410 lines
16 KiB
Go
410 lines
16 KiB
Go
package proposal_generation
|
|
|
|
import (
|
|
"context"
|
|
"encoding/json"
|
|
"errors"
|
|
"os"
|
|
"path/filepath"
|
|
"reflect"
|
|
"strings"
|
|
"testing"
|
|
|
|
ck "gitea.maximumdirect.net/eric/ConfigKit"
|
|
"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/responseschema"
|
|
"gitea.maximumdirect.net/eric/audita/internal/framework/structuredoutput"
|
|
)
|
|
|
|
type fakeStructuredClient struct {
|
|
responses []StructuredCorrectionSet
|
|
err error
|
|
completion contracts.StructuredCompletionResponse
|
|
calls []contracts.StructuredCompletionRequest
|
|
}
|
|
|
|
func (f *fakeStructuredClient) CompleteStructured(ctx context.Context, req contracts.StructuredCompletionRequest, out any) (contracts.StructuredCompletionResponse, error) {
|
|
_ = ctx
|
|
f.calls = append(f.calls, req)
|
|
if f.err != nil {
|
|
return f.completion, f.err
|
|
}
|
|
target, ok := out.(*StructuredCorrectionSet)
|
|
if !ok {
|
|
return contracts.StructuredCompletionResponse{}, errors.New("unexpected output type")
|
|
}
|
|
if len(f.responses) == 0 {
|
|
return contracts.StructuredCompletionResponse{}, errors.New("unexpected call")
|
|
}
|
|
*target = f.responses[0]
|
|
f.responses = f.responses[1:]
|
|
var content strings.Builder
|
|
for _, value := range req.Variables {
|
|
content.WriteString(value)
|
|
}
|
|
response := f.completion
|
|
response.Prepared = &structuredoutput.PreparedSnapshot{Messages: []structuredoutput.Message{{Role: "user", Content: content.String()}}}
|
|
return response, nil
|
|
}
|
|
|
|
type captureDiagnosticsWriter struct {
|
|
err error
|
|
lastStage string
|
|
lastRequestMetadata any
|
|
lastRequestPayload any
|
|
}
|
|
|
|
func (w *captureDiagnosticsWriter) WriteInteraction(stage string, requestMetadata any, requestPayload any, responsePayload any, errorPayload any) (InteractionArtifacts, error) {
|
|
w.lastStage = stage
|
|
w.lastRequestMetadata = requestMetadata
|
|
w.lastRequestPayload = requestPayload
|
|
_ = responsePayload
|
|
_ = errorPayload
|
|
return InteractionArtifacts{}, w.err
|
|
}
|
|
|
|
func defaultRequest(t *testing.T) Request {
|
|
t.Helper()
|
|
cfg := config.Config{}
|
|
return Request{
|
|
ModuleKey: "test_module",
|
|
ModuleInstance: "test_module",
|
|
ReplacementPolicy: proposals.ReplacementPolicyRequireUnique,
|
|
Config: &cfg,
|
|
PromptID: "modules.grammar.proposal",
|
|
PromptVersion: "v1",
|
|
Variables: map[string]string{"SectionJSON": "[]"},
|
|
WorkingTranscript: &schema.Transcript{Segments: []schema.Segment{{ID: 7, Text: "teh"}, {ID: 25, Text: "other"}}},
|
|
StartIndex: 0,
|
|
}
|
|
}
|
|
|
|
func TestGenerateCandidatesSuccess(t *testing.T) {
|
|
client := &fakeStructuredClient{
|
|
responses: []StructuredCorrectionSet{
|
|
{
|
|
Corrections: []StructuredCorrectionProposal{
|
|
{SegmentReference: 1, OriginalText: "teh", CorrectedText: "the", Confidence: 0.9},
|
|
},
|
|
},
|
|
},
|
|
}
|
|
section := contracts.SectionMetadata{Index: 2}
|
|
req := defaultRequest(t)
|
|
req.LLMClient = client
|
|
req.StartIndex = 10
|
|
req.Section = §ion
|
|
|
|
got, err := GenerateCandidates(context.Background(), req)
|
|
if err != nil {
|
|
t.Fatalf("GenerateCandidates error: %v", err)
|
|
}
|
|
