Migrate production modules to raw outputs

This commit is contained in:
2026-07-07 19:23:35 +00:00
parent cc6b050367
commit aa14faa3cb
8 changed files with 247 additions and 78 deletions

View File

@@ -4,6 +4,7 @@ import (
"context"
"encoding/json"
"fmt"
"strings"
"gitea.maximumdirect.net/eric/notarius/internal/framework/contracts"
"gitea.maximumdirect.net/eric/notarius/internal/framework/pipeline"
@@ -90,22 +91,24 @@ func (e *Extractor) Extract(ctx context.Context, req contracts.ExtractionRequest
}
var response extractionResponse
if _, err := req.LLMClient.CompleteStructured(ctx, contracts.StructuredCompletionRequest{
completion, err := req.LLMClient.CompleteStructured(ctx, contracts.StructuredCompletionRequest{
StageName: Key,
PromptID: PromptID,
PromptVersion: SchemaVersion,
ProfileID: req.LLMProfile,
SessionID: req.SessionID,
Inputs: dnd.PromptInputs(req.SourceInput, req.References),
}, &response); err != nil {
}, &response)
if err != nil {
return contracts.ExtractionResult{}, extractorErrorf("complete structured output: %w", err)
}
if response.SpellCasts == nil {
return contracts.ExtractionResult{}, extractorErrorf("malformed structured output: spell_casts must be present")
}
content, err := json.Marshal(response)
if err != nil {
return contracts.ExtractionResult{}, extractorErrorf("marshal raw output: %w", err)
content := append([]byte(nil), completion.Content...)
if len(strings.TrimSpace(string(content))) == 0 {
var err error
content, err = json.Marshal(response)
if err != nil {
return contracts.ExtractionResult{}, extractorErrorf("marshal raw output: %w", err)
}
}
return contracts.ExtractionResult{
Output: contracts.ExtractOutput{

View File

@@ -24,6 +24,7 @@ func TestExtractReturnsRawOutputFromStructuredResponse(t *testing.T) {
},
},
},
content: []byte(`{"spell_casts":[{"caster":" Aria ","spell":" Cure Wounds ","effect":" Heals an injured ally. ","narrative_description":" Aria restores the fighter after the fight. ","source_refs":[{"source_id":"session-alpha","start_unit_id":1,"end_unit_id":2}]}],"raw_marker":true}`),
}
result, err := New().Extract(context.Background(), extractionRequestWithClient(client))
@@ -58,6 +59,9 @@ func TestExtractReturnsRawOutputFromStructuredResponse(t *testing.T) {
if result.Output.Schema.ID != ResponseSchemaID || result.Output.Schema.Name != ResponseSchemaName || result.Output.Schema.Version != SchemaVersion {
t.Fatalf("schema = %#v, want response schema provenance", result.Output.Schema)
}
if got := string(result.Output.Payload.Content); got != string(client.content) {
t.Fatalf("content = %q, want exact raw completion content", got)
}
var payload extractionResponse
if err := json.Unmarshal(result.Output.Payload.Content, &payload); err != nil {
@@ -179,15 +183,15 @@ func TestExtractReturnsRawOutputForEmptyResponse(t *testing.T) {
}
}
func TestExtractRejectsMissingSpellCasts(t *testing.T) {
func TestExtractCarriesMalformedStructuredContentAsRawOutput(t *testing.T) {
client := &fakeSpellsLLMClient{response: extractionResponse{}}
_, err := New().Extract(context.Background(), extractionRequestWithClient(client))
if err == nil {
t.Fatal("Extract() error = nil, want malformed output error")
result, err := New().Extract(context.Background(), extractionRequestWithClient(client))
if err != nil {
t.Fatalf("Extract() error = %v, want nil", err)
}
if !strings.Contains(err.Error(), "dnd spells") || !strings.Contains(err.Error(), "spell_casts") {
t.Fatalf("Extract() error = %q, want spell_casts context", err.Error())
if string(result.Output.Payload.Content) != `{"spell_casts":null}` {
t.Fatalf("content = %s, want raw structured output", result.Output.Payload.Content)
}
}
@@ -329,6 +333,7 @@ func emptyChunkRequest(req contracts.ExtractionRequest) contracts.ExtractionRequ
type fakeSpellsLLMClient struct {
response extractionResponse
content []byte
err error
requests []contracts.StructuredCompletionRequest
}
@@ -344,9 +349,13 @@ func (client *fakeSpellsLLMClient) CompleteStructured(ctx context.Context, req c
return contracts.StructuredCompletionResponse{}, errors.New("unexpected output target")
}
*target = client.response
content, err := json.Marshal(client.response)
if err != nil {
return contracts.StructuredCompletionResponse{}, err
content := append([]byte(nil), client.content...)
if len(content) == 0 {
var err error
content, err = json.Marshal(client.response)
if err != nil {
return contracts.StructuredCompletionResponse{}, err
}
}
return contracts.StructuredCompletionResponse{Content: content}, nil
}

