Migrate item reconciliation to shared engine

This commit is contained in:
2026-08-09 16:46:43 +00:00
parent 8c071800cf
commit c7f79fb38e
7 changed files with 286 additions and 233 deletions

View File

@@ -3,7 +3,6 @@ package itemregistry
import (
"context"
"errors"
"fmt"
"reflect"
"sort"
@@ -13,20 +12,19 @@ import (
"gitea.maximumdirect.net/eric/notarius/internal/core/source"
"gitea.maximumdirect.net/eric/notarius/internal/framework/contracts"
"gitea.maximumdirect.net/eric/notarius/internal/framework/pipeline"
"gitea.maximumdirect.net/eric/notarius/internal/framework/semanticreconcile"
"gitea.maximumdirect.net/eric/notarius/internal/modules/dnd"
"gitea.maximumdirect.net/eric/notarius/internal/modules/dnd/items/identity"
"gitea.maximumdirect.net/eric/notarius/internal/modules/dnd/shared"
"gitea.maximumdirect.net/eric/notarius/internal/modules/dnd/shared/diagnostics"
"gitea.maximumdirect.net/eric/notarius/internal/modules/dnd/shared/entityreconcile"
)
const (
Key = "dnd/item-registry"
PromptID = "dnd.item_registry.normalize"
normalizationPolicy = "dnd.item_registry.normalize.v2"
semanticContextPolicy = "dnd.entity_reconcile.context.v1"
semanticContextRadius = 2
NormalizationPolicy = normalizationPolicy
Key = "dnd/item-registry"
PromptID = "dnd.item_registry.normalize"
PromptVersion = "v1"
normalizationPolicy = "dnd.item_registry.normalize.v3"
NormalizationPolicy = normalizationPolicy
ReasonCodeItemFieldsNormalized = "item_fields_normalized"
ReasonCodeItemIDRecomputed = "item_id_recomputed"
@@ -47,9 +45,7 @@ var _ pipeline.CheckpointFingerprintProvider = (*Normalizer)(nil)
type Options struct{}
type Normalizer struct {
llm contracts.StructuredLLMClient
promptSHA string
responseSchemaSHA string
engine *semanticreconcile.Engine
}
func New(llmClient contracts.StructuredLLMClient, _ Options) (*Normalizer, error) {
@@ -60,46 +56,45 @@ func New(llmClient contracts.StructuredLLMClient, _ Options) (*Normalizer, error
if err != nil {
return nil, normalizerErrorf("load prompt metadata: %w", err)
}
responseSchema, err := entityreconcile.LoadResponseSchema()
engine, err := semanticreconcile.NewEngine(llmClient, semanticreconcile.PromptSpec{
ID: PromptID, Version: PromptVersion, SHA256: promptSHA,
}, semanticreconcile.DefaultLimits())
if err != nil {
return nil, normalizerErrorf("load response schema: %w", err)
return nil, normalizerErrorf("construct semantic reconciliation engine: %w", err)
}
return &Normalizer{llm: llmClient, promptSHA: promptSHA, responseSchemaSHA: responseSchema.SHA256}, nil
return &Normalizer{engine: engine}, nil
}
func (n *Normalizer) Key() string { return Key }
func (n *Normalizer) ReferenceSlots() []contracts.ReferenceSlot { return nil }
func (n *Normalizer) ManifestMetadata() map[string]any {
if n == nil {
if n == nil || n.engine == nil {
return nil
}
return map[string]any{
"prompt_id": PromptID, "prompt_version": entityreconcile.SchemaVersion, "prompt_sha256": n.promptSHA,
"response_schema_key": string(entityreconcile.ResponseSchemaKey), "response_schema_id": entityreconcile.ResponseSchemaID,
"response_schema_name": entityreconcile.ResponseSchemaName, "response_schema_version": entityreconcile.SchemaVersion,
"response_schema_sha256": n.responseSchemaSHA, "identity_policy": identity.Policy,
"normalization_policy": normalizationPolicy, "semantic_context_policy": semanticContextPolicy, "semantic_context_radius": semanticContextRadius,
}
metadata := n.engine.ManifestMetadata()
metadata["identity_policy"] = identity.Policy
metadata["normalization_policy"] = normalizationPolicy
return metadata
}
func (n *Normalizer) CheckpointFingerprints() []pipeline.CheckpointFingerprint {
if n == nil {
if n == nil || n.engine == nil {
return nil
}
return []pipeline.CheckpointFingerprint{
{Name: "prompt", Value: n.promptSHA}, {Name: "response_schema", Value: n.responseSchemaSHA},
{Name: "identity_policy", Value: identity.Policy}, {Name: "normalization_policy", Value: normalizationPolicy},
{Name: "semantic_context_policy", Value: fmt.Sprintf("%s:%d", semanticContextPolicy, semanticContextRadius)},
