Fix the five resilience review findings
This commit is contained in:
@@ -61,11 +61,13 @@ func SectionMetadataFromSection(section chunking.Section) SectionMetadata {
|
||||
|
||||
// ExecutionContext carries shared execution state for proposal generation and validation.
|
||||
type ExecutionContext struct {
|
||||
Config *config.Config `json:"-"`
|
||||
WorkingTranscript *schema.Transcript `json:"-"`
|
||||
Glossary *schema.Glossary `json:"-"`
|
||||
Section *SectionMetadata `json:"section,omitempty"`
|
||||
DiagnosticsDir string `json:"diagnostics_dir,omitempty"`
|
||||
// ReportDiagnosticError may be called concurrently; it must not affect processing.
|
||||
ReportDiagnosticError func(error) `json:"-"`
|
||||
Config *config.Config `json:"-"`
|
||||
WorkingTranscript *schema.Transcript `json:"-"`
|
||||
Glossary *schema.Glossary `json:"-"`
|
||||
Section *SectionMetadata `json:"section,omitempty"`
|
||||
DiagnosticsDir string `json:"diagnostics_dir,omitempty"`
|
||||
}
|
||||
|
||||
// ProposalRequest is the input to module proposal generation.
|
||||
|
||||
99
internal/framework/processreport/partial_validation_test.go
Normal file
99
internal/framework/processreport/partial_validation_test.go
Normal file
@@ -0,0 +1,99 @@
|
||||
package processreport
|
||||
|
||||
import (
|
||||
"context"
|
||||
"errors"
|
||||
"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/contracts"
|
||||
"gitea.maximumdirect.net/eric/audita/internal/framework/outcomes"
|
||||
"gitea.maximumdirect.net/eric/audita/internal/framework/proposals"
|
||||
"gitea.maximumdirect.net/eric/audita/internal/framework/runner"
|
||||
"gitea.maximumdirect.net/eric/audita/internal/framework/structuredoutput"
|
||||
"gitea.maximumdirect.net/eric/audita/internal/framework/validators"
|
||||
)
|
||||
|
||||
type partialModule struct{ validator contracts.Validator }
|
||||
|
||||
func (m partialModule) Key() string { return "grammar" }
|
||||
func (m partialModule) ReplacementPolicy() proposals.ReplacementPolicy {
|
||||
return proposals.ReplacementPolicyRequireUnique
|
||||
}
|
||||
func (m partialModule) Validators() []contracts.Validator { return []contracts.Validator{m.validator} }
|
||||
func (m partialModule) ModuleForSpec(contracts.ModuleRunSpec) (contracts.TranscriptModule, error) {
|
||||
return m, nil
|
||||
}
|
||||
func (m partialModule) Propose(context.Context, contracts.ProposalRequest) (contracts.ProposalResult, error) {
|
||||
result := contracts.ProposalResult{}
|
||||
for _, text := range []string{"Hello", "Greetings", "Hi"} {
|
||||
result.Proposals = append(result.Proposals, proposals.CorrectionProposal{TargetSegmentID: 1, OriginalText: "hello", CorrectedText: text, Confidence: 1})
|
||||
}
|
||||
return result, nil
|
||||
}
|
||||
|
||||
type oneItemEstimator struct{}
|
||||
|
||||
func (oneItemEstimator) EstimateTokens(payload string) int {
|
||||
return 100 * strings.Count(payload, `"correction_index"`)
|
||||
}
|
||||
|
||||
type interruptedValidationClient struct {
|
||||
cause error
|
||||
calls int
|
||||
}
|
||||
|
||||
func (c *interruptedValidationClient) CompleteStructured(_ context.Context, req structuredoutput.Request, target any) (structuredoutput.Response, error) {
|
||||
c.calls++
|
||||
if *req.Context.BatchIndex == 2 {
|
||||
return structuredoutput.Response{}, &structuredoutput.Failure{Class: structuredoutput.FailureFatal, Reason: outcomes.ReasonSetupFailure, Message: "validation interrupted", Cause: c.cause}
|
||||
}
|
||||
index := *req.Context.BatchIndex
|
||||
*target.(*validators.LLMValidationResponse) = validators.LLMValidationResponse{Validations: []validators.LLMValidationDecision{{CorrectionIndex: index, Approved: index == 1, Confidence: 1, Reason: "completed assessment"}}}