View File

@@ -279,7 +279,7 @@ func dndSpellsReferenceSet(party string, glossary string) contracts.ReferenceSet
return contracts.ReferenceSet{Slots: slots}
}
func TestRunnerFailsWhenDNDSpellsExtractorReturnsMalformedOutput(t *testing.T) {
func TestRunnerCarriesMalformedDNDSpellsExtractorOutput(t *testing.T) {
raw := readDNDSpellsFixture(t)
resolved := resolveDNDSpellsPipeline(t)
llmClient := &fakeSpellsLLMClient{response: extractionResponse{}}
@@ -289,16 +289,17 @@ func TestRunnerFailsWhenDNDSpellsExtractorReturnsMalformedOutput(t *testing.T) {
RawInput: raw,
LLMClient: llmClient,
})
if err == nil {
t.Fatal("Run() error = nil, want malformed extraction error")
if err != nil {
t.Fatalf("Run() error = %v, want nil", err)
}
if !strings.Contains(err.Error(), "extract lane") ||
!strings.Contains(err.Error(), "dnd spells") ||
!strings.Contains(err.Error(), "spell_casts") {
t.Fatalf("Run() error = %q, want D&D spells extraction context", err.Error())
if len(output.NormalizeOutputs) != 1 {
t.Fatalf("len(NormalizeOutputs) = %d, want raw output", len(output.NormalizeOutputs))
}
if output.Manifest.ValidationStatus != "failed" {
t.Fatalf("ValidationStatus = %q, want failed", output.Manifest.ValidationStatus)
if string(output.NormalizeOutputs[0].Payload.Content) != `{"spell_casts":null}` {
t.Fatalf("content = %s, want raw structured output", output.NormalizeOutputs[0].Payload.Content)
}
if output.Manifest.ValidationStatus != "approved" {
t.Fatalf("ValidationStatus = %q, want approved", output.Manifest.ValidationStatus)
}
}

