Bind extraction reuse to transcript identity

This commit is contained in:
2026-08-10 22:15:53 +00:00
parent 4e4eff6ba7
commit f39fc94610
14 changed files with 256 additions and 41 deletions

View File

@@ -31,7 +31,11 @@ func (c *ArtifactCatalog) HydrateExtractionArtifacts(
if c == nil || m == nil || len(configured) == 0 {
return
}
proof := InspectExtractionEvidence(paths, m, configured)
input, err := ResolveExtractionInputIdentity(paths, m)
if err != nil {
return
}
proof := InspectExtractionEvidence(paths, m, configured, input)
if proof.State != ExtractionEvidenceValid {
return
}

View File

@@ -79,9 +79,10 @@ func TestHydrateExtractionArtifactsAcceptsOnlyCompleteCurrentBundle(t *testing.T
func TestInspectExtractionEvidenceClassifiesBundleStates(t *testing.T) {
tests := []struct {
name string
state ExtractionEvidenceState
mutate func(*testing.T, *SessionPaths, *manifest.Manifest)
name string
state ExtractionEvidenceState
recomputeInputIdentity bool
mutate func(*testing.T, *SessionPaths, *manifest.Manifest)
}{
{name: "valid", state: ExtractionEvidenceValid, mutate: func(_ *testing.T, _ *SessionPaths, _ *manifest.Manifest) {}},
{name: "absent", state: ExtractionEvidenceAbsent, mutate: func(_ *testing.T, _ *SessionPaths, m *manifest.Manifest) {
@@ -90,6 +91,12 @@ func TestInspectExtractionEvidenceClassifiesBundleStates(t *testing.T) {
{name: "obsolete version", state: ExtractionEvidenceObsolete, mutate: func(_ *testing.T, _ *SessionPaths, m *manifest.Manifest) {
m.Stages["extract"].Outputs[0].Contract.SchemaVersion = "99"
}},
{name: "changed direct transcript", state: ExtractionEvidenceObsolete, recomputeInputIdentity: true, mutate: func(t *testing.T, _ *SessionPaths, m *manifest.Manifest) {
writeExtractionFixtureFile(t, m.Stages["trim"].Outputs[0].LocalPath, `{"segments":[{"id":"changed"}]}`)
}},
{name: "changed direct producer", state: ExtractionEvidenceObsolete, recomputeInputIdentity: true, mutate: func(_ *testing.T, _ *SessionPaths, m *manifest.Manifest) {
m.Stages["trim"].Outputs[0].ProducerRunID = "trim-run-2"
}},
{name: "incomplete", state: ExtractionEvidenceObsolete, mutate: func(_ *testing.T, _ *SessionPaths, m *manifest.Manifest) {
m.Stages["extract"].Outputs = m.Stages["extract"].Outputs[1:]
}},
@@ -115,8 +122,18 @@ func TestInspectExtractionEvidenceClassifiesBundleStates(t *testing.T) {
for _, test := range tests {
t.Run(test.name, func(t *testing.T) {
paths, currentManifest, definitions := validExtractionCatalogFixture(t)
input, err := ResolveExtractionInputIdentity(paths, currentManifest)
if err != nil {
t.Fatal(err)
}
test.mutate(t, &paths, currentManifest)
proof := InspectExtractionEvidence(paths, currentManifest, definitions)
if test.recomputeInputIdentity {
input, err = ResolveExtractionInputIdentity(paths, currentManifest)
if err != nil {
t.Fatal(err)
}
}
proof := InspectExtractionEvidence(paths, currentManifest, definitions, input)
if proof.State != test.state {
t.Fatalf("proof = %#v, want %q", proof, test.state)
}
@@ -231,18 +248,31 @@ func validExtractionCatalogFixture(t *testing.T) (SessionPaths, *manifest.Manife
t.Helper()
root := t.TempDir()
paths := SessionPaths{Root: root, ArtifactsDir: filepath.Join(root, "artifacts")}
trimmedPath := filepath.Join(paths.Root, filepath.FromSlash(TranscriptPathFinalTrimmed))
bundleRoot := filepath.Join(paths.ArtifactsDir, "notarius", "extract-run-1")
lanePath := filepath.Join(bundleRoot, "lanes", "encounters.json")
indexPath := filepath.Join(bundleRoot, "index.json")
writeExtractionFixtureFile(t, lanePath, `{"encounters":[]}`)
writeExtractionFixtureFile(t, indexPath, `{"lanes":[]}`)
writeExtractionFixtureFile(t, trimmedPath, `{"segments":[]}`)
definitions := extractionFixtureDefinitions()
m := manifest.New("session", fixtureTime)
m.Stages["trim"] = &manifest.StageRecord{
Name: "trim", Status: manifest.StatusSucceeded,
Outputs: []manifest.ArtifactRecord{{
Kind: TranscriptOutputKindFinalTrimmed, LocalPath: trimmedPath, ProducerRunID: "trim-run-1",
}},
}
input, err := ResolveExtractionInputIdentity(paths, m)
if err != nil {
t.Fatal(err)
}
m.Stages["extract"] = &manifest.StageRecord{
Name: "extract", Status: manifest.StatusSucceeded,
Metadata: map[string]any{
"narratio_run_id": "extract-run-1", "bundle_root": bundleRoot,
"receipt": map[string]any{"run_id": "notarius-run-1", "pipeline_id": "campaign.extract"},
"receipt": map[string]any{"run_id": "notarius-run-1", "pipeline_id": "campaign.extract"},
extractionMetadataInput: input.Metadata(),
},
Outputs: []manifest.ArtifactRecord{
{