if len(got.Corrections) != 1 || len(got.Enriched) != 1 {
|
|
t.Fatalf("unexpected proposal lengths: %+v", got)
|
|
}
|
|
if got.Corrections[0].TargetSegmentID != 7 {
|
|
t.Fatalf("local reference not resolved: %+v", got.Corrections)
|
|
}
|
|
if got.Enriched[0].ProposalIndex != 10 {
|
|
t.Fatalf("expected proposal index 10, got %d", got.Enriched[0].ProposalIndex)
|
|
}
|
|
if got.Enriched[0].SectionIndex == nil || *got.Enriched[0].SectionIndex != 2 {
|
|
t.Fatalf("expected section index 2, got %v", got.Enriched[0].SectionIndex)
|
|
}
|
|
if len(got.Operations) != 1 || !got.Operations[0].Succeeded || got.Operations[0].Reference.SectionIndex != 2 || got.Operations[0].Reference.Kind != outcomes.OperationProposal {
|
|
t.Fatalf("proposal operation identity: %+v", got.Operations)
|
|
}
|
|
}
|
|
|
|
func TestGenerateCandidatesUsesCorrectionSetSchema(t *testing.T) {
|
|
client := &fakeStructuredClient{
|
|
responses: []StructuredCorrectionSet{
|
|
{Corrections: []StructuredCorrectionProposal{{SegmentReference: 1, OriginalText: "teh", CorrectedText: "the", Confidence: 0.9}}},
|
|
},
|
|
}
|
|
req := defaultRequest(t)
|
|
req.LLMClient = client
|
|
|
|
_, err := GenerateCandidates(context.Background(), req)
|
|
if err != nil {
|
|
t.Fatalf("GenerateCandidates error: %v", err)
|
|
}
|
|
if len(client.calls) != 1 {
|
|
t.Fatalf("expected 1 LLM call, got %d", len(client.calls))
|
|
}
|
|
call := client.calls[0]
|
|
if call.Schema != responseschema.CorrectionSetKey || call.PromptID != req.PromptID || call.Context.Role != "proposal" {
|
|
t.Fatalf("unexpected call contract: %+v", call)
|
|
}
|
|
}
|
|
|
|
func TestGenerateCandidatesDiagnosticsIncludeSchemaMetadata(t *testing.T) {
|
|
client := &fakeStructuredClient{
|
|
responses: []StructuredCorrectionSet{
|
|
{Corrections: []StructuredCorrectionProposal{{SegmentReference: 1, OriginalText: "teh", CorrectedText: "the", Confidence: 0.9}}},
|
|
},
|
|
}
|
|
diag := &captureDiagnosticsWriter{}
|
|
req := defaultRequest(t)
|
|
req.LLMClient = client
|
|
req.DiagnosticsWriter = diag
|
|
|
|
_, err := GenerateCandidates(context.Background(), req)
|
|
if err != nil {
|
|
t.Fatalf("GenerateCandidates error: %v", err)
|
|
}
|
|
metadata, ok := diag.lastRequestMetadata.(map[string]any)
|
|
if !ok {
|
|
t.Fatalf("expected request metadata map, got %T", diag.lastRequestMetadata)
|
|
}
|
|
schemaMap, ok := metadata["response_schema"].(map[string]any)
|
|
if !ok {
|
|
t.Fatalf("expected response_schema metadata map, got %T", metadata["response_schema"])
|
|
}
|
|
want := responseschema.MustLookup(responseschema.CorrectionSetKey)
|
|
if schemaMap["id"] != want.ID || schemaMap["version"] != want.Version || schemaMap["name"] != want.Name || schemaMap["sha256"] != want.SHA256 {
|
|
t.Fatalf("unexpected diagnostics schema metadata: got=%v want=%+v", schemaMap, want)
|
|
}
|
|
|
|
}
|
|
|
|
func TestGenerateCandidatesInvalidCorrectionIsPreservedForLaterValidation(t *testing.T) {
|
|
client := &fakeStructuredClient{
|
|
responses: []StructuredCorrectionSet{
|
|
{
|
|
Corrections: []StructuredCorrectionProposal{
|
|
{SegmentReference: 0, OriginalText: "x", CorrectedText: "", Confidence: 1.2},
|
|
},
|
|
},
|
|
},
|
|
}
|
|
req := defaultRequest(t)
|
|