View File

@@ -4,6 +4,9 @@ import (
"context"
"encoding/json"
"fmt"
"mime"
"sort"
"strings"
"gitea.maximumdirect.net/eric/notarius/internal/framework/contracts"
"gitea.maximumdirect.net/eric/notarius/internal/framework/pipeline"
@@ -34,20 +37,24 @@ func (m *Merger) Merge(ctx context.Context, req contracts.MergeRequest) (contrac
return contracts.MergeResult{}, mergerErrorf("context error before merge: %w", err)
}
if len(req.ExtractOutputs) == 1 {
payload := cloneRawPayload(req.ExtractOutputs[0].Payload)
outputs, err := orderedOutputs(req.ExtractOutputs)
if err != nil {
return contracts.MergeResult{}, err
}
if len(outputs) == 1 {
payload := cloneRawPayload(outputs[0].Payload)
return contracts.MergeResult{
Output: contracts.MergeOutput{
LaneID: req.LaneID,
MergerKey: Key,
SourceID: req.ExtractOutputs[0].SourceID,
Schema: req.ExtractOutputs[0].Schema,
SourceID: outputs[0].SourceID,
Schema: outputs[0].Schema,
Payload: payload,
},
}, nil
}
content, err := orderedContent(req.ExtractOutputs)
content, err := mergedContent(outputs)
if err != nil {
return contracts.MergeResult{}, err
}
@@ -55,7 +62,8 @@ func (m *Merger) Merge(ctx context.Context, req contracts.MergeRequest) (contrac
Output: contracts.MergeOutput{
LaneID: req.LaneID,
MergerKey: Key,
SourceID: sourceID(req.ExtractOutputs),
SourceID: sourceID(outputs),
Schema: commonSchema(outputs),
Payload: contracts.RawPayload{
Content: content,
MediaType: "application/json",
@@ -78,30 +86,94 @@ func Register(registry *pipeline.MergerRegistry) error {
})
}
func orderedContent(outputs []contracts.ExtractOutput) ([]byte, error) {
items := make([]map[string]any, 0, len(outputs))
func orderedOutputs(outputs []contracts.ExtractOutput) ([]contracts.ExtractOutput, error) {
ordered := make([]contracts.ExtractOutput, 0, len(outputs))
for _, output := range outputs {
item := map[string]any{
"chunk_id": output.ChunkID,
"chunk_index": output.ChunkIndex,
"media_type": output.Payload.MediaType,
if !isJSONMediaType(output.Payload.MediaType) {
return nil, mergerErrorf("extract output for chunk %q has unsupported media type %q", output.ChunkID, output.Payload.MediaType)
}
if json.Valid(output.Payload.Content) {
item["content"] = json.RawMessage(append([]byte(nil), output.Payload.Content...))
} else {
item["content"] = string(output.Payload.Content)
if !json.Valid(output.Payload.Content) {
return nil, mergerErrorf("extract output for chunk %q contains invalid JSON", output.ChunkID)
}
items = append(items, item)
ordered = append(ordered, cloneExtractOutput(output))
}
content, err := json.Marshal(struct {
Outputs []map[string]any `json:"outputs"`
}{Outputs: items})
sort.SliceStable(ordered, func(i, j int) bool {
return ordered[i].ChunkIndex < ordered[j].ChunkIndex
})
return ordered, nil
}
func mergedContent(outputs []contracts.ExtractOutput) ([]byte, error) {
values := make([]any, 0, len(outputs))
objects := make([]map[string]any, 0, len(outputs))
for _, output := range outputs {
var value any
if err := json.Unmarshal(output.Payload.Content, &value); err != nil {
return nil, mergerErrorf("decode extract output for chunk %q: %w", output.ChunkID, err)
}
values = append(values, value)
object, ok := value.(map[string]any)
if !ok {
continue
}
objects = append(objects, object)
}
if len(objects) == len(outputs) {
if field, ok := commonArrayField(objects); ok {
merged := make([]any, 0)
for _, object := range objects {
items := object[field].([]any)
merged = append(merged, items...)
}
return marshalMerged(map[string]any{field: merged})
}
}
return marshalMerged(values)
}
func commonArrayField(objects []map[string]any) (string, bool) {
if len(objects) == 0 {
return "", false
}
candidates := map[string]struct{}{}
for key, value := range objects[0] {
if _, ok := value.([]any); ok {
candidates[key] = struct{}{}
}
}
for _, object := range objects[1:] {
for key := range candidates {
if _, ok := object[key].([]any); !ok {
delete(candidates, key)
}
}
}
if len(candidates) != 1 {
return "", false
}
for key := range candidates {
return key, true
}
return "", false
}
func marshalMerged(value any) ([]byte, error) {
content, err := json.Marshal(value)
if err != nil {
return nil, mergerErrorf("encode merged output: %w", err)
}
return content, nil
}
func isJSONMediaType(mediaType string) bool {
base, _, err := mime.ParseMediaType(strings.TrimSpace(mediaType))
if err != nil {
base = strings.TrimSpace(mediaType)
}
return strings.EqualFold(base, "application/json")
}
func sourceID(outputs []contracts.ExtractOutput) string {
for _, output := range outputs {
if output.SourceID != "" {
@@ -111,6 +183,24 @@ func sourceID(outputs []contracts.ExtractOutput) string {
return ""
}
func commonSchema(outputs []contracts.ExtractOutput) contracts.ResponseSchema {
if len(outputs) == 0 {
return contracts.ResponseSchema{}
}
schema := outputs[0].Schema
for _, output := range outputs[1:] {
if output.Schema != schema {
return contracts.ResponseSchema{}
}
}
return schema
}
func cloneExtractOutput(output contracts.ExtractOutput) contracts.ExtractOutput {
output.Payload = cloneRawPayload(output.Payload)
return output
}
func cloneRawPayload(payload contracts.RawPayload) contracts.RawPayload {
return contracts.RawPayload{
Content: append([]byte(nil), payload.Content...),