|
||||
return structuredoutput.Response{}, nil
|
||||
}
|
||||
|
||||
func TestFatalBatchRetainsEarlierDecisionsInLedger(t *testing.T) {
|
||||
validator, err := validators.NewLLMBackedValidator("editorial_review", validators.LLMValidatorTypeEditorialReview)
|
||||
if err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
validator.SetTokenEstimator(oneItemEstimator{})
|
||||
origin := errors.New("later batch failed")
|
||||
client := &interruptedValidationClient{cause: origin}
|
||||
cfg := config.Config{Chunking: config.Chunking{MaxSectionTokens: 8192, MinSectionTokens: 1}, Validation: config.Validation{MaxPromptTokens: 150}}
|
||||
result, err := runner.New(partialModule{validator}).Run(context.Background(), runner.RunInput{
|
||||
Config: &cfg, Transcript: &schema.Transcript{Segments: []schema.Segment{{ID: 1, Text: "hello"}}},
|
||||
ModuleSpecs: []contracts.ModuleRunSpec{{ModuleKey: "grammar", InstanceName: "grammar"}}, ValidationLLMClient: client, CaptureCandidates: true,
|
||||
})
|
||||
if !errors.Is(err, origin) || client.calls != 3 {
|
||||
t.Fatalf("fatal cause/calls: %v %d", err, client.calls)
|
||||
}
|
||||
if result.FinalTranscript.Segments[0].Text != "hello" || len(result.ModuleResults[0].AppliedChanges) != 0 {
|
||||
t.Fatal("fatal partial approvals changed transcript")
|
||||
}
|
||||
if len(result.WorkUnits[0].Operations) != 3 || result.WorkUnits[0].State != outcomes.WorkInterrupted {
|
||||
t.Fatalf("partial facts: %+v", result.WorkUnits)
|
||||
}
|
||||
ledger := BuildCorrectionLedger(CorrectionLedgerInput{RunOutput: &result})
|
||||
candidates := make(map[int]CorrectionLedgerEntry)
|
||||
for _, entry := range ledger {
|
||||
if entry.Disposition != correctionDispositionFailed {
|
||||
candidates[entry.ProposalIndex] = entry
|
||||
}
|
||||
}
|
||||
if len(candidates) != 3 {
|
||||
t.Fatalf("candidate evidence: %+v", ledger)
|
||||
}
|
||||
for index := 0; index < 2; index++ {
|
||||
entry := candidates[index]
|
||||
if len(entry.LLMValidatorResults) != 1 || entry.LLMValidatorResults[0].Message != "completed assessment" || entry.LLMValidatorResults[0].Approved != (index == 1) {
|
||||
t.Errorf("lost batch %d decision: %+v", index, entry)
|
||||
}
|
||||
}
|
||||
if candidates[0].Disposition != correctionDispositionRejected || candidates[1].Disposition != correctionDispositionInterrupted || candidates[2].Disposition != correctionDispositionInterrupted {
|
||||
t.Fatalf("wrong partial dispositions: %+v", ledger)
|
||||
}
|
||||
}
|
||||
@@ -50,21 +50,23 @@ type StructuredCorrectionSet struct {
|
||||
|
||||
// Request captures reusable proposal-generation inputs for future modules.
|
||||
type Request struct {
|
||||
ModuleKey string `json:"module_key"`
|
||||
ModuleInstance string `json:"module_instance"`
|
||||
ReplacementPolicy proposals.ReplacementPolicy `json:"replacement_policy"`
|
||||
WorkingTranscript *schema.Transcript `json:"-"`
|
||||
Section *contracts.SectionMetadata `json:"section,omitempty"`
|
||||
Glossary *schema.Glossary `json:"-"`
|
||||
Config *config.Config `json:"-"`
|
||||
PromptID string `json:"prompt_id"`
|
||||
PromptVersion string `json:"prompt_version"`
|
||||
Variables map[string]string `json:"variables"`
|
||||
StageName string `json:"stage_name,omitempty"`
|
||||
StartIndex int `json:"start_index"`
|
||||
LLMClient contracts.StructuredLLMClient
|
||||
DiagnosticsDir string
|
||||
DiagnosticsWriter InteractionDiagnosticsWriter
|
||||
// ReportDiagnosticError may be called concurrently; it must not affect processing.