}
fingerprints := n.engine.CheckpointFingerprints()
return append(fingerprints,
pipeline.CheckpointFingerprint{Name: "identity_policy", Value: identity.Policy},
pipeline.CheckpointFingerprint{Name: "normalization_policy", Value: normalizationPolicy},
)
}
func (n *Normalizer) Normalize(ctx context.Context, req contracts.TypedNormalizeRequest[dnd.ItemRegistry]) (contracts.TypedNormalizeResult[dnd.ItemRegistry], error) {
if n == nil {
return contracts.TypedNormalizeResult[dnd.ItemRegistry]{}, normalizerErrorf("normalizer must not be nil")
}
if n.llm == nil {
return contracts.TypedNormalizeResult[dnd.ItemRegistry]{}, normalizerErrorf("LLM client must not be nil")
if n.engine == nil {
return contracts.TypedNormalizeResult[dnd.ItemRegistry]{}, normalizerErrorf("semantic reconciliation engine must not be nil")
}
if ctx == nil {
return contracts.TypedNormalizeResult[dnd.ItemRegistry]{}, normalizerErrorf("context must not be nil")
@@ -111,35 +106,44 @@ func (n *Normalizer) Normalize(ctx context.Context, req contracts.TypedNormalize
order := shared.NewSourceRefOrder(req.Source)
records, warnings := preprocessRecords(req.MergeOutput.Value, order)
deterministic := recordList(records)
materials, ready, err := entityreconcile.BuildContext(req.Source, reconciliationCandidates(records), semanticContextRadius)
if err != nil {
return contracts.TypedNormalizeResult[dnd.ItemRegistry]{}, normalizerErrorf("build semantic context: %w", err)
}
if !ready {
if len(records) < 2 || req.Source == nil {
return contracts.TypedNormalizeResult[dnd.ItemRegistry]{Value: deterministic, Warnings: limitWarnings(warnings)}, nil
}
var response entityreconcile.ProposalResponse
if _, err := n.llm.CompleteStructured(ctx, contracts.StructuredCompletionRequest{
StageName: Key, PromptID: PromptID, PromptVersion: entityreconcile.SchemaVersion,
candidates, envelopes, err := reconciliationInputs(records)
if err != nil {
return contracts.TypedNormalizeResult[dnd.ItemRegistry]{}, normalizerErrorf("prepare semantic reconciliation inputs: %w", err)
}
reconciliation, err := n.engine.Reconcile(ctx, semanticreconcile.Request{
StageName: Key, Source: req.Source, Candidates: candidates,
ProfileID: req.LLMProfile, SessionID: req.SessionID,
Inputs: contracts.LLMInputSet{"candidates": materials.Candidates, "transcript": materials.Transcript},
}, &response); err != nil {
if errors.Is(err, contracts.ErrInvalidStructuredOutput) {
return n.invalidStructuredResult(deterministic, warnings), nil
}
return contracts.TypedNormalizeResult[dnd.ItemRegistry]{}, normalizerErrorf("complete structured output: %w", err)
})
if err != nil {
return contracts.TypedNormalizeResult[dnd.ItemRegistry]{}, normalizerErrorf("reconcile semantic duplicates: %w", err)
}
switch reconciliation.Disposition() {
case semanticreconcile.SkippedInsufficientCandidates:
return contracts.TypedNormalizeResult[dnd.ItemRegistry]{Value: deterministic, Warnings: limitWarnings(warnings)}, nil
case semanticreconcile.SkippedLimitExceeded:
return contracts.TypedNormalizeResult[dnd.ItemRegistry]{Value: deterministic, Warnings: limitWarningsWithSemanticFallback(warnings)}, nil
case semanticreconcile.RetryableInvalidStructuredOutput:
return n.invalidStructuredResult(deterministic, warnings), nil
case semanticreconcile.Complete, semanticreconcile.RetryableDiscardedProposalGroups:
default:
return contracts.TypedNormalizeResult[dnd.ItemRegistry]{}, normalizerErrorf("unknown semantic reconciliation disposition %d", reconciliation.Disposition())
}
applied, semanticWarnings, rejectedGroups, err := applyReconciliationPlan(reconciliation.Plan(), records, envelopes, order)
if err != nil {
return contracts.TypedNormalizeResult[dnd.ItemRegistry]{}, normalizerErrorf("apply semantic reconciliation plan: %w", err)
}
assessment := materials.Assess(response)
applied, semanticWarnings, rejectedGroups := applySafeGroups(records, reconciliationGroups(assessment, materials.CandidateKeys()), order)
warnings = append(warnings, semanticWarnings...)
