package pipeline import ( "context" "encoding/json" "os" "path/filepath" "strings" "testing" ) // runner_memory_test.go: the шаг-4 acceptance — the memory bank v2 mechanism end-to-end // on a seeded glossary against the mock provider (injection → translate → post-check → // retrieval-state → resume $0 → resnapshot on an approved change). Per §8, acceptance is // "the mechanism is correct on a seeded glossary", not "≥98% on a real book". const suzukiSeed = ` terms: - src: 鈴木 dst: Судзуки type: name status: approved decl: { invariant: true, forms: ["Судзуки"] } ` // suzukiSource is one ja chunk that names 鈴木 (so the glossary fires). const suzukiSource = "鈴木は静かな図書館へ行った。" type wireMsg struct { Role string Content string } func bodyMessages(t *testing.T, body string) []wireMsg { t.Helper() var req struct { Messages []wireMsg `json:"messages"` } if err := json.Unmarshal([]byte(body), &req); err != nil { t.Fatalf("bad request body: %v\n%s", err, body) } return req.Messages } // translatorBody returns the recorded draft(translator) request body. func translatorBody(t *testing.T, rec *reqRec) string { t.Helper() for _, b := range rec.all() { if !isEditBody(b) { return b } } t.Fatal("no translator request recorded") return "" } func TestRunnerMemoryInjectionAndPostcheck(t *testing.T) { rec := &reqRec{} srv := newJSONProvider(rec, func(body string) (string, string) { if isEditBody(body) { return "ОТРЕДАКТИРОВАННЫЙ ПЕРЕВОД про Судзуки.", "stop" } return "Судзуки пошёл в тихую библиотеку.", "stop" // renders the approved dst }) defer srv.Close() bookPath := setupProjectOpts(t, srv.URL, projectOpts{regenerate: 1, source: suzukiSource, glossarySeed: suzukiSeed}) ctx := context.Background() r := newRunner(t, bookPath) res, err := r.TranslateBook(ctx) if err != nil { t.Fatal(err) } if res.Flagged != 0 { t.Fatalf("flagger default must not flag a correct render: flagged=%d", res.Flagged) } // The TRANSLATOR request carries the injection as its own system message between the // stable prefix and the user chunk; the dst appears only via injection (not in the ja // source), so its presence proves the injection landed on the wire. tb := translatorBody(t, rec) msgs := bodyMessages(t, tb) if len(msgs) != 3 { t.Fatalf("translator must have 3 messages (system, injection, user), got %d: %+v", len(msgs), msgs) } if msgs[1].Role != "system" || !strings.Contains(msgs[1].Content, "Судзуки") || !strings.Contains(msgs[1].Content, "ГЛОССАРИЙ") { t.Errorf("injection message wrong: %+v", msgs[1]) } if strings.Contains(msgs[2].Content, "ГЛОССАРИЙ") { t.Errorf("glossary leaked into the user (ch.Text) message: %q", msgs[2].Content) } // The EDITOR gets no injection in v1. for _, b := range rec.all() { if isEditBody(b) && strings.Contains(b, "ГЛОССАРИЙ") { t.Error("editor received a glossary injection (not in v1 scope)") } } // retrieval-state: one exact hit, post-check ran with 0 misses. rs, err := r.Store.GetRetrievalState("test-book", 1, 0) if err != nil || rs == nil { t.Fatalf("retrieval-state missing: %v", err) } if rs.NExactHits != 1 || rs.NPostcheckMiss != 0 { t.Errorf("retrieval-state wrong: exact=%d miss=%d", rs.NExactHits, rs.NPostcheckMiss) } callsRun1 := rec.count() r.Close() // Resume: no new provider calls, $0 (the deterministic memory path re-derives for free). r2 := newRunner(t, bookPath) defer r2.Close() res2, err := r2.TranslateBook(ctx) if err != nil { t.Fatal(err) } if rec.count() != callsRun1 || res2.TotalUSD != 0 { t.Fatalf("resume must be $0 with no new calls: calls=%d (run1=%d) usd=%v", rec.count(), callsRun1, res2.TotalUSD) } // retrieval-state re-derived identically. rs2, _ := r2.Store.GetRetrievalState("test-book", 1, 0) if rs2 == nil || rs2.NExactHits != 1 || rs2.NPostcheckMiss != 0 { t.Errorf("retrieval-state not stable on resume: %+v", rs2) } } func TestRunnerMemoryFlaggerRecordsMissWithoutFlagging(t *testing.T) { rec := &reqRec{} srv := newJSONProvider(rec, func(body string) (string, string) { if isEditBody(body) { return "ОТРЕДАКТИРОВАННЫЙ.", "stop" } return "Некто пошёл в библиотеку.", "stop" // DROPS the approved name → a post-check miss }) defer srv.Close() // Default flagger mode (postcheckGate: false). bookPath := setupProjectOpts(t, srv.URL, projectOpts{regenerate: 1, source: suzukiSource, glossarySeed: suzukiSeed}) r := newRunner(t, bookPath) defer r.Close() res, err := r.TranslateBook(context.Background()) if err != nil { t.Fatal(err) } // Flagger mode: the chunk is NOT flagged (translation still exported), but the miss IS // recorded (silent degradation converted to a loud, visible record). if res.Flagged != 0 { t.Errorf("flagger mode must not flag the chunk: flagged