|
||||
ReportDiagnosticError func(error) `json:"-"`
|
||||
ModuleKey string `json:"module_key"`
|
||||
ModuleInstance string `json:"module_instance"`
|
||||
ReplacementPolicy proposals.ReplacementPolicy `json:"replacement_policy"`
|
||||
WorkingTranscript *schema.Transcript `json:"-"`
|
||||
Section *contracts.SectionMetadata `json:"section,omitempty"`
|
||||
Glossary *schema.Glossary `json:"-"`
|
||||
Config *config.Config `json:"-"`
|
||||
PromptID string `json:"prompt_id"`
|
||||
PromptVersion string `json:"prompt_version"`
|
||||
Variables map[string]string `json:"variables"`
|
||||
StageName string `json:"stage_name,omitempty"`
|
||||
StartIndex int `json:"start_index"`
|
||||
LLMClient contracts.StructuredLLMClient
|
||||
DiagnosticsDir string
|
||||
DiagnosticsWriter InteractionDiagnosticsWriter
|
||||
}
|
||||
|
||||
// Result contains generated candidate proposals and optional diagnostics paths.
|
||||
@@ -145,13 +147,17 @@ func GenerateCandidates(ctx context.Context, req Request) (Result, error) {
|
||||
requestMetadata["response_schema"] = responseschema.MustLookup(responseschema.CorrectionSetKey).DiagnosticsMap()
|
||||
|
||||
if writer != nil {
|
||||
artifacts, _ = writer.WriteInteraction(
|
||||
var diagnosticErr error
|
||||
artifacts, diagnosticErr = writer.WriteInteraction(
|
||||
stage,
|
||||
requestMetadata,
|
||||
completion.Prepared,
|
||||
map[string]any{"decoded": response, "final_content": completion.FinalContent, "model": completion.Model},
|
||||
errPayload(callErr),
|
||||
)
|
||||
if diagnosticErr != nil && req.ReportDiagnosticError != nil {
|
||||
req.ReportDiagnosticError(fmt.Errorf("proposal interaction %s: %w", stage, diagnosticErr))
|
||||
}
|
||||
}
|
||||
|
||||
if callErr != nil {
|
||||
|
||||
@@ -51,6 +51,7 @@ func (f *fakeStructuredClient) CompleteStructured(ctx context.Context, req contr
|
||||
}
|
||||
|
||||
type captureDiagnosticsWriter struct {
|
||||
err error
|
||||
lastStage string
|
||||
lastRequestMetadata any
|
||||
lastRequestPayload any
|
||||
@@ -62,7 +63,7 @@ func (w *captureDiagnosticsWriter) WriteInteraction(stage string, requestMetadat
|
||||
w.lastRequestPayload = requestPayload
|
||||
_ = responsePayload
|
||||
_ = errorPayload
|
||||
return InteractionArtifacts{}, nil
|
||||
return InteractionArtifacts{}, w.err
|
||||
}
|
||||
|
||||
func defaultRequest(t *testing.T) Request {
|
||||
@@ -383,3 +384,16 @@ func TestGenerateCandidatesClientError(t *testing.T) {
|
||||
t.Fatalf("expected error diagnostics artifact under %s", req.DiagnosticsDir)
|
||||
}
|
||||
}
|
||||
|
||||
func TestProposalDiagnosticFailureDoesNotChangeOperation(t *testing.T) {
|
||||
cause := errors.New("diagnostic storage unavailable")
|
||||
req := defaultRequest(t)
|
||||
req.LLMClient = &fakeStructuredClient{responses: []StructuredCorrectionSet{{}}}
|
||||
req.DiagnosticsWriter = &captureDiagnosticsWriter{err: cause}
|
||||
var reported error
|
||||
req.ReportDiagnosticError = func(err error) { reported = err }
|
||||
result, err := GenerateCandidates(context.Background(), req)
|
||||
if err != nil || !errors.Is(reported, cause) || len(result.Operations) != 1 || !result.Operations[0].Succeeded || len(result.Operations[0].RecoveryReasons) != 0 || result.Artifacts.ResponsePayloadPath != "" {
