report")
 	}
+	if len(runDirReport.ModuleResults[1].ValidatorDecisions) == 0 {
+		t.Fatalf("expected validator decisions in run-dir report")
+	}
 }

 func TestRunProcessInjectedFactorySkippedKeepsAutoRetention(t *testing.T) {
--- a/internal/core/reporting/report.go
+++ b/internal/core/reporting/report.go
@@ -33,16 +33,38 @@ type ProcessReport struct {
 }

 type ModuleReport struct {
-	ModuleKey         string                    `json:"module_key"`
-	ModuleInstance    string                    `json:"module_instance"`
-	ReplacementPolicy string                    `json:"replacement_policy,omitempty"`
-	Status            string                    `json:"status"`
-	ProposalCount     int                       `json:"proposal_count"`
-	AppliedChanges    []proposals.AppliedChange `json:"applied_changes,omitempty"`
-	SkippedChanges    []proposals.SkippedChange `json:"skipped_changes,omitempty"`
-	ErrorMessage      string                    `json:"error_message,omitempty"`
-	StartedAt         *time.Time                `json:"started_at,omitempty"`
-	CompletedAt       *time.Time                `json:"completed_at,omitempty"`
+	ModuleKey          string                    `json:"module_key"`
+	ModuleInstance     string                    `json:"module_instance"`
+	ReplacementPolicy  string                    `json:"replacement_policy,omitempty"`
+	Status             string                    `json:"status"`
+	ProposalCount      int                       `json:"proposal_count"`
+	ValidatorDecisions []ValidatorDecisionReport `json:"validator_decisions,omitempty"`
+	ValidatorRejected  []ValidatorRejectedReport `json:"validator_rejected,omitempty"`
+	AppliedChanges     []proposals.AppliedChange `json:"applied_changes,omitempty"`
+	SkippedChanges     []proposals.SkippedChange `json:"skipped_changes,omitempty"`
+	ErrorMessage       string                    `json:"error_message,omitempty"`
+	StartedAt          *time.Time                `json:"started_at,omitempty"`
+	CompletedAt        *time.Time                `json:"completed_at,omitempty"`
+}
+
+type ValidatorDecisionReport struct {
+	ValidatorName string `json:"validator_name"`
+	ProposalIndex int    `json:"proposal_index"`
+	Approved      bool   `json:"approved"`
+	ReasonCode    string `json:"reason_code"`
+	Message       string `json:"message"`
+}
+
+type ValidatorRejectedReport struct {
+	ValidatorName   string `json:"validator_name"`
+	ProposalIndex   int    `json:"proposal_index"`
+	ModuleKey       string `json:"module_key"`
+	ModuleInstance  string `json:"module_instance"`
+	TargetSegmentID int    `json:"target_segment_id"`
+	OriginalText    string `json:"original_text"`
+	CorrectedText   string `json:"corrected_text"`
+	ReasonCode      string `json:"reason_code"`
+	Message         string `json:"message"`
 }

 type ModulesSummary struct {
--- a/internal/core/reporting/report_test.go
+++ b/internal/core/reporting/report_test.go
@@ -11,7 +11,7 @@ import (
 func TestProcessReportModuleResultsJSONSuccessAndSkipped(t *testing.T) {
 	now := time.Now().UTC()
 	report := ProcessReport{
-		Phase:  "phase7-runner",
+		Phase:  "phase8-validators",
 		Status: "success",
 		ModuleResults: []ModuleReport{
 			{
@@ -20,6 +20,24 @@ func TestProcessReportModuleResultsJSONSuccessAndSkipped(t *testing.T) {
 				ReplacementPolicy: "require_unique",
 				Status:            "success",
 				ProposalCount:     2,
+				ValidatorDecisions: []ValidatorDecisionReport{{
+					ValidatorName: "confidence_threshold",
+					ProposalIndex: 0,
+					Approved:      true,
+					ReasonCode:    "approved",
+					Message:       "approved",
+				}},
+				ValidatorRejected: []ValidatorRejectedReport{{
+					ValidatorName:   "confidence_threshold",
+					ProposalIndex:   1,
+					ModuleKey:       "grammar",
+					ModuleInstance:  "grammar_1",
+					TargetSegmentID: 1,
+					OriginalText:    "bad",
+					CorrectedText:   "worse",
+					ReasonCode:      "low_confidence",
+					Message:         "rejected",
+				}},
 				AppliedChanges: []proposals.AppliedChange{{
 					ProposalIndex: 0,
 					ModuleKey:     "grammar",
@@ -51,6 +69,9 @@ func TestProcessReportModuleResultsJSONSuccessAndSkipped(t *testing.T) {
 	if len(parsed.ModuleResults) != 1 {
 		t.Fatalf("expected one module result, got %d", len(parsed.ModuleResults))
 	}
+	if len(parsed.ModuleResults[0].ValidatorDecisions) != 1 || len(parsed.ModuleResults[0].ValidatorRejected) != 1 {
+		t.Fatalf("expected validator result payloads in module report")
+	}
 	if parsed.ModulesSummary == nil || parsed.ModulesSummary.TotalSkippedChanges != 1 {
 		t.Fatalf("unexpected module summary: %+v", parsed.ModulesSummary)
 	}
@@ -59,7 +80,7 @@ func TestProcessReportModuleResultsJSONSuccessAndSkipped(t *testing.T) {
 func TestProcessReportModuleResultsJSONFailedModule(t *testing.T) {
 	now := time.Now().UTC()
 	report := ProcessReport{
-		Phase:  "phase7-runner",
+		Phase:  "phase8-validators",
 		Status: "failed",
 		ModuleResults: []ModuleReport{
 			{
--- a/internal/framework/contracts/contracts.go
+++ b/internal/framework/contracts/contracts.go
@@ -9,6 +9,7 @@ import (
 	"gitea.maximumdirect.net/eric/audita/internal/core/config"
 	"gitea.maximumdirect.net/eric/audita/internal/core/schema"
 	"gitea.maximumdirect.net/eric/audita/internal/framework/proposals"
+	"gitea.maximumdirect.net/eric/audita/internal/framework/validators"
 )

