Simplify contextual entity grounding

This commit is contained in:
2026-08-08 15:47:41 +00:00
parent 20397ef710
commit d9b87347b8
6 changed files with 45 additions and 89 deletions

View File

@@ -5,8 +5,8 @@ across records, or infer ownership that the transcript does not establish.
For every occurrence, use the supplied canonical item `name`. Record a stated
quantity as an integer and leave it null when the transcript does not state
one. Use a concise observed item name and preserve the stated currency
denomination.
one. Preserve the stated currency denomination through the selected canonical
registry name.
Use `discovered` when the party learns of or encounters an item without
establishing possession. Use `acquired` when the party or a party member gains

View File

@@ -274,8 +274,9 @@ path.
registry and the current `*source.SourceDocument`. Its API must provide:
- a cloned `contracts.LLMInputMaterial` for the `location_registry` slot;
- deterministic resolution of a returned selector to one cloned
`dnd.Location`; and
- the digest of the exact model projection.
`dnd.Location`.
The prompt material's existing `Digest` field owns the digest of the exact
model projection; do not expose a second grounding-specific digest API.
4. Construct the projection in registry order. Group entries by the existing
location comparison key:
- every projection entry has exactly the required fields `name`,
@@ -299,10 +300,10 @@ path.
7. Separate deterministic identity fingerprinting from LLM material. Add
`IdentityDigest()` over the registry's ordered `{id,name}` identity
projection, and update the location normalizer and registry-validator
checkpoint consumers to use it. The new operation grounding owns the model
projection digest. Retain the old ID-bearing prompt accessor only as a
documented transitional dependency of the still-unchanged location
occurrence extractor; do not add new callers.
checkpoint consumers to use it. The new operation grounding carries the
model projection digest in its `LLMInputMaterial`. Retain the old ID-bearing
prompt accessor only as a documented transitional dependency of the
still-unchanged location occurrence extractor; do not add new callers.
8. Add focused tests for unique names, same-name context and selectors,
canonical range order, deterministic projection/digest, defensive copies,
exact selector resolution, nil/foreign/invalid references, selector
@@ -367,10 +368,12 @@ remove durable location IDs from the prompt and private response.
7. Change `mappingPolicy` to
`dnd.location_occurrences.extract_mapping.v2`.
8. Remove the legacy ID-bearing registry `PromptInput` and its model-projection
digest once the extractor uses operation grounding. Update the occurrence
checkpoint to combine `IdentityDigest()` with the new grounding projection
digest, prompt/schema fingerprint, mapping policy, and its existing inputs;
do not retain dead compatibility aliases.
digest once the extractor uses operation grounding. Keep the occurrence's
static module fingerprint based on `IdentityDigest()`, prompt/schema
fingerprints, and mapping policy. The operation-scoped projection is already
covered by source/chunk identity and configured or generated reference
dependencies, while its `LLMInputMaterial.Digest` identifies the exact model
input; do not add a second digest API or operation-aware static fingerprint.
9. Update focused schema, prompt, extractor, canonicalization, checkpoint, and
generated-reference tests. Cover unique-name empty selectors, successful
same-name selection, failure for an unsupported ambiguous mention,