|
||||
t.Fatalf("debug failure changed outcome: %+v %v reported=%v", result, err, reported)
|
||||
}
|
||||
}
|
||||
|
||||
@@ -55,20 +55,21 @@ func ExecuteModuleProposal(ctx context.Context, req ModuleProposalRequest) (cont
|
||||
}
|
||||
|
||||
generated, err := GenerateCandidates(ctx, Request{
|
||||
ModuleKey: req.ProposalRequest.RunSpec.ModuleKey,
|
||||
ModuleInstance: req.ProposalRequest.RunSpec.InstanceName,
|
||||
ReplacementPolicy: req.ProposalRequest.RunSpec.ReplacementPolicy,
|
||||
WorkingTranscript: req.ProposalRequest.WorkingTranscript,
|
||||
Section: req.ProposalRequest.Section,
|
||||
Glossary: req.ProposalRequest.Glossary,
|
||||
Config: req.ProposalRequest.Config,
|
||||
PromptID: promptMetadata.PromptID,
|
||||
PromptVersion: promptMetadata.PromptVersion,
|
||||
Variables: variables,
|
||||
StageName: stagename.ModuleProposal(req.ProposalRequest.RunSpec.InstanceName, sectionIndexPtr(req.ProposalRequest.Section)),
|
||||
StartIndex: 0,
|
||||
LLMClient: req.ProposalRequest.LLMClient,
|
||||
DiagnosticsDir: req.ProposalRequest.DiagnosticsDir,
|
||||
ModuleKey: req.ProposalRequest.RunSpec.ModuleKey,
|
||||
ModuleInstance: req.ProposalRequest.RunSpec.InstanceName,
|
||||
ReplacementPolicy: req.ProposalRequest.RunSpec.ReplacementPolicy,
|
||||
WorkingTranscript: req.ProposalRequest.WorkingTranscript,
|
||||
Section: req.ProposalRequest.Section,
|
||||
Glossary: req.ProposalRequest.Glossary,
|
||||
Config: req.ProposalRequest.Config,
|
||||
PromptID: promptMetadata.PromptID,
|
||||
PromptVersion: promptMetadata.PromptVersion,
|
||||
Variables: variables,
|
||||
StageName: stagename.ModuleProposal(req.ProposalRequest.RunSpec.InstanceName, sectionIndexPtr(req.ProposalRequest.Section)),
|
||||
StartIndex: 0,
|
||||
LLMClient: req.ProposalRequest.LLMClient,
|
||||
DiagnosticsDir: req.ProposalRequest.DiagnosticsDir,
|
||||
ReportDiagnosticError: req.ProposalRequest.ReportDiagnosticError,
|
||||
})
|
||||
if err != nil {
|
||||
return contracts.ProposalResult{Operations: generated.Operations}, err
|
||||
|
||||
@@ -2,6 +2,7 @@ package runner
|
||||
|
||||
import (
|
||||
"context"
|
||||
"errors"
|
||||
"fmt"
|
||||
"path/filepath"
|
||||
"sort"
|
||||
@@ -86,6 +87,8 @@ type ValidatorRejectedChange struct {
|
||||
|
||||
// RunInput is the deterministic runner input.
|
||||
type RunInput struct {
|
||||
// ReportDiagnosticError may be called concurrently; it must not affect processing.
|
||||
ReportDiagnosticError func(error) `json:"-"`
|
||||
Config *config.Config
|
||||
Transcript *schema.Transcript
|
||||
Glossary *schema.Glossary
|
||||
@@ -188,6 +191,7 @@ func (r *Runner) run(ctx context.Context, input RunInput) (RunOutput, error) {
|
||||
ValidationClient: input.ValidationLLMClient,
|
||||
ValidationDiagnosticsDir: input.ValidationDiagnosticsDir,
|
||||
Timings: input.timings,
|
||||
ReportDiagnosticError: input.ReportDiagnosticError,
|
||||
})
|
||||
if input.CaptureCandidates {
|
||||
candidates = append(candidates, pipelineResult.Candidates...)
|
||||
@@ -293,6 +297,8 @@ func applyWithinSections(sections []chunking.Section, approved []proposals.Enric
|
||||
}
|
||||
|
||||
type collectSectionProposalsInput struct {
|
||||
// ReportDiagnosticError may be called concurrently; it must not affect processing.