 // StructuredLLMClient provides provider-agnostic structured completion.
@@ -27,7 +28,7 @@ type TranscriptModule interface {
 // Validator evaluates candidate proposals and returns one decision per proposal index.
 type Validator interface {
 	Name() string
-	Validate(ctx context.Context, req ValidationRequest) ([]ValidationDecision, error)
+	Validate(ctx context.Context, req ValidationRequest) (validators.Result, error)
 }

 // StructuredCompletionRequest is a transport-neutral structured completion request.
@@ -90,19 +91,7 @@ type ProposalRequest struct {
 }

 // ValidationRequest is the input to validator execution.
-type ValidationRequest struct {
-	ExecutionContext
-	RunSpec            ModuleRunSpec                          `json:"run_spec"`
-	CandidateProposals []proposals.EnrichedCorrectionProposal `json:"candidate_proposals"`
-}
-
-// ValidationDecision is one validator decision for one proposal index.
-type ValidationDecision struct {
-	ProposalIndex int      `json:"proposal_index"`
-	Approved      bool     `json:"approved"`
-	Confidence    *float64 `json:"confidence,omitempty"`
-	Reason        string   `json:"reason,omitempty"`
-}
+type ValidationRequest = validators.Request

 // ResolveModuleRunSpecs deterministically resolves instance names from logical keys.
 // Repeated keys are suffixed with _<n> (1-based), while singleton keys keep their raw key.
--- a/internal/framework/contracts/contracts_test.go
+++ b/internal/framework/contracts/contracts_test.go
@@ -8,6 +8,7 @@ import (

 	"gitea.maximumdirect.net/eric/audita/internal/core/chunking"
 	"gitea.maximumdirect.net/eric/audita/internal/framework/proposals"
+	"gitea.maximumdirect.net/eric/audita/internal/framework/validators"
 )

 type fakeLLMClient struct{}
@@ -22,13 +23,13 @@ type fakeValidator struct{}

 func (f *fakeValidator) Name() string { return "fake-validator" }

-func (f *fakeValidator) Validate(ctx context.Context, req ValidationRequest) ([]ValidationDecision, error) {
+func (f *fakeValidator) Validate(ctx context.Context, req ValidationRequest) (validators.Result, error) {
 	_ = ctx
-	decisions := make([]ValidationDecision, len(req.CandidateProposals))
-	for i, proposal := range req.CandidateProposals {
-		decisions[i] = ValidationDecision{ProposalIndex: proposal.ProposalIndex, Approved: true}
+	decisions := make([]validators.Decision, len(req.CandidateProposal))
+	for i, proposal := range req.CandidateProposal {
+		decisions[i] = validators.Decision{ProposalIndex: proposal.ProposalIndex, Approved: true, ReasonCode: validators.ReasonApproved, Message: "approved"}
 	}
-	return decisions, nil
+	return validators.Result{ValidatorName: f.Name(), Decisions: decisions}, nil
 }

 type fakeModule struct{}
--- a/internal/framework/runner/runner.go
+++ b/internal/framework/runner/runner.go
@@ -9,6 +9,7 @@ import (
 	"gitea.maximumdirect.net/eric/audita/internal/core/schema"
 	"gitea.maximumdirect.net/eric/audita/internal/framework/contracts"
 	"gitea.maximumdirect.net/eric/audita/internal/framework/proposals"
+	"gitea.maximumdirect.net/eric/audita/internal/framework/validators"
 )

 const (
@@ -28,16 +29,38 @@ type Runner struct {

 // ModuleResult captures deterministic per-module execution output.
 type ModuleResult struct {
-	ModuleKey         string                      `json:"module_key"`
-	ModuleInstance    string                      `json:"module_instance"`
-	ReplacementPolicy proposals.ReplacementPolicy `json:"replacement_policy"`
-	Status            string                      `json:"status"`
-	ProposalCount     int                         `json:"proposal_count"`
-	AppliedChanges    []proposals.AppliedChange   `json:"applied_changes,omitempty"`
-	SkippedChanges    []proposals.SkippedChange   `json:"skipped_changes,omitempty"`
-	ErrorMessage      string                      `json:"error_message,omitempty"`
-	StartedAt         time.Time                   `json:"started_at"`
-	CompletedAt       time.Time                   `json:"completed_at"`
+	ModuleKey          string                      `json:"module_key"`
+	ModuleInstance     string                      `json:"module_instance"`
+	ReplacementPolicy  proposals.ReplacementPolicy `json:"replacement_policy"`
+	Status             string                      `json:"status"`
+	ProposalCount      int                         `json:"proposal_count"`
+	ValidatorDecisions []ValidatorDecisionRecord   `json:"validator_decisions,omitempty"`
+	ValidatorRejected  []ValidatorRejectedChange   `json:"validator_rejected,omitempty"`
+	AppliedChanges     []proposals.AppliedChange   `json:"applied_changes,omitempty"`
+	SkippedChanges     []proposals.SkippedChange   `json:"skipped_changes,omitempty"`
+	ErrorMessage       string                      `json:"error_message,omitempty"`
+	StartedAt          time.Time                   `json:"started_at"`
+	CompletedAt        time.Time                   `json:"completed_at"`
+}
+
+type ValidatorDecisionRecord struct {
+	ValidatorName string `json:"validator_name"`
+	ProposalIndex int    `json:"proposal_index"`
+	Approved      bool   `json:"approved"`
+	ReasonCode    string `json:"reason_code"`
+	Message       string `json:"message"`
+}
+
+type ValidatorRejectedChange struct {
+	ValidatorName   string `json:"validator_name"`
+	ProposalIndex   int    `json:"proposal_index"`
+	ModuleKey       string `json:"module_key"`
+	ModuleInstance  string `json:"module_instance"`
+	TargetSegmentID int    `json:"target_segment_id"`
+	OriginalText    string `json:"original_text"`
+	CorrectedText   string `json:"corrected_text"`
+	ReasonCode      string `json:"reason_code"`
+	Message         string `json:"message"`
 }

 // RunInput is the deterministic runner input.
@@ -121,19 +144,100 @@ func (r *Runner) Run(ctx context.Context, input RunInput) (RunOutput, error) {
 			})
 		}