View File

@@ -10,26 +10,34 @@ import (
"gitea.maximumdirect.net/eric/notarius/internal/modules/dnd/shared"
)
type orderedItemOccurrenceResponse struct {
value itemOccurrenceResponse
type orderedItemOccurrence struct {
value dnd.ItemOccurrence
earliest int
hasEvidence bool
}
func canonicalizeResponse(response *extractionResponse, order shared.SourceRefOrder, sourceID string, registry *itemregistry.Registry) error {
if response == nil {
return nil
func canonicalItemOccurrenceList(response extractionResponse, order shared.SourceRefOrder, sourceID string, registry *itemregistry.Registry) (dnd.ItemOccurrenceList, error) {
if response.Occurrences == nil {
return dnd.ItemOccurrenceList{}, nil
}
ordered := make([]orderedItemOccurrenceResponse, len(response.Occurrences))
for index := range response.Occurrences {
canonical, found := registry.Lookup(response.Occurrences[index].Name)
ordered := make([]orderedItemOccurrence, len(response.Occurrences))
for index, occurrence := range response.Occurrences {
item, found := registry.Lookup(occurrence.Name)
if !found {
return fmt.Errorf("occurrences[%d].name is not in the item registry", index)
return dnd.ItemOccurrenceList{}, fmt.Errorf("occurrences[%d].name is not in the item registry", index)
}
response.Occurrences[index].Name = canonical.Name
earliest, hasEvidence := canonicalizeItemOccurrence(&response.Occurrences[index], order, sourceID)
ordered[index] = orderedItemOccurrenceResponse{
value: response.Occurrences[index],
refs := order.Canonicalize(itemOccurrenceSourceRefs(occurrence.SourceRefs, sourceID))
earliest, hasEvidence := order.EarliestValid(refs)
ordered[index] = orderedItemOccurrence{
value: dnd.ItemOccurrence{
ItemID: item.ID,
Name: item.Name,
Kind: dnd.ItemOccurrenceKind(occurrence.Kind),
Quantity: cloneQuantity(occurrence.Quantity),
From: occurrence.From,
To: occurrence.To,
SourceRefs: refs,
},
earliest: earliest,
hasEvidence: hasEvidence,
}
@@ -43,40 +51,9 @@ func canonicalizeResponse(response *extractionResponse, order shared.SourceRefOr
}
return ordered[left].earliest < ordered[right].earliest
})
for index := range ordered {
response.Occurrences[index] = ordered[index].value
}
return nil
}
func canonicalizeItemOccurrence(occurrence *itemOccurrenceResponse, order shared.SourceRefOrder, sourceID string) (int, bool) {
if occurrence == nil {
return 0, false
}
refs := order.Canonicalize(itemOccurrenceSourceRefs(occurrence.SourceRefs, sourceID))
occurrence.SourceRefs = itemOccurrenceResponseRefs(refs)
return order.EarliestValid(refs)
}
func canonicalItemOccurrenceList(response extractionResponse, sourceID string, registry *itemregistry.Registry) (dnd.ItemOccurrenceList, error) {
if response.Occurrences == nil {
return dnd.ItemOccurrenceList{}, nil
}
occurrences := make([]dnd.ItemOccurrence, 0, len(response.Occurrences))
for index, occurrence := range response.Occurrences {
item, found := registry.Lookup(occurrence.Name)
if !found {
return dnd.ItemOccurrenceList{}, fmt.Errorf("occurrences[%d].name is not in the item registry", index)
}
occurrences = append(occurrences, dnd.ItemOccurrence{
ItemID: item.ID,
Name: item.Name,
Kind: dnd.ItemOccurrenceKind(occurrence.Kind),
Quantity: cloneQuantity(occurrence.Quantity),
From: occurrence.From,
To: occurrence.To,
SourceRefs: itemOccurrenceSourceRefs(occurrence.SourceRefs, sourceID),
})
occurrences := make([]dnd.ItemOccurrence, len(ordered))
for index, occurrence := range ordered {
occurrences[index] = occurrence.value
}
return dnd.ItemOccurrenceList{Occurrences: occurrences}, nil
}
@@ -92,17 +69,6 @@ func itemOccurrenceSourceRefs(refs []itemOccurrenceSourceRefResponse, sourceID s
return values
}
func itemOccurrenceResponseRefs(refs []source.SourceRef) []itemOccurrenceSourceRefResponse {
if refs == nil {
return nil
}
values := make([]itemOccurrenceSourceRefResponse, len(refs))
for index, ref := range refs {
values[index] = itemOccurrenceSourceRefResponse{StartSegment: ref.StartUnitID, EndSegment: ref.EndUnitID}
}
return values
}
func cloneQuantity(value *int) *int {
if value == nil {
return nil