req.LLMClient = client
|
|
|
|
result, err := GenerateCandidates(context.Background(), req)
|
|
if err != nil {
|
|
t.Fatalf("expected invalid correction to survive generation, got %v", err)
|
|
}
|
|
if len(result.Corrections) != 1 {
|
|
t.Fatalf("expected one correction, got %+v", result)
|
|
}
|
|
if result.Corrections[0].TargetSegmentID != 0 || result.Corrections[0].Confidence != 1.2 {
|
|
t.Fatalf("unexpected preserved correction: %+v", result.Corrections[0])
|
|
}
|
|
if len(result.Operations) != 1 || !result.Operations[0].Succeeded {
|
|
t.Fatalf("invalid individual correction should remain a usable response: %+v", result.Operations)
|
|
}
|
|
}
|
|
|
|
func TestGenerateCandidatesMalformedStructuredOutputReturnsFailedOperation(t *testing.T) {
|
|
client := &fakeStructuredClient{err: &structuredoutput.Failure{Class: structuredoutput.FailureRecoverable, Reason: outcomes.ReasonInvalidOutputExhausted, Message: "structured output remained invalid"}}
|
|
req := defaultRequest(t)
|
|
req.LLMClient = client
|
|
|
|
result, err := GenerateCandidates(context.Background(), req)
|
|
if err != nil {
|
|
t.Fatalf("expected recoverable model operation, got %v", err)
|
|
}
|
|
if len(result.Corrections) != 0 || len(result.Enriched) != 0 {
|
|
t.Fatalf("expected no proposals on malformed response, got %+v", result)
|
|
}
|
|
if len(result.Operations) != 1 || result.Operations[0].Succeeded || result.Operations[0].FailureReason != outcomes.ReasonInvalidOutputExhausted {
|
|
t.Fatalf("expected one failed operation, got %+v", result.Operations)
|
|
}
|
|
}
|
|
|
|
func TestGenerateCandidatesFatalSetupPreservesCallFacts(t *testing.T) {
|
|
client := &fakeStructuredClient{err: &structuredoutput.Failure{Class: structuredoutput.FailureFatal, Reason: outcomes.ReasonSetupFailure, Message: "structured model setup failed", Cause: errors.New("broken prompt")}, completion: structuredoutput.Response{Execution: outcomes.ExecutionCounts{Available: true, ProviderAttempts: 0}}}
|
|
req := defaultRequest(t)
|
|
req.LLMClient = client
|
|
result, err := GenerateCandidates(context.Background(), req)
|
|
if err == nil || len(result.Operations) != 1 || result.Operations[0].FailureReason != outcomes.ReasonSetupFailure || !result.Operations[0].Counts.Available {
|
|
t.Fatalf("fatal setup lost call facts: %+v, %v", result, err)
|
|
}
|
|
}
|
|
|
|
func TestGenerateCandidatesValidEmptyResponseIsSuccessful(t *testing.T) {
|
|
client := &fakeStructuredClient{responses: []StructuredCorrectionSet{{}}}
|
|
req := defaultRequest(t)
|
|
req.LLMClient = client
|
|
result, err := GenerateCandidates(context.Background(), req)
|
|
if err != nil || len(result.Corrections) != 0 || len(result.Operations) != 1 || !result.Operations[0].Succeeded {
|
|
t.Fatalf("valid empty response: %+v, %v", result, err)
|
|
}
|
|
}
|
|
|
|
func TestGenerateCandidatesDeterministicIndexAssignment(t *testing.T) {
|
|
baseResponse := StructuredCorrectionSet{
|
|
Corrections: []StructuredCorrectionProposal{
|
|
{SegmentReference: 1, OriginalText: "a", CorrectedText: "A", Confidence: 0.9},
|
|
{SegmentReference: 2, OriginalText: "b", CorrectedText: "B", Confidence: 0.9},
|
|
},
|
|
}
|
|
clientA := &fakeStructuredClient{responses: []StructuredCorrectionSet{baseResponse}}