-		applyResult := proposals.ApplyProposals(working, enriched, policy)
+		validatorDecisions := make([]ValidatorDecisionRecord, 0)
+		validatorRejected := make([]ValidatorRejectedChange, 0)
+		eligible := enriched
+		for _, validator := range module.Validators() {
+			vResult, vErr := validator.Validate(ctx, contracts.ValidationRequest{
+				WorkingTranscript: working,
+				CandidateProposal: eligible,
+				ModuleKey:         spec.ModuleKey,
+				ModuleInstance:    spec.InstanceName,
+				ReplacementPolicy: policy,
+				Glossary:          input.Glossary,
+				Config:            input.Config,
+			})
+			if vErr != nil {
+				failed := ModuleResult{
+					ModuleKey:          spec.ModuleKey,
+					ModuleInstance:     spec.InstanceName,
+					ReplacementPolicy:  policy,
+					Status:             ModuleStatusFailed,
+					ProposalCount:      len(enriched),
+					ValidatorDecisions: validatorDecisions,
+					ValidatorRejected:  validatorRejected,
+					ErrorMessage:       vErr.Error(),
+					StartedAt:          startedAt,
+					CompletedAt:        time.Now().UTC(),
+				}
+				results = append(results, failed)
+				return RunOutput{FinalTranscript: working, ModuleResults: results}, fmt.Errorf("module %q validator %q failed: %w", spec.InstanceName, validator.Name(), vErr)
+			}
+			if err := validators.EnforceDecisionCardinality(eligible, vResult.Decisions); err != nil {
+				failed := ModuleResult{
+					ModuleKey:          spec.ModuleKey,
+					ModuleInstance:     spec.InstanceName,
+					ReplacementPolicy:  policy,
+					Status:             ModuleStatusFailed,
+					ProposalCount:      len(enriched),
+					ValidatorDecisions: validatorDecisions,
+					ValidatorRejected:  validatorRejected,
+					ErrorMessage:       err.Error(),
+					StartedAt:          startedAt,
+					CompletedAt:        time.Now().UTC(),
+				}
+				results = append(results, failed)
+				return RunOutput{FinalTranscript: working, ModuleResults: results}, fmt.Errorf("module %q validator %q cardinality failed: %w", spec.InstanceName, validator.Name(), err)
+			}
+
+			nextEligible := make([]proposals.EnrichedCorrectionProposal, 0, len(eligible))
+			byIndex := make(map[int]proposals.EnrichedCorrectionProposal, len(eligible))
+			for _, p := range eligible {
+				byIndex[p.ProposalIndex] = p
+			}
+			for _, d := range vResult.Decisions {
+				validatorDecisions = append(validatorDecisions, ValidatorDecisionRecord{
+					ValidatorName: validator.Name(),
+					ProposalIndex: d.ProposalIndex,
+					Approved:      d.Approved,
+					ReasonCode:    d.ReasonCode,
+					Message:       d.Message,
+				})
+				if d.Approved {
+					nextEligible = append(nextEligible, byIndex[d.ProposalIndex])
+					continue
+				}
+				p := byIndex[d.ProposalIndex]
+				validatorRejected = append(validatorRejected, ValidatorRejectedChange{
+					ValidatorName:   validator.Name(),
+					ProposalIndex:   p.ProposalIndex,
+					ModuleKey:       p.ModuleKey,
+					ModuleInstance:  p.ModuleInstance,
+					TargetSegmentID: p.TargetSegmentID,
+					OriginalText:    p.OriginalText,
+					CorrectedText:   p.CorrectedText,
+					ReasonCode:      d.ReasonCode,
+					Message:         d.Message,
+				})
+			}
+			eligible = nextEligible
+		}
+
+		applyResult := proposals.ApplyProposals(working, eligible, policy)
 		working = applyResult.Transcript

 		results = append(results, ModuleResult{
-			ModuleKey:         spec.ModuleKey,
-			ModuleInstance:    spec.InstanceName,
-			ReplacementPolicy: policy,
-			Status:            ModuleStatusSuccess,
-			ProposalCount:     len(enriched),
-			AppliedChanges:    applyResult.Applied,
-			SkippedChanges:    applyResult.Skipped,
-			StartedAt:         startedAt,
-			CompletedAt:       time.Now().UTC(),
+			ModuleKey:          spec.ModuleKey,
+			ModuleInstance:     spec.InstanceName,
+			ReplacementPolicy:  policy,
+			Status:             ModuleStatusSuccess,
+			ProposalCount:      len(enriched),
+			ValidatorDecisions: validatorDecisions,
+			ValidatorRejected:  validatorRejected,
+			AppliedChanges:     applyResult.Applied,
+			SkippedChanges:     applyResult.Skipped,
+			StartedAt:          startedAt,
+			CompletedAt:        time.Now().UTC(),
 		})
 	}

--- a/internal/framework/runner/runner_test.go
+++ b/internal/framework/runner/runner_test.go
@@ -9,6 +9,7 @@ import (
 	"gitea.maximumdirect.net/eric/audita/internal/core/schema"
 	"gitea.maximumdirect.net/eric/audita/internal/framework/contracts"
 	"gitea.maximumdirect.net/eric/audita/internal/framework/proposals"
+	"gitea.maximumdirect.net/eric/audita/internal/framework/validators"
 )

 type fakeFactory struct {
@@ -24,14 +25,15 @@ func (f fakeFactory) ModuleForSpec(spec contracts.ModuleRunSpec) (contracts.Tran
 }

 type fakeModule struct {
-	key      string
-	policy   proposals.ReplacementPolicy
-	proposeF func(req contracts.ProposalRequest) ([]proposals.CorrectionProposal, error)
+	key        string
+	policy     proposals.ReplacementPolicy
+	validators []contracts.Validator
+	proposeF   func(req contracts.ProposalRequest) ([]proposals.CorrectionProposal, error)
 }

 func (m fakeModule) Key() string                                    { return m.key }
 func (m fakeModule) ReplacementPolicy() proposals.ReplacementPolicy { return m.policy }
-func (m fakeModule) Validators() []contracts.Validator              { return nil }
+func (m fakeModule) Validators() []contracts.Validator              { return m.validators }
 func (m fakeModule) Propose(ctx context.Context, req contracts.ProposalRequest) ([]proposals.CorrectionProposal, error) {
 	if m.proposeF == nil {
 		return nil, nil
@@ -39,6 +41,17 @@ func (m fakeModule) Propose(ctx context.Context, req contracts.ProposalRequest)
 	return m.proposeF(req)
 }