|
||||
ReportDiagnosticError func(error) `json:"-"`
|
||||
Module contracts.TranscriptModule
|
||||
Spec contracts.ModuleRunSpec
|
||||
Policy proposals.ReplacementPolicy
|
||||
@@ -356,11 +362,12 @@ func collectSectionProposals(ctx context.Context, input collectSectionProposalsI
|
||||
meta := contracts.SectionMetadataFromSection(section)
|
||||
proposalResult, err := input.Module.Propose(runCtx, contracts.ProposalRequest{
|
||||
ExecutionContext: contracts.ExecutionContext{
|
||||
Config: input.Config,
|
||||
WorkingTranscript: transcriptFromSection(section),
|
||||
Glossary: input.Glossary,
|
||||
Section: &meta,
|
||||
DiagnosticsDir: input.DiagnosticsDir,
|
||||
Config: input.Config,
|
||||
WorkingTranscript: transcriptFromSection(section),
|
||||
Glossary: input.Glossary,
|
||||
Section: &meta,
|
||||
DiagnosticsDir: input.DiagnosticsDir,
|
||||
ReportDiagnosticError: input.ReportDiagnosticError,
|
||||
},
|
||||
RunSpec: contracts.ModuleRunSpec{
|
||||
ModuleKey: input.Spec.ModuleKey,
|
||||
@@ -495,17 +502,18 @@ func runModulePipeline(ctx context.Context, input collectSectionProposalsInput)
|
||||
go func(sectionPos int, enriched []proposals.EnrichedCorrectionProposal, sectionMetadata contracts.SectionMetadata) {
|
||||
defer vwg.Done()
|
||||
validated, err := validateSectionCandidates(runCtx, validateSectionCandidatesInput{
|
||||
Spec: input.Spec,
|
||||
Policy: input.Policy,
|
||||
Glossary: input.Glossary,
|
||||
Config: input.Config,
|
||||
WorkingTranscript: transcriptFromSection(sectionResult.section),
|
||||
SectionIndex: §ionMetadata.Index,
|
||||
Validators: validatorsOrdered,
|
||||
SectionEnriched: enriched,
|
||||
ValidationLLMClient: input.ValidationClient,
|
||||
DiagnosticsDir: input.ValidationDiagnosticsDir,
|
||||
Timings: input.Timings,
|
||||
Spec: input.Spec,
|
||||
Policy: input.Policy,
|
||||
Glossary: input.Glossary,
|
||||
Config: input.Config,
|
||||
WorkingTranscript: transcriptFromSection(sectionResult.section),
|
||||
SectionIndex: §ionMetadata.Index,
|
||||
Validators: validatorsOrdered,
|
||||
SectionEnriched: enriched,
|
||||
ValidationLLMClient: input.ValidationClient,
|
||||
DiagnosticsDir: input.ValidationDiagnosticsDir,
|
||||
Timings: input.Timings,
|
||||
ReportDiagnosticError: input.ReportDiagnosticError,
|
||||
})
|
||||
setErr(err)
|
||||
validationResults <- sectionValidationResult{
|
||||
@@ -578,17 +586,19 @@ func runModulePipeline(ctx context.Context, input collectSectionProposalsInput)
|
||||
}
|
||||
|
||||
type validateSectionCandidatesInput struct {
|
||||
Spec contracts.ModuleRunSpec
|
||||
Policy proposals.ReplacementPolicy
|
||||
Glossary *schema.Glossary
|
||||
Config *config.Config
|
||||
WorkingTranscript *schema.Transcript
|
||||
SectionIndex *int
|
||||
Validators []contracts.Validator
|
||||
SectionEnriched []proposals.EnrichedCorrectionProposal
|
||||
ValidationLLMClient contracts.StructuredLLMClient
|
||||
DiagnosticsDir string
|
||||
Timings *validatorTimingCollector
|
||||
// ReportDiagnosticError may be called concurrently; it must not affect processing.