|
|
clientB := &fakeStructuredClient{responses: []StructuredCorrectionSet{baseResponse}}
|
|
|
|
reqA := defaultRequest(t)
|
|
reqA.LLMClient = clientA
|
|
reqA.StartIndex = 3
|
|
first, err := GenerateCandidates(context.Background(), reqA)
|
|
if err != nil {
|
|
t.Fatalf("first generation failed: %v", err)
|
|
}
|
|
|
|
reqB := defaultRequest(t)
|
|
reqB.LLMClient = clientB
|
|
reqB.StartIndex = 3
|
|
second, err := GenerateCandidates(context.Background(), reqB)
|
|
if err != nil {
|
|
t.Fatalf("second generation failed: %v", err)
|
|
}
|
|
|
|
if !reflect.DeepEqual(first.Enriched, second.Enriched) {
|
|
t.Fatalf("expected stable enriched proposals\nfirst=%+v\nsecond=%+v", first.Enriched, second.Enriched)
|
|
}
|
|
}
|
|
|
|
func TestGenerateCandidatesMultipleSectionsStableMetadata(t *testing.T) {
|
|
client := &fakeStructuredClient{
|
|
responses: []StructuredCorrectionSet{
|
|
{Corrections: []StructuredCorrectionProposal{{SegmentReference: 1, OriginalText: "alpha", CorrectedText: "ALPHA", Confidence: 0.9}}},
|
|
{Corrections: []StructuredCorrectionProposal{{SegmentReference: 1, OriginalText: "charlie", CorrectedText: "CHARLIE", Confidence: 0.9}}},
|
|
},
|
|
}
|
|
section0 := contracts.SectionMetadata{Index: 0}
|
|
section1 := contracts.SectionMetadata{Index: 1}
|
|
|
|
req0 := defaultRequest(t)
|
|
req0.LLMClient = client
|
|
req0.Section = §ion0
|
|
req0.WorkingTranscript = &schema.Transcript{Segments: []schema.Segment{{ID: 71, Text: "alpha"}}}
|
|
req0.StartIndex = 0
|
|
part0, err := GenerateCandidates(context.Background(), req0)
|
|
if err != nil {
|
|
t.Fatalf("section 0 generation failed: %v", err)
|
|
}
|
|
|
|
req1 := defaultRequest(t)
|
|
req1.LLMClient = client
|
|
req1.Section = §ion1
|
|
req1.WorkingTranscript = &schema.Transcript{Segments: []schema.Segment{{ID: 903, Text: "charlie"}}}
|
|
req1.StartIndex = len(part0.Enriched)
|
|
part1, err := GenerateCandidates(context.Background(), req1)
|
|
if err != nil {
|
|
t.Fatalf("section 1 generation failed: %v", err)
|
|
}
|
|
|
|
all := append(append([]proposals.EnrichedCorrectionProposal(nil), part0.Enriched...), part1.Enriched...)
|
|
if len(all) != 2 {
|
|
t.Fatalf("expected 2 proposals, got %d", len(all))
|
|
}
|
|
if all[0].TargetSegmentID != 71 || all[1].TargetSegmentID != 903 {
|
|
t.Fatalf("section references did not reset: %+v", all)
|
|
}
|
|
if all[0].ProposalIndex != 0 || all[1].ProposalIndex != 1 {
|
|
t.Fatalf("unexpected proposal indexes: %d, %d", all[0].ProposalIndex, all[1].ProposalIndex)
|
|
}
|
|
if all[0].SectionIndex == nil || *all[0].SectionIndex != 0 || all[1].SectionIndex == nil || *all[1].SectionIndex != 1 {
|
|
t.Fatalf("unexpected section metadata: %+v", all)
|
|
}
|
|
}
|
|
|
|
func TestGenerateCandidatesDiagnosticsWrittenAndRedacted(t *testing.T) {
|
|
primarySecret := "proposal-primary-secret"
|
|
validationSecret := "proposal-validation-secret"
|
|
client := &fakeStructuredClient{
|
|
responses: []StructuredCorrectionSet{
|
|
{Corrections: []StructuredCorrectionProposal{{SegmentReference: 1, OriginalText: validationSecret, CorrectedText: "safe", Confidence: 0.9}}},
|
|
},
|
|
}
|
|
t.Setenv("PRIMARY_SECRET", primarySecret)
|
|
t.Setenv("VALIDATION_SECRET", validationSecret)