+type fakeValidator struct {
+	name      string
+	validateF func(req contracts.ValidationRequest) (validators.Result, error)
+}
+
+func (v fakeValidator) Name() string { return v.name }
+func (v fakeValidator) Validate(ctx context.Context, req contracts.ValidationRequest) (validators.Result, error) {
+	_ = ctx
+	return v.validateF(req)
+}
+
 func TestRunnerOneModuleAppliesProposal(t *testing.T) {
 	transcript := &schema.Transcript{Segments: []schema.Segment{{ID: 1, Speaker: "A", Start: 0, End: 1, Text: "teh cat"}}}
 	r := New(fakeFactory{modules: map[string]contracts.TranscriptModule{
@@ -149,4 +162,148 @@ func TestRunnerRepeatedModuleInstanceNames(t *testing.T) {
 	}
 }

+func TestRunnerValidatorApprovedProposalApplied(t *testing.T) {
+	transcript := &schema.Transcript{Segments: []schema.Segment{{ID: 1, Text: "teh cat"}}}
+	allowAll := fakeValidator{name: "allow", validateF: func(req contracts.ValidationRequest) (validators.Result, error) {
+		decisions := make([]validators.Decision, len(req.CandidateProposal))
+		for i, p := range req.CandidateProposal {
+			decisions[i] = validators.Decision{ProposalIndex: p.ProposalIndex, Approved: true, ReasonCode: validators.ReasonApproved, Message: "approved"}
+		}
+		return validators.Result{ValidatorName: "allow", Decisions: decisions}, nil
+	}}
+	r := New(fakeFactory{modules: map[string]contracts.TranscriptModule{
+		"m": fakeModule{
+			key:        "m",
+			policy:     proposals.ReplacementPolicyRequireUnique,
+			validators: []contracts.Validator{allowAll},
+			proposeF: func(req contracts.ProposalRequest) ([]proposals.CorrectionProposal, error) {
+				return []proposals.CorrectionProposal{{TargetSegmentID: 1, OriginalText: "teh", CorrectedText: "the", Confidence: 1}}, nil
+			},
+		},
+	}})
+
+	out, err := r.Run(context.Background(), RunInput{Transcript: transcript, ModuleSpecs: []contracts.ModuleRunSpec{{ModuleKey: "m", InstanceName: "m"}}})
+	if err != nil {
+		t.Fatalf("Run error: %v", err)
+	}
+	if out.FinalTranscript.Segments[0].Text != "the cat" {
+		t.Fatalf("expected proposal applied, got %q", out.FinalTranscript.Segments[0].Text)
+	}
+	if len(out.ModuleResults[0].ValidatorDecisions) != 1 {
+		t.Fatalf("expected validator decisions recorded")
+	}
+}
+
+func TestRunnerValidatorRejectedProposalNotApplied(t *testing.T) {
+	transcript := &schema.Transcript{Segments: []schema.Segment{{ID: 1, Text: "teh cat"}}}
+	rejectAll := fakeValidator{name: "reject", validateF: func(req contracts.ValidationRequest) (validators.Result, error) {
+		decisions := make([]validators.Decision, len(req.CandidateProposal))
+		for i, p := range req.CandidateProposal {
+			decisions[i] = validators.Decision{ProposalIndex: p.ProposalIndex, Approved: false, ReasonCode: validators.ReasonNoEffect, Message: "rejected"}
+		}
+		return validators.Result{ValidatorName: "reject", Decisions: decisions}, nil
+	}}
+	r := New(fakeFactory{modules: map[string]contracts.TranscriptModule{
+		"m": fakeModule{key: "m", policy: proposals.ReplacementPolicyRequireUnique, validators: []contracts.Validator{rejectAll}, proposeF: func(req contracts.ProposalRequest) ([]proposals.CorrectionProposal, error) {
+			return []proposals.CorrectionProposal{{TargetSegmentID: 1, OriginalText: "teh", CorrectedText: "the", Confidence: 1}}, nil
+		}},
+	}})
+	out, err := r.Run(context.Background(), RunInput{Transcript: transcript, ModuleSpecs: []contracts.ModuleRunSpec{{ModuleKey: "m", InstanceName: "m"}}})
+	if err != nil {
+		t.Fatalf("Run error: %v", err)
+	}
+	if out.FinalTranscript.Segments[0].Text != "teh cat" {
+		t.Fatalf("expected rejected proposal not applied, got %q", out.FinalTranscript.Segments[0].Text)
+	}
+	if len(out.ModuleResults[0].ValidatorRejected) != 1 {
+		t.Fatalf("expected validator rejection recorded, got %+v", out.ModuleResults[0].ValidatorRejected)
+	}
+}
+
+func TestRunnerMultipleValidatorsRunInOrderAndFilterSurvivors(t *testing.T) {
+	transcript := &schema.Transcript{Segments: []schema.Segment{{ID: 1, Text: "one two"}}}
+	first := fakeValidator{name: "first", validateF: func(req contracts.ValidationRequest) (validators.Result, error) {
+		if len(req.CandidateProposal) != 2 {
+			t.Fatalf("expected first validator to see 2 candidates, got %d", len(req.CandidateProposal))
+		}
+		return validators.Result{
+			ValidatorName: "first",
+			Decisions: []validators.Decision{
+				{ProposalIndex: req.CandidateProposal[0].ProposalIndex, Approved: true, ReasonCode: validators.ReasonApproved, Message: "ok"},
+				{ProposalIndex: req.CandidateProposal[1].ProposalIndex, Approved: false, ReasonCode: validators.ReasonNoEffect, Message: "reject"},
+			},
+		}, nil
+	}}
+	second := fakeValidator{name: "second", validateF: func(req contracts.ValidationRequest) (validators.Result, error) {
+		if len(req.CandidateProposal) != 1 {
+			t.Fatalf("expected second validator to see only survivors, got %d", len(req.CandidateProposal))
+		}
+		return validators.Result{ValidatorName: "second", Decisions: []validators.Decision{{ProposalIndex: req.CandidateProposal[0].ProposalIndex, Approved: true, ReasonCode: validators.ReasonApproved, Message: "ok"}}}, nil
+	}}
+	r := New(fakeFactory{modules: map[string]contracts.TranscriptModule{