View File

@@ -26,8 +26,13 @@ type ExtractionEvidence struct {
Outputs map[string]string
}
// InspectExtractionEvidence verifies structure, confinement, identity, contracts, and payload bytes.
func InspectExtractionEvidence(paths SessionPaths, m *manifest.Manifest, configured map[string]ExtractionArtifactDefinition) ExtractionEvidence {
// InspectExtractionEvidence verifies structure, confinement, identities, contracts, and payload bytes.
func InspectExtractionEvidence(
paths SessionPaths,
m *manifest.Manifest,
configured map[string]ExtractionArtifactDefinition,
input ExtractionInputIdentity,
) ExtractionEvidence {
if m == nil || len(configured) == 0 {
return ExtractionEvidence{State: ExtractionEvidenceAbsent, Reason: "extraction evidence is absent"}
}
@@ -35,6 +40,9 @@ func InspectExtractionEvidence(paths SessionPaths, m *manifest.Manifest, configu
if r == nil || r.Name != extractStageName || r.Status != manifest.StatusSucceeded {
return ExtractionEvidence{State: ExtractionEvidenceAbsent, Reason: "extract stage has no succeeded result"}
}
if !extractionInputMatchesMetadata(r.Metadata, input) {
return ExtractionEvidence{State: ExtractionEvidenceObsolete, Reason: "extract result does not match the current trimmed transcript"}
}
runID := extractionMetadataString(r.Metadata, extractionMetadataRun)
if !safeExtractionPathSegment(runID) {
return ExtractionEvidence{State: ExtractionEvidenceObsolete, Reason: "extract result has an invalid producing run ID"}

View File

@@ -0,0 +1,76 @@
package artifacts
import (
"crypto/sha256"
"encoding/hex"
"fmt"
"path/filepath"
"strings"
"gitea.maximumdirect.net/eric/narratio/internal/fileops"
"gitea.maximumdirect.net/eric/narratio/internal/manifest"
)
const extractionMetadataInput = "direct_input"
// ExtractionInputIdentity identifies the exact trimmed transcript used for extraction.
type ExtractionInputIdentity struct {
Path string `json:"-"`
Checksum string `json:"checksum"`
SourceID string `json:"source_id"`
ProducerStage string `json:"producer_stage"`
OutputKind string `json:"output_kind"`
ProducerRunID string `json:"producer_run_id"`
Provenance string `json:"provenance"`
}
// ResolveExtractionInputIdentity resolves and hashes the direct transcript input.
func ResolveExtractionInputIdentity(paths SessionPaths, m *manifest.Manifest) (ExtractionInputIdentity, error) {
resolved, err := ResolveSessionArtifact(paths, m, ArtifactTranscriptFinalTrimmed)
if err != nil {
return ExtractionInputIdentity{}, err
}
path, err := filepath.Abs(resolved.Path)
if err != nil {
return ExtractionInputIdentity{}, fmt.Errorf("resolve trimmed transcript path: %w", err)
}
bytes, err := fileops.ReadRegularFile(path, MaxResolvedArtifactBytes)
if err != nil {
return ExtractionInputIdentity{}, fmt.Errorf("read trimmed transcript: %w", err)
}
digest := sha256.Sum256(bytes)
return ExtractionInputIdentity{
Path: filepath.Clean(path),
Checksum: hex.EncodeToString(digest[:]),
SourceID: resolved.ID,
ProducerStage: resolved.ProducerStage,
OutputKind: resolved.OutputKind,
ProducerRunID: resolved.ProducerRunID,
Provenance: resolved.Provenance,
}, nil
}
// Metadata returns the durable fields needed to compare this identity on reuse.
func (identity ExtractionInputIdentity) Metadata() map[string]any {
return map[string]any{
"checksum": identity.Checksum,
"source_id": identity.SourceID,
"producer_stage": identity.ProducerStage,
"output_kind": identity.OutputKind,
"producer_run_id": identity.ProducerRunID,
"provenance": identity.Provenance,
}
}
func extractionInputMatchesMetadata(metadata map[string]any, current ExtractionInputIdentity) bool {
stored, ok := metadata[extractionMetadataInput].(map[string]any)
if !ok || strings.TrimSpace(current.Checksum) == "" || strings.TrimSpace(current.SourceID) == "" {
return false
}
for key, want := range current.Metadata() {
if extractionMetadataString(stored, key) != want {
return false
}
}
return true
}