View File

@@ -161,12 +161,9 @@ func (e *Extractor) Extract(ctx context.Context, req contracts.TypedExtractionRe
}, &response); err != nil {
return contracts.TypedExtractionResult[dnd.ItemOccurrenceList]{}, extractorErrorf("complete structured output: %w", err)
}
if err := canonicalizeResponse(&response, order, req.Source.ID, registry); err != nil {
return contracts.TypedExtractionResult[dnd.ItemOccurrenceList]{}, extractorErrorf("map item occurrence response: %w", err)
}
value, err := canonicalItemOccurrenceList(response, req.Source.ID, registry)
value, err := canonicalItemOccurrenceList(response, order, req.Source.ID, registry)
if err != nil {
return contracts.TypedExtractionResult[dnd.ItemOccurrenceList]{}, extractorErrorf("resolve canonical item occurrence names: %w", err)
return contracts.TypedExtractionResult[dnd.ItemOccurrenceList]{}, extractorErrorf("map item occurrence response: %w", err)
}
return contracts.TypedExtractionResult[dnd.ItemOccurrenceList]{Value: value}, nil
}

View File

@@ -36,7 +36,6 @@ type ContextUnit struct {
// Its prompt input contains no durable IDs or source IDs.
type Grounding struct {
promptInput contracts.LLMInputMaterial
projectionDigest string
locationsBySelector map[string]dnd.Location
}
@@ -104,7 +103,6 @@ func NewGrounding(registry *Registry, doc *source.SourceDocument) (*Grounding, e
digest := semanticDigest(content)
return &Grounding{
promptInput: contracts.NewLLMInputMaterial(ReferenceSlot, locationcodec.MediaType, content, digest, ""),
projectionDigest: digest,
locationsBySelector: locationsBySelector,
}, nil
}
@@ -118,14 +116,6 @@ func (g *Grounding) PromptInput() contracts.LLMInputMaterial {
return g.promptInput.Clone()
}
// ProjectionDigest returns the digest of the exact contextual prompt input.
func (g *Grounding) ProjectionDigest() string {
if g == nil {
return ""
}
return g.projectionDigest
}
// Resolve maps a contextual selector to one canonical location.
func (g *Grounding) Resolve(selector Selector) (dnd.Location, bool) {
if g == nil {

View File

@@ -216,8 +216,8 @@ func TestGroundingProjectsAndResolvesUniqueAndSameNameLocations(t *testing.T) {
if bytes.Contains(grounding.PromptInput().Content, []byte("location:sha256:")) || bytes.Contains(grounding.PromptInput().Content, []byte(`"source_id"`)) {
t.Fatalf("grounding projection exposed durable identity: %s", grounding.PromptInput().Content)
}
if grounding.ProjectionDigest() == "" || grounding.ProjectionDigest() == registry.IdentityDigest() || grounding.PromptInput().Digest != grounding.ProjectionDigest() {
t.Fatalf("grounding/identity digests = %q/%q", grounding.ProjectionDigest(), registry.IdentityDigest())
if grounding.PromptInput().Digest == "" || grounding.PromptInput().Digest == registry.IdentityDigest() {
t.Fatalf("grounding/identity digests = %q/%q", grounding.PromptInput().Digest, registry.IdentityDigest())
}
resolved, ok := grounding.Resolve(Selector{Name: " the tavern ", RegistryRefs: []RegistryRef{{StartUnitID: 30, EndUnitID: 30}, {StartUnitID: 20, EndUnitID: 20}}})
@@ -276,8 +276,8 @@ func TestGroundingHandlesEmptyRegistriesAndIdentityFingerprints(t *testing.T) {
if err != nil {
t.Fatal(err)
}
if base.IdentityDigest() != expanded.IdentityDigest() || baseGrounding.ProjectionDigest() != expandedGrounding.ProjectionDigest() {
t.Fatalf("identity/model fingerprints = %q/%q and %q/%q", base.IdentityDigest(), expanded.IdentityDigest(), baseGrounding.ProjectionDigest(), expandedGrounding.ProjectionDigest())
if base.IdentityDigest() != expanded.IdentityDigest() || baseGrounding.PromptInput().Digest != expandedGrounding.PromptInput().Digest {
t.Fatalf("identity/model fingerprints = %q/%q and %q/%q", base.IdentityDigest(), expanded.IdentityDigest(), baseGrounding.PromptInput().Digest, expandedGrounding.PromptInput().Digest)
}
}