+		"m": fakeModule{key: "m", policy: proposals.ReplacementPolicyRequireUnique, validators: []contracts.Validator{first, second}, proposeF: func(req contracts.ProposalRequest) ([]proposals.CorrectionProposal, error) {
+			return []proposals.CorrectionProposal{
+				{TargetSegmentID: 1, OriginalText: "one", CorrectedText: "ONE", Confidence: 1},
+				{TargetSegmentID: 1, OriginalText: "two", CorrectedText: "TWO", Confidence: 1},
+			}, nil
+		}},
+	}})
+
+	out, err := r.Run(context.Background(), RunInput{Transcript: transcript, ModuleSpecs: []contracts.ModuleRunSpec{{ModuleKey: "m", InstanceName: "m"}}})
+	if err != nil {
+		t.Fatalf("Run error: %v", err)
+	}
+	if out.FinalTranscript.Segments[0].Text != "ONE two" {
+		t.Fatalf("expected only survivor applied, got %q", out.FinalTranscript.Segments[0].Text)
+	}
+}
+
+func TestRunnerValidatorCardinalityErrorStopsPipelineWithPartialProgress(t *testing.T) {
+	transcript := &schema.Transcript{Segments: []schema.Segment{{ID: 1, Text: "teh cat"}}}
+	good := fakeModule{key: "m1", policy: proposals.ReplacementPolicyRequireUnique, proposeF: func(req contracts.ProposalRequest) ([]proposals.CorrectionProposal, error) {
+		return []proposals.CorrectionProposal{{TargetSegmentID: 1, OriginalText: "teh", CorrectedText: "the", Confidence: 1}}, nil
+	}}
+	badValidator := fakeValidator{name: "bad", validateF: func(req contracts.ValidationRequest) (validators.Result, error) {
+		// Missing one decision triggers cardinality error.
+		return validators.Result{ValidatorName: "bad", Decisions: nil}, nil
+	}}
+	bad := fakeModule{key: "m2", policy: proposals.ReplacementPolicyRequireUnique, validators: []contracts.Validator{badValidator}, proposeF: func(req contracts.ProposalRequest) ([]proposals.CorrectionProposal, error) {
+		return []proposals.CorrectionProposal{{TargetSegmentID: 1, OriginalText: "cat", CorrectedText: "dog", Confidence: 1}}, nil
+	}}
+	r := New(fakeFactory{modules: map[string]contracts.TranscriptModule{"m1": good, "m2": bad}})
+
+	out, err := r.Run(context.Background(), RunInput{Transcript: transcript, ModuleSpecs: []contracts.ModuleRunSpec{{ModuleKey: "m1", InstanceName: "m1"}, {ModuleKey: "m2", InstanceName: "m2"}}})
+	if err == nil {
+		t.Fatal("expected cardinality error")
+	}
+	if out.FinalTranscript.Segments[0].Text != "the cat" {
+		t.Fatalf("expected partial progress preserved, got %q", out.FinalTranscript.Segments[0].Text)
+	}
+	if len(out.ModuleResults) != 2 || out.ModuleResults[1].Status != ModuleStatusFailed {
+		t.Fatalf("expected second module failed")
+	}
+}
+
+func TestRunnerApplicationSkipAfterValidatorApprovalReported(t *testing.T) {
+	transcript := &schema.Transcript{Segments: []schema.Segment{{ID: 1, Text: "word word"}}}
+	allow := fakeValidator{name: "allow", validateF: func(req contracts.ValidationRequest) (validators.Result, error) {
+		return validators.Result{ValidatorName: "allow", Decisions: []validators.Decision{{ProposalIndex: req.CandidateProposal[0].ProposalIndex, Approved: true, ReasonCode: validators.ReasonApproved, Message: "ok"}}}, nil
+	}}
+	r := New(fakeFactory{modules: map[string]contracts.TranscriptModule{
+		"m": fakeModule{key: "m", policy: proposals.ReplacementPolicyRequireUnique, validators: []contracts.Validator{allow}, proposeF: func(req contracts.ProposalRequest) ([]proposals.CorrectionProposal, error) {
+			return []proposals.CorrectionProposal{{TargetSegmentID: 1, OriginalText: "word", CorrectedText: "term", Confidence: 1}}, nil
+		}},
+	}})
+	out, err := r.Run(context.Background(), RunInput{Transcript: transcript, ModuleSpecs: []contracts.ModuleRunSpec{{ModuleKey: "m", InstanceName: "m"}}})
+	if err != nil {
+		t.Fatalf("Run error: %v", err)
+	}
+	if len(out.ModuleResults[0].SkippedChanges) != 1 {
+		t.Fatalf("expected application skip recorded")
+	}
+	if len(out.ModuleResults[0].ValidatorRejected) != 0 {
+		t.Fatalf("expected no validator rejection")
+	}
+}
+
 func ptrConfig(c config.Config) *config.Config { return &c }
--- a/internal/framework/validators/deterministic.go
+++ b/internal/framework/validators/deterministic.go
@@ -0,0 +1,161 @@
+package validators
+
+import (
+	"context"
+	"fmt"
+	"strings"
+
+	"gitea.maximumdirect.net/eric/audita/internal/framework/proposals"
+)
+
+type ConfidenceThresholdValidator struct{}
+
+func (v ConfidenceThresholdValidator) Name() string { return "confidence_threshold" }
+
+func (v ConfidenceThresholdValidator) Validate(_ context.Context, req Request) (Result, error) {
+	threshold := confidenceThresholdForModule(req.ModuleKey, req.Config)
+	decisions := make([]Decision, 0, len(req.CandidateProposal))
+	for _, c := range req.CandidateProposal {
+		if c.Confidence < threshold {
+			decisions = append(decisions, rejection(c.ProposalIndex, ReasonLowConfidence, fmt.Sprintf("confidence %.4f below threshold %.4f", c.Confidence, threshold)))
+			continue
+		}
+		decisions = append(decisions, approval(c.ProposalIndex))
+	}
+	if err := EnforceDecisionCardinality(req.CandidateProposal, decisions); err != nil {
+		return Result{}, err
+	}
+	return Result{ValidatorName: v.Name(), Decisions: decisions}, nil
+}
+
+type OriginalTextPresenceValidator struct{}