|
||||
ReportDiagnosticError func(error) `json:"-"`
|
||||
Spec contracts.ModuleRunSpec
|
||||
Policy proposals.ReplacementPolicy
|
||||
Glossary *schema.Glossary
|
||||
Config *config.Config
|
||||
WorkingTranscript *schema.Transcript
|
||||
SectionIndex *int
|
||||
Validators []contracts.Validator
|
||||
SectionEnriched []proposals.EnrichedCorrectionProposal
|
||||
ValidationLLMClient contracts.StructuredLLMClient
|
||||
DiagnosticsDir string
|
||||
Timings *validatorTimingCollector
|
||||
}
|
||||
|
||||
type validateSectionCandidatesResult struct {
|
||||
@@ -622,38 +632,34 @@ func validateSectionCandidates(ctx context.Context, input validateSectionCandida
|
||||
|
||||
validatorStarted := time.Now()
|
||||
vResult, err := validator.Validate(ctx, contracts.ValidationRequest{
|
||||
WorkingTranscript: input.WorkingTranscript,
|
||||
CandidateProposal: eligible,
|
||||
ModuleKey: input.Spec.ModuleKey,
|
||||
ModuleInstance: input.Spec.InstanceName,
|
||||
SectionIndex: input.SectionIndex,
|
||||
ValidatorOrder: &validatorOrder,
|
||||
ReplacementPolicy: input.Policy,
|
||||
Glossary: input.Glossary,
|
||||
Config: input.Config,
|
||||
LLMClient: input.ValidationLLMClient,
|
||||
DiagnosticsWriter: diagnosticsWriter,
|
||||
WorkingTranscript: input.WorkingTranscript,
|
||||
CandidateProposal: eligible,
|
||||
ModuleKey: input.Spec.ModuleKey,
|
||||
ModuleInstance: input.Spec.InstanceName,
|
||||
SectionIndex: input.SectionIndex,
|
||||
ValidatorOrder: &validatorOrder,
|
||||
ReplacementPolicy: input.Policy,
|
||||
Glossary: input.Glossary,
|
||||
Config: input.Config,
|
||||
LLMClient: input.ValidationLLMClient,
|
||||
DiagnosticsWriter: diagnosticsWriter,
|
||||
ReportDiagnosticError: input.ReportDiagnosticError,
|
||||
})
|
||||
if input.Timings != nil {
|
||||
input.Timings.record(input.Spec.ModuleKey, input.Spec.InstanceName, validator.Name(), validatormetadata.ClassOf(validator) == validatormetadata.ExecutionClassLLMBacked, time.Since(validatorStarted))
|
||||
}
|
||||
operations = append(operations, vResult.Operations...)
|
||||
|
||||
var decisionErr error
|
||||
if err != nil {
|
||||
return validateSectionCandidatesResult{
|
||||
decisions: decisions,
|
||||
rejected: rejected,
|
||||
operations: operations,
|
||||
filtered: filtered,
|
||||
}, fmt.Errorf("validator %q failed: %w", validator.Name(), err)
|
||||
decisionErr = validators.EnforcePartialDecisionCardinality(eligible, vResult.Decisions)
|
||||
} else {
|
||||
decisionErr = validators.EnforceDecisionCardinality(eligible, vResult.Decisions)
|
||||
}
|
||||
if err := validators.EnforceDecisionCardinality(eligible, vResult.Decisions); err != nil {
|
||||
if decisionErr != nil {
|
||||
return validateSectionCandidatesResult{
|
||||
decisions: decisions,
|
||||
rejected: rejected,
|
||||
operations: operations,
|
||||
filtered: filtered,
|
||||
}, fmt.Errorf("validator %q cardinality failed: %w", validator.Name(), err)
|
||||
decisions: decisions, rejected: rejected, operations: operations, filtered: filtered,
|
||||
}, fmt.Errorf("validator %q cardinality failed: %w", validator.Name(), errors.Join(err, decisionErr))
|
||||
}
|
||||
|
||||
nextEligible := make([]proposals.EnrichedCorrectionProposal, 0, len(eligible))
|
||||
@@ -694,6 +700,13 @@ func validateSectionCandidates(ctx context.Context, input validateSectionCandida
|
||||
})
|
||||
}
|
||||
|
||||
if err != nil {
|
||||
// Completed decisions are evidence only. A fatal return never grants
|
||||
// approval to candidates that have not finished the full chain.