View File

@@ -4,6 +4,7 @@ import (
"context"
"encoding/json"
"reflect"
"strings"
"testing"
"gitea.maximumdirect.net/eric/notarius/internal/framework/contracts"
@@ -77,12 +78,12 @@ func TestMergeDefensivelyCopiesRawPayload(t *testing.T) {
}
}
func TestMergeWrapsMultipleOutputsInChunkOrder(t *testing.T) {
func TestMergeConcatenatesCommonTopLevelArrayFieldInChunkOrder(t *testing.T) {
result, err := New().Merge(context.Background(), contracts.MergeRequest{
LaneID: "events",
ExtractOutputs: []contracts.ExtractOutput{
extractOutput("chunk-0", 0, `{"name":"first"}`),
extractOutput("chunk-1", 1, `{"name":"second"}`),
extractOutput("chunk-1", 1, `{"events":[{"name":"second"}]}`),
extractOutput("chunk-0", 0, `{"events":[{"name":"first"}]}`),
},
})
if err != nil {
@@ -93,27 +94,83 @@ func TestMergeWrapsMultipleOutputsInChunkOrder(t *testing.T) {
}
var decoded struct {
Outputs []struct {
ChunkID string `json:"chunk_id"`
ChunkIndex int `json:"chunk_index"`
Content json.RawMessage `json:"content"`
} `json:"outputs"`
Events []struct {
Name string `json:"name"`
} `json:"events"`
}
if err := json.Unmarshal(result.Output.Payload.Content, &decoded); err != nil {
t.Fatalf("Unmarshal() error = %v, want nil", err)
}
if len(decoded.Outputs) != 2 || decoded.Outputs[0].ChunkID != "chunk-0" || decoded.Outputs[1].ChunkID != "chunk-1" {
t.Fatalf("outputs = %#v, want chunk order", decoded.Outputs)
if len(decoded.Events) != 2 || decoded.Events[0].Name != "first" || decoded.Events[1].Name != "second" {
t.Fatalf("events = %#v, want concatenated chunk order", decoded.Events)
}
if result.Output.Schema.ID != "schema-id" {
t.Fatalf("schema = %#v, want common extract schema", result.Output.Schema)
}
}
func TestMergeHandlesEmptyInput(t *testing.T) {
result, err := New().Merge(context.Background(), contracts.MergeRequest{LaneID: "events"})
func TestMergeFallsBackToOrderedJSONValueArrayWhenShapesDiffer(t *testing.T) {
result, err := New().Merge(context.Background(), contracts.MergeRequest{
LaneID: "events",
ExtractOutputs: []contracts.ExtractOutput{
extractOutput("chunk-1", 1, `{"notes":["second"]}`),
extractOutput("chunk-0", 0, `{"events":[{"name":"first"}]}`),
},
})
if err != nil {
t.Fatalf("Merge() error = %v, want nil", err)
}
if string(result.Output.Payload.Content) != `{"outputs":[]}` {
t.Fatalf("content = %s, want empty outputs", result.Output.Payload.Content)
var decoded []map[string]any
if err := json.Unmarshal(result.Output.Payload.Content, &decoded); err != nil {
t.Fatalf("Unmarshal() error = %v, want nil", err)
}
if len(decoded) != 2 {
t.Fatalf("len(decoded) = %d, want 2", len(decoded))
}
if _, ok := decoded[0]["events"]; !ok {
t.Fatalf("decoded[0] = %#v, want first chunk value", decoded[0])
}
if _, ok := decoded[1]["notes"]; !ok {
t.Fatalf("decoded[1] = %#v, want second chunk value", decoded[1])
}
}
func TestMergeRejectsInvalidJSONAndNonJSONMediaTypes(t *testing.T) {
tests := []struct {
name string
output contracts.ExtractOutput
want string
}{
{
name: "invalid JSON",
output: extractOutput("chunk-0", 0, `{"events":[`),
want: "invalid JSON",
},
{
name: "non JSON media type",
output: func() contracts.ExtractOutput {
output := extractOutput("chunk-0", 0, `{"events":[]}`)
output.Payload.MediaType = "text/plain"
return output
}(),
want: "unsupported media type",
},
}
for _, test := range tests {
t.Run(test.name, func(t *testing.T) {
_, err := New().Merge(context.Background(), contracts.MergeRequest{
LaneID: "events",
ExtractOutputs: []contracts.ExtractOutput{test.output},
})
if err == nil {
t.Fatal("Merge() error = nil, want error")
}
if !strings.Contains(err.Error(), test.want) {
t.Fatalf("Merge() error = %q, want %q", err.Error(), test.want)
}
})
}
}