+
+func (v OriginalTextPresenceValidator) Name() string { return "original_text_presence" }
+
+func (v OriginalTextPresenceValidator) Validate(_ context.Context, req Request) (Result, error) {
+	byID := make(map[int]string)
+	if req.WorkingTranscript != nil {
+		for _, seg := range req.WorkingTranscript.Segments {
+			byID[seg.ID] = seg.Text
+		}
+	}
+
+	decisions := make([]Decision, 0, len(req.CandidateProposal))
+	for _, c := range req.CandidateProposal {
+		text, ok := byID[c.TargetSegmentID]
+		if !ok {
+			decisions = append(decisions, rejection(c.ProposalIndex, ReasonMissingTargetSegment, "target segment was not found"))
+			continue
+		}
+		if !strings.Contains(text, c.OriginalText) {
+			decisions = append(decisions, rejection(c.ProposalIndex, ReasonMissingOriginalText, "original_text was not found in current segment text"))
+			continue
+		}
+		decisions = append(decisions, approval(c.ProposalIndex))
+	}
+	if err := EnforceDecisionCardinality(req.CandidateProposal, decisions); err != nil {
+		return Result{}, err
+	}
+	return Result{ValidatorName: v.Name(), Decisions: decisions}, nil
+}
+
+type NonEmptyCorrectionValidator struct{}
+
+func (v NonEmptyCorrectionValidator) Name() string { return "non_empty_correction" }
+
+func (v NonEmptyCorrectionValidator) Validate(_ context.Context, req Request) (Result, error) {
+	decisions := make([]Decision, 0, len(req.CandidateProposal))
+	for _, c := range req.CandidateProposal {
+		if strings.TrimSpace(c.CorrectedText) == "" {
+			decisions = append(decisions, rejection(c.ProposalIndex, ReasonEmptyCorrectedText, "corrected_text must not be empty"))
+			continue
+		}
+		decisions = append(decisions, approval(c.ProposalIndex))
+	}
+	if err := EnforceDecisionCardinality(req.CandidateProposal, decisions); err != nil {
+		return Result{}, err
+	}
+	return Result{ValidatorName: v.Name(), Decisions: decisions}, nil
+}
+
+type NoEffectValidator struct{}
+
+func (v NoEffectValidator) Name() string { return "no_effect" }
+
+func (v NoEffectValidator) Validate(_ context.Context, req Request) (Result, error) {
+	decisions := make([]Decision, 0, len(req.CandidateProposal))
+	for _, c := range req.CandidateProposal {
+		if c.OriginalText == c.CorrectedText {
+			decisions = append(decisions, rejection(c.ProposalIndex, ReasonNoEffect, "original_text and corrected_text are identical"))
+			continue
+		}
+		decisions = append(decisions, approval(c.ProposalIndex))
+	}
+	if err := EnforceDecisionCardinality(req.CandidateProposal, decisions); err != nil {
+		return Result{}, err
+	}
+	return Result{ValidatorName: v.Name(), Decisions: decisions}, nil
+}
+
+type ProtectedGlossaryTermValidator struct{}
+
+func (v ProtectedGlossaryTermValidator) Name() string { return "protected_glossary_terms" }
+
+func (v ProtectedGlossaryTermValidator) Validate(_ context.Context, req Request) (Result, error) {
+	if req.ModuleKey == "glossary" {
+		decisions := make([]Decision, 0, len(req.CandidateProposal))
+		for _, c := range req.CandidateProposal {
+			decisions = append(decisions, approval(c.ProposalIndex))
+		}
+		return Result{ValidatorName: v.Name(), Decisions: decisions}, nil
+	}
+
+	terms := glossaryTerms(req)
+	decisions := make([]Decision, 0, len(req.CandidateProposal))
+	for _, c := range req.CandidateProposal {
+		if altersProtectedTerm(c.CorrectionProposal, terms) {
+			decisions = append(decisions, rejection(c.ProposalIndex, ReasonProtectedGlossaryTerm, "proposal may alter protected glossary terminology"))
+			continue
+		}
+		decisions = append(decisions, approval(c.ProposalIndex))
+	}
+	if err := EnforceDecisionCardinality(req.CandidateProposal, decisions); err != nil {
+		return Result{}, err
+	}
+	return Result{ValidatorName: v.Name(), Decisions: decisions}, nil
+}
+
+func glossaryTerms(req Request) []string {
+	if req.Glossary == nil {
+		return nil
+	}
+	out := make([]string, 0)
+	for _, e := range req.Glossary.Entries {
+		if t := strings.TrimSpace(strings.ToLower(e.Name)); t != "" {
+			out = append(out, t)
+		}
+		for _, a := range e.Aliases {
+			if t := strings.TrimSpace(strings.ToLower(a)); t != "" {
+				out = append(out, t)
+			}
+		}
+		if t := strings.TrimSpace(strings.ToLower(e.Plural)); t != "" {
+			out = append(out, t)
+		}
+	}
+	return out
+}
+
+func altersProtectedTerm(p proposals.CorrectionProposal, terms []string) bool {
+	if len(terms) == 0 {
+		return false
+	}
+	orig := strings.ToLower(p.OriginalText)
+	corr := strings.ToLower(p.CorrectedText)
+	for _, t := range terms {
+		if strings.Contains(orig, t) && !strings.Contains(corr, t) {
+			return true
+		}
+	}
+	return false
+}
--- a/internal/framework/validators/models.go
+++ b/internal/framework/validators/models.go
@@ -0,0 +1,105 @@
+package validators
+
+import (
+	"context"
+	"fmt"
+	"strings"
+
+	"gitea.maximumdirect.net/eric/audita/internal/core/config"
+	"gitea.maximumdirect.net/eric/audita/internal/core/schema"
+	"gitea.maximumdirect.net/eric/audita/internal/framework/proposals"
+)
+
+const (
+	ReasonApproved              = "approved"
+	ReasonLowConfidence         = "low_confidence"
+	ReasonMissingOriginalText   = "missing_original_text"