if rejectedGroups > 0 {
return currencyRetryResult(recordList(applied), warnings, rejectedGroups), nil
}
if assessment.DiscardedGroups() == 0 {
discardedGroups := reconciliation.DiscardedGroupCount() + rejectedGroups
if discardedGroups == 0 {
return contracts.TypedNormalizeResult[dnd.ItemRegistry]{Value: recordList(applied), Warnings: limitWarnings(warnings)}, nil
}
return retryResult(recordList(applied), warnings, assessment), nil
return retryResult(recordList(applied), warnings, reconciliation, rejectedGroups), nil
}
func (n *Normalizer) invalidStructuredResult(value dnd.ItemRegistry, warnings []contracts.Warning) contracts.TypedNormalizeResult[dnd.ItemRegistry] {
@@ -149,17 +153,15 @@ func (n *Normalizer) invalidStructuredResult(value dnd.ItemRegistry, warnings []
}}
}
func retryResult(value dnd.ItemRegistry, warnings []contracts.Warning, assessment entityreconcile.Assessment) contracts.TypedNormalizeResult[dnd.ItemRegistry] {
func retryResult(value dnd.ItemRegistry, warnings []contracts.Warning, reconciliation semanticreconcile.Result, rejectedGroups int) contracts.TypedNormalizeResult[dnd.ItemRegistry] {
details := reconciliationIssues(reconciliation.Issues())
if rejectedGroups > 0 {
details = append(details, "currency may only be consolidated with aliases of one denomination")
}
discardedGroups := reconciliation.DiscardedGroupCount() + rejectedGroups
return contracts.TypedNormalizeResult[dnd.ItemRegistry]{Value: value, Warnings: limitWarningsForRetry(warnings), Retry: &contracts.NormalizeRetry{
ReasonCode: ReasonCodeItemSemanticProposalInvalid, Message: diagnostics.Aggregate("semantic proposal requires retry", reconciliationIssues(assessment)),
FallbackWarnings: []contracts.Warning{semanticFallbackWarning(assessment.DiscardedGroups())},
}}
}
func currencyRetryResult(value dnd.ItemRegistry, warnings []contracts.Warning, rejectedGroups int) contracts.TypedNormalizeResult[dnd.ItemRegistry] {
return contracts.TypedNormalizeResult[dnd.ItemRegistry]{Value: value, Warnings: limitWarningsForRetry(warnings), Retry: &contracts.NormalizeRetry{
ReasonCode: ReasonCodeItemSemanticProposalInvalid, Message: "semantic proposal requires retry: currency may only be consolidated with aliases of one denomination",
FallbackWarnings: []contracts.Warning{semanticFallbackWarning(rejectedGroups)},
ReasonCode: ReasonCodeItemSemanticProposalInvalid, Message: diagnostics.Aggregate("semantic proposal requires retry", details),
FallbackWarnings: []contracts.Warning{semanticFallbackWarning(discardedGroups)},
}}
}
@@ -187,6 +189,10 @@ func limitWarningsForRetry(warnings []contracts.Warning) []contracts.Warning {
return append(bounded, contracts.Warning{Scope: "items", ReasonCode: ReasonCodeItemNormalizationWarningsOmitted, Message: fmt.Sprintf("%d additional warning(s) omitted", len(warnings)-displayed)})
}
func limitWarningsWithSemanticFallback(warnings []contracts.Warning) []contracts.Warning {
return append(limitWarningsForRetry(warnings), semanticFallbackWarning(-1))
}
type normalizedRecord struct {
item dnd.Item
inputIndexes []int