|
|
loaded, loadErr := config.Load(ck.Bytes("config.yml", []byte("version: 2\nbackends:\n primary:\n endpoint: https://primary.invalid/v1\n credential: PRIMARY_SECRET\n validation:\n endpoint: https://validation.invalid/v1\n credential: VALIDATION_SECRET\nllm:\n proposal_profile: base\n")), "")
|
|
if loadErr != nil {
|
|
t.Fatal(loadErr)
|
|
}
|
|
cfg := loaded.Config
|
|
req := defaultRequest(t)
|
|
req.Config = &cfg
|
|
req.LLMClient = client
|
|
req.DiagnosticsDir = t.TempDir()
|
|
req.Variables = map[string]string{"SectionJSON": "include secret " + primarySecret}
|
|
|
|
got, err := GenerateCandidates(context.Background(), req)
|
|
if err != nil {
|
|
t.Fatalf("GenerateCandidates error: %v", err)
|
|
}
|
|
if got.Artifacts.ResponsePayloadPath == "" || got.Artifacts.RequestPayloadPath == "" {
|
|
t.Fatalf("expected diagnostics artifact paths, got %+v", got.Artifacts)
|
|
}
|
|
|
|
for _, path := range []string{got.Artifacts.RequestPayloadPath, got.Artifacts.ResponsePayloadPath} {
|
|
raw, readErr := os.ReadFile(path)
|
|
if readErr != nil {
|
|
t.Fatalf("read artifact %q: %v", path, readErr)
|
|
}
|
|
if strings.Contains(string(raw), primarySecret) || strings.Contains(string(raw), validationSecret) {
|
|
t.Fatalf("artifact leaked configured secret in %q: %s", path, string(raw))
|
|
}
|
|
if !strings.Contains(string(raw), "[REDACTED]") {
|
|
t.Fatalf("expected redaction marker in artifact %q: %s", path, string(raw))
|
|
}
|
|
}
|
|
|
|
raw, readErr := os.ReadFile(got.Artifacts.RequestMetadataPath)
|
|
if readErr != nil {
|
|
t.Fatalf("read metadata artifact %q: %v", got.Artifacts.RequestMetadataPath, readErr)
|
|
}
|
|
var metadata map[string]any
|
|
if err := json.Unmarshal(raw, &metadata); err != nil {
|
|
t.Fatalf("unmarshal metadata artifact: %v", err)
|
|
}
|
|
schemaMap, ok := metadata["response_schema"].(map[string]any)
|
|
if !ok {
|
|
t.Fatalf("expected response_schema metadata in diagnostics, got %T", metadata["response_schema"])
|
|
}
|
|
want := responseschema.MustLookup(responseschema.CorrectionSetKey)
|
|
if schemaMap["id"] != want.ID || schemaMap["version"] != want.Version || schemaMap["name"] != want.Name || schemaMap["sha256"] != want.SHA256 {
|
|
t.Fatalf("unexpected schema metadata in diagnostics: got=%v want=%+v", schemaMap, want)
|
|
}
|
|
}
|
|
|
|
func TestGenerateCandidatesClientError(t *testing.T) {
|
|
client := &fakeStructuredClient{err: errors.New("boom")}
|
|
req := defaultRequest(t)
|
|
req.LLMClient = client
|
|
req.DiagnosticsDir = t.TempDir()
|
|
|
|
got, err := GenerateCandidates(context.Background(), req)
|
|
if err == nil || !strings.Contains(err.Error(), "completion failed") {
|
|
t.Fatalf("expected completion failure, got %v", err)
|
|
}
|
|
if len(got.Operations) != 1 || got.Operations[0].FailureReason != outcomes.ReasonInternalFailure || len(got.Corrections) != 0 {
|
|
t.Fatalf("expected fatal call facts without proposals, got %+v", got)
|
|
}
|
|
matches, globErr := filepath.Glob(filepath.Join(req.DiagnosticsDir, req.ModuleInstance, "*error-payload.json"))
|
|
if globErr != nil {
|
|
t.Fatalf("glob error: %v", globErr)
|
|
}
|
|
if len(matches) == 0 {
|
|
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)
|
|
}
|
|
}
|