|
||||
return validateSectionCandidatesResult{
|
||||
decisions: decisions, rejected: rejected, operations: operations, filtered: filtered,
|
||||
}, fmt.Errorf("validator %q failed: %w", validator.Name(), err)
|
||||
}
|
||||
eligible = nextEligible
|
||||
}
|
||||
|
||||
|
||||
@@ -1280,3 +1280,20 @@ func TestRunnerGlossaryModuleUsesConfidenceThreshold(t *testing.T) {
|
||||
t.Fatalf("expected low confidence reason, got %+v", out.ModuleResults[0].ValidatorRejected[0])
|
||||
}
|
||||
}
|
||||
|
||||
func TestFatalPartialDecisionsRejectUnknownAndDuplicateIdentity(t *testing.T) {
|
||||
origin := errors.New("original fatal failure")
|
||||
candidate := proposals.EnrichedCorrectionProposal{ProposalMetadata: proposals.ProposalMetadata{ProposalIndex: 7}}
|
||||
for _, decisions := range [][]validators.Decision{
|
||||
{{ProposalIndex: 8, Approved: true}},
|
||||
{{ProposalIndex: 7, Approved: true}, {ProposalIndex: 7, Approved: true}},
|
||||
} {
|
||||
v := fakeValidator{name: "broken", validateF: func(contracts.ValidationRequest) (validators.Result, error) {
|
||||
return validators.Result{Decisions: decisions}, origin
|
||||
}}
|
||||
result, err := validateSectionCandidates(context.Background(), validateSectionCandidatesInput{Validators: []contracts.Validator{v}, SectionEnriched: []proposals.EnrichedCorrectionProposal{candidate}})
|
||||
if !errors.Is(err, origin) || len(result.decisions) != 0 || len(result.approved) != 0 {
|
||||
t.Fatalf("invalid partial evidence accepted or origin lost: %+v %v", result, err)
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
@@ -151,7 +151,8 @@ func (v *LLMBackedValidator) Validate(ctx context.Context, req Request) (Result,
|
||||
|
||||
artifacts := InteractionArtifacts{}
|
||||
if req.DiagnosticsWriter != nil {
|
||||
artifacts, _ = req.DiagnosticsWriter.WriteInteraction(
|
||||
var diagnosticErr error
|
||||
artifacts, diagnosticErr = req.DiagnosticsWriter.WriteInteraction(
|
||||
stage,
|
||||
map[string]any{
|
||||
"validator_name": v.name,
|
||||
@@ -164,6 +165,9 @@ func (v *LLMBackedValidator) Validate(ctx context.Context, req Request) (Result,
|
||||
map[string]any{"decoded": response, "final_content": completion.FinalContent, "model": completion.Model},
|
||||
errPayload(err),
|
||||
)
|
||||
if diagnosticErr != nil && req.ReportDiagnosticError != nil {
|
||||
req.ReportDiagnosticError(fmt.Errorf("validation interaction %s: %w", stage, diagnosticErr))
|
||||
}
|
||||
}
|
||||
if err != nil {
|
||||
var failure *structuredoutput.Failure
|
||||
|
||||
@@ -23,6 +23,7 @@ type fakeStructuredLLMClient struct {
|
||||
}
|
||||
|
||||
type captureValidationDiagnosticsWriter struct {
|
||||
err error
|
||||
lastRequestMetadata any
|
||||
}
|
||||
|
||||
@@ -32,7 +33,7 @@ func (w *captureValidationDiagnosticsWriter) WriteInteraction(stage string, requ
|
||||
_ = responsePayload
|
||||
_ = errorPayload
|
||||
w.lastRequestMetadata = requestMetadata
|
||||
return InteractionArtifacts{}, nil
|
||||
return InteractionArtifacts{}, w.err
|
||||
}
|
||||
|
||||
func (f *fakeStructuredLLMClient) CompleteStructured(ctx context.Context, req StructuredCompletionRequest, out any) (StructuredCompletionResponse, error) {
|
||||
@@ -321,3 +322,17 @@ func TestEveryValidationTypeSelectsEmbeddedPromptAndVariables(t *testing.T) {
|
||||
})
|
||||
}
|
||||
}
|
||||
|
||||
func TestValidationDiagnosticFailureDoesNotChangeDecision(t *testing.T) {
|
||||
cause := errors.New("diagnostic storage unavailable")
|
||||
req := makeReq([]proposals.EnrichedCorrectionProposal{mk(0, "gestures", "Jesters")})
|
||||
req.LLMClient = &fakeStructuredLLMClient{responses: []LLMValidationResponse{{Validations: []LLMValidationDecision{{CorrectionIndex: 0, Approved: true, Confidence: 1, Reason: "ok"}}}}}
|
||||
req.DiagnosticsWriter = &captureValidationDiagnosticsWriter{err: cause}
|
||||
var reported error
|
||||
req.ReportDiagnosticError = func(err error) { reported = err }
|
||||
validator, _ := NewLLMBackedValidator("editorial_review", LLMValidatorTypeEditorialReview)
|
||||
result, err := validator.Validate(context.Background(), req)
|
||||
if err != nil || !errors.Is(reported, cause) || len(result.Decisions) != 1 || !result.Decisions[0].Approved || result.Decisions[0].DiagnosticArtifactPath != "" || len(result.Operations) != 1 || !result.Operations[0].Succeeded || len(result.Operations[0].RecoveryReasons) != 0 {
|
||||
t.Fatalf("debug failure changed outcome: %+v %v reported=%v", result, err, reported)
|
||||
}
|
||||
}
|
||||
|
||||
@@ -29,17 +29,19 @@ const (
|
||||
|
||||
// Request is the runtime input shared by deterministic validators.