+	ReasonMissingTargetSegment  = "missing_target_segment"
+	ReasonEmptyCorrectedText    = "empty_corrected_text"
+	ReasonNoEffect              = "no_effect"
+	ReasonProtectedGlossaryTerm = "protected_glossary_term"
+)
+
+// Request is the runtime input shared by deterministic validators.
+type Request struct {
+	WorkingTranscript *schema.Transcript                     `json:"-"`
+	CandidateProposal []proposals.EnrichedCorrectionProposal `json:"candidate_proposals"`
+	ModuleKey         string                                 `json:"module_key"`
+	ModuleInstance    string                                 `json:"module_instance"`
+	ReplacementPolicy proposals.ReplacementPolicy            `json:"replacement_policy"`
+	Glossary          *schema.Glossary                       `json:"-"`
+	Config            *config.Config                         `json:"-"`
+}
+
+// Decision is one validator decision for one proposal index.
+type Decision struct {
+	ProposalIndex int    `json:"proposal_index"`
+	Approved      bool   `json:"approved"`
+	ReasonCode    string `json:"reason_code"`
+	Message       string `json:"message"`
+}
+
+// Result is one validator output containing exactly one decision per proposal index.
+type Result struct {
+	ValidatorName string     `json:"validator_name"`
+	Decisions     []Decision `json:"decisions"`
+}
+
+// Validator is the deterministic runtime validator interface.
+type Validator interface {
+	Name() string
+	Validate(ctx context.Context, req Request) (Result, error)
+}
+
+// EnforceDecisionCardinality verifies every candidate proposal index receives exactly one decision.
+func EnforceDecisionCardinality(candidate []proposals.EnrichedCorrectionProposal, decisions []Decision) error {
+	expected := make(map[int]struct{}, len(candidate))
+	for _, c := range candidate {
+		expected[c.ProposalIndex] = struct{}{}
+	}
+
+	seen := make(map[int]int, len(decisions))
+	for _, d := range decisions {
+		if _, ok := expected[d.ProposalIndex]; !ok {
+			return fmt.Errorf("unknown decision proposal index %d", d.ProposalIndex)
+		}
+		seen[d.ProposalIndex]++
+		if seen[d.ProposalIndex] > 1 {
+			return fmt.Errorf("duplicate decision proposal index %d", d.ProposalIndex)
+		}
+	}
+
+	for idx := range expected {
+		if seen[idx] == 0 {
+			return fmt.Errorf("missing decision proposal index %d", idx)
+		}
+	}
+
+	return nil
+}
+
+func approval(index int) Decision {
+	return Decision{ProposalIndex: index, Approved: true, ReasonCode: ReasonApproved, Message: "approved"}
+}
+
+func rejection(index int, reasonCode string, msg string) Decision {
+	return Decision{ProposalIndex: index, Approved: false, ReasonCode: reasonCode, Message: strings.TrimSpace(msg)}
+}
+
+func confidenceThresholdForModule(moduleKey string, cfg *config.Config) float64 {
+	if cfg == nil {
+		return 0.0
+	}
+	switch moduleKey {
+	case "glossary":
+		return cfg.Thresholds.Glossary
+	case "grammar":
+		return cfg.Thresholds.Grammar
+	case "homophones":
+		return cfg.Thresholds.Homophones
+	case "spoken_word":
+		return cfg.Thresholds.SpokenWord
+	default:
+		return 0.0
+	}
+}
--- a/internal/framework/validators/validators_test.go
+++ b/internal/framework/validators/validators_test.go
@@ -0,0 +1,181 @@
+package validators
+
+import (
+	"context"
+	"strings"
+	"testing"
+
+	"gitea.maximumdirect.net/eric/audita/internal/core/config"
+	"gitea.maximumdirect.net/eric/audita/internal/core/schema"
+	"gitea.maximumdirect.net/eric/audita/internal/framework/proposals"
+)
+
+func mkCandidate(index int, segID int, orig, corr string, conf float64) proposals.EnrichedCorrectionProposal {
+	return proposals.EnrichedCorrectionProposal{
+		CorrectionProposal: proposals.CorrectionProposal{TargetSegmentID: segID, OriginalText: orig, CorrectedText: corr, Confidence: conf},
+		ProposalMetadata:   proposals.ProposalMetadata{ProposalIndex: index, ModuleKey: "grammar", ModuleInstance: "grammar_1"},
+	}
+}
+
+func TestEnforceDecisionCardinalitySuccess(t *testing.T) {
+	candidates := []proposals.EnrichedCorrectionProposal{mkCandidate(0, 1, "teh", "the", 0.9), mkCandidate(1, 1, "recieve", "receive", 0.9)}
+	decisions := []Decision{{ProposalIndex: 0, Approved: true}, {ProposalIndex: 1, Approved: false}}
+	if err := EnforceDecisionCardinality(candidates, decisions); err != nil {
+		t.Fatalf("unexpected error: %v", err)
+	}
+}
+
+func TestEnforceDecisionCardinalityMissingDecision(t *testing.T) {
+	candidates := []proposals.EnrichedCorrectionProposal{mkCandidate(0, 1, "teh", "the", 0.9), mkCandidate(1, 1, "recieve", "receive", 0.9)}
+	err := EnforceDecisionCardinality(candidates, []Decision{{ProposalIndex: 0, Approved: true}})
+	if err == nil || !strings.Contains(err.Error(), "missing decision") {
+		t.Fatalf("expected missing decision error, got %v", err)
+	}
+}
+
+func TestEnforceDecisionCardinalityDuplicateDecision(t *testing.T) {
+	candidates := []proposals.EnrichedCorrectionProposal{mkCandidate(0, 1, "teh", "the", 0.9)}
+	err := EnforceDecisionCardinality(candidates, []Decision{{ProposalIndex: 0, Approved: true}, {ProposalIndex: 0, Approved: false}})
+	if err == nil || !strings.Contains(err.Error(), "duplicate decision") {
+		t.Fatalf("expected duplicate decision error, got %v", err)