|
||||
type Request struct {
|
||||
SectionIndex *int `json:"section_index,omitempty"`
|
||||
WorkingTranscript *schema.Transcript `json:"-"`
|
||||
CandidateProposal []proposals.EnrichedCorrectionProposal `json:"candidate_proposals"`
|
||||
ModuleKey string `json:"module_key"`
|
||||
ModuleInstance string `json:"module_instance"`
|
||||
ValidatorOrder *int `json:"validator_order,omitempty"`
|
||||
ReplacementPolicy proposals.ReplacementPolicy `json:"replacement_policy"`
|
||||
Glossary *schema.Glossary `json:"-"`
|
||||
Config *config.Config `json:"-"`
|
||||
LLMClient StructuredLLMClient `json:"-"`
|
||||
DiagnosticsWriter InteractionDiagnosticsWriter `json:"-"`
|
||||
// ReportDiagnosticError may be called concurrently; it must not affect processing.
|
||||
ReportDiagnosticError func(error) `json:"-"`
|
||||
SectionIndex *int `json:"section_index,omitempty"`
|
||||
WorkingTranscript *schema.Transcript `json:"-"`
|
||||
CandidateProposal []proposals.EnrichedCorrectionProposal `json:"candidate_proposals"`
|
||||
ModuleKey string `json:"module_key"`
|
||||
ModuleInstance string `json:"module_instance"`
|
||||
ValidatorOrder *int `json:"validator_order,omitempty"`
|
||||
ReplacementPolicy proposals.ReplacementPolicy `json:"replacement_policy"`
|
||||
Glossary *schema.Glossary `json:"-"`
|
||||
Config *config.Config `json:"-"`
|
||||
LLMClient StructuredLLMClient `json:"-"`
|
||||
DiagnosticsWriter InteractionDiagnosticsWriter `json:"-"`
|
||||
}
|
||||
|
||||
// Decision is one validator decision for one proposal index.
|
||||
@@ -62,7 +64,8 @@ const (
|
||||
DispositionUnavailable DecisionDisposition = "unavailable"
|
||||
)
|
||||
|
||||
// Result is one validator output containing exactly one decision per proposal index.
|
||||
// Result contains one decision per proposal index on success. Fatal returns may
|
||||
// include completed partial decisions as evidence, never as application approval.
|
||||
type Result struct {
|
||||
ValidatorName string `json:"validator_name"`
|
||||
Decisions []Decision `json:"decisions"`
|
||||
@@ -88,6 +91,16 @@ type Validator interface {
|
||||
|
||||
// EnforceDecisionCardinality verifies every candidate proposal index receives exactly one decision.
|
||||
func EnforceDecisionCardinality(candidate []proposals.EnrichedCorrectionProposal, decisions []Decision) error {
|
||||
return enforceDecisionCardinality(candidate, decisions, true)
|
||||
}
|
||||
|
||||
// EnforcePartialDecisionCardinality accepts missing decisions on fatal returns,
|
||||
// but still rejects duplicate and unknown proposal identities.
|
||||
func EnforcePartialDecisionCardinality(candidate []proposals.EnrichedCorrectionProposal, decisions []Decision) error {
|
||||
return enforceDecisionCardinality(candidate, decisions, false)
|
||||
}
|
||||
|
||||
func enforceDecisionCardinality(candidate []proposals.EnrichedCorrectionProposal, decisions []Decision, requireAll bool) error {
|
||||
expected := make(map[int]struct{}, len(candidate))
|
||||
for _, c := range candidate {
|
||||
expected[c.ProposalIndex] = struct{}{}
|
||||
@@ -105,7 +118,7 @@ func EnforceDecisionCardinality(candidate []proposals.EnrichedCorrectionProposal
|
||||
}
|
||||
|
||||
for idx := range expected {
|
||||
if seen[idx] == 0 {
|
||||
if requireAll && seen[idx] == 0 {
|
||||
return fmt.Errorf("missing decision proposal index %d", idx)
|
||||
}
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user