+	}
+}
+
+func TestEnforceDecisionCardinalityUnknownDecision(t *testing.T) {
+	candidates := []proposals.EnrichedCorrectionProposal{mkCandidate(0, 1, "teh", "the", 0.9)}
+	err := EnforceDecisionCardinality(candidates, []Decision{{ProposalIndex: 99, Approved: true}})
+	if err == nil || !strings.Contains(err.Error(), "unknown decision") {
+		t.Fatalf("expected unknown decision error, got %v", err)
+	}
+}
+
+func TestConfidenceThresholdValidator(t *testing.T) {
+	cfg := config.Default()
+	cfg.Thresholds.Grammar = 0.8
+	req := Request{ModuleKey: "grammar", Config: &cfg, CandidateProposal: []proposals.EnrichedCorrectionProposal{
+		mkCandidate(0, 1, "teh", "the", 0.9),
+		mkCandidate(1, 1, "recieve", "receive", 0.7),
+	}}
+	res, err := (ConfidenceThresholdValidator{}).Validate(context.Background(), req)
+	if err != nil {
+		t.Fatalf("Validate error: %v", err)
+	}
+	if !res.Decisions[0].Approved || res.Decisions[0].ReasonCode != ReasonApproved {
+		t.Fatalf("expected first decision approved, got %+v", res.Decisions[0])
+	}
+	if res.Decisions[1].Approved || res.Decisions[1].ReasonCode != ReasonLowConfidence {
+		t.Fatalf("expected second decision low confidence reject, got %+v", res.Decisions[1])
+	}
+}
+
+func TestOriginalTextPresenceValidator(t *testing.T) {
+	req := Request{WorkingTranscript: &schema.Transcript{Segments: []schema.Segment{{ID: 1, Text: "hello world"}}}, CandidateProposal: []proposals.EnrichedCorrectionProposal{
+		mkCandidate(0, 1, "hello", "hi", 0.9),
+		mkCandidate(1, 1, "missing", "x", 0.9),
+		mkCandidate(2, 5, "hello", "hi", 0.9),
+	}}
+	res, err := (OriginalTextPresenceValidator{}).Validate(context.Background(), req)
+	if err != nil {
+		t.Fatalf("Validate error: %v", err)
+	}
+	if !res.Decisions[0].Approved {
+		t.Fatalf("expected proposal 0 approved")
+	}
+	if res.Decisions[1].ReasonCode != ReasonMissingOriginalText {
+		t.Fatalf("expected missing_original_text, got %+v", res.Decisions[1])
+	}
+	if res.Decisions[2].ReasonCode != ReasonMissingTargetSegment {
+		t.Fatalf("expected missing_target_segment, got %+v", res.Decisions[2])
+	}
+}
+
+func TestNonEmptyCorrectionValidator(t *testing.T) {
+	req := Request{CandidateProposal: []proposals.EnrichedCorrectionProposal{
+		mkCandidate(0, 1, "hello", "hi", 0.9),
+		mkCandidate(1, 1, "hello", "   ", 0.9),
+	}}
+	res, err := (NonEmptyCorrectionValidator{}).Validate(context.Background(), req)
+	if err != nil {
+		t.Fatalf("Validate error: %v", err)
+	}
+	if !res.Decisions[0].Approved {
+		t.Fatalf("expected proposal 0 approved")
+	}
+	if res.Decisions[1].ReasonCode != ReasonEmptyCorrectedText {
+		t.Fatalf("expected empty_corrected_text, got %+v", res.Decisions[1])
+	}
+}
+
+func TestNoEffectValidator(t *testing.T) {
+	req := Request{CandidateProposal: []proposals.EnrichedCorrectionProposal{
+		mkCandidate(0, 1, "hello", "hello", 0.9),
+		mkCandidate(1, 1, "hello", "hi", 0.9),
+	}}
+	res, err := (NoEffectValidator{}).Validate(context.Background(), req)
+	if err != nil {
+		t.Fatalf("Validate error: %v", err)
+	}
+	if res.Decisions[0].ReasonCode != ReasonNoEffect || res.Decisions[0].Approved {
+		t.Fatalf("expected no_effect rejection, got %+v", res.Decisions[0])
+	}
+	if !res.Decisions[1].Approved {
+		t.Fatalf("expected proposal 1 approved")
+	}
+}
+
+func TestProtectedGlossaryTermValidator(t *testing.T) {
+	glossary := &schema.Glossary{Entries: []schema.GlossaryEntry{{Name: "OpenAI", Aliases: []string{"Open AI"}, Plural: "OpenAIs", Category: "brand", Summary: "brand"}}}
+	req := Request{
+		Glossary:  glossary,
+		ModuleKey: "grammar",
+		CandidateProposal: []proposals.EnrichedCorrectionProposal{
+			mkCandidate(0, 1, "OpenAI", "Open A I", 0.9),
+			mkCandidate(1, 1, "teh", "the", 0.9),
+		},
+	}
+	res, err := (ProtectedGlossaryTermValidator{}).Validate(context.Background(), req)
+	if err != nil {
+		t.Fatalf("Validate error: %v", err)
+	}
+	if res.Decisions[0].ReasonCode != ReasonProtectedGlossaryTerm || res.Decisions[0].Approved {
+		t.Fatalf("expected protected glossary rejection, got %+v", res.Decisions[0])
+	}
+	if !res.Decisions[1].Approved {
+		t.Fatalf("expected non-glossary proposal approved")
+	}
+}
+
+func TestProtectedGlossaryTermValidatorAllowsGlossaryModule(t *testing.T) {
+	glossary := &schema.Glossary{Entries: []schema.GlossaryEntry{{Name: "OpenAI", Category: "brand", Summary: "brand"}}}
+	req := Request{
+		Glossary:  glossary,
+		ModuleKey: "glossary",
+		CandidateProposal: []proposals.EnrichedCorrectionProposal{
+			mkCandidate(0, 1, "OpenAI", "Open A I", 0.9),
+		},
+	}
+	res, err := (ProtectedGlossaryTermValidator{}).Validate(context.Background(), req)
+	if err != nil {
+		t.Fatalf("Validate error: %v", err)
+	}
+	if !res.Decisions[0].Approved {
+		t.Fatalf("expected glossary module approval, got %+v", res.Decisions[0])
+	}
+}
+
+func TestStableReasonCodes(t *testing.T) {
+	codes := []string{
+		ReasonApproved,
+		ReasonLowConfidence,
+		ReasonMissingOriginalText,
+		ReasonMissingTargetSegment,
+		ReasonEmptyCorrectedText,
+		ReasonNoEffect,
+		ReasonProtectedGlossaryTerm,
+	}
+	for _, code := range codes {
+		if strings.TrimSpace(code) == "" {
+			t.Fatalf("reason code must not be empty")
+		}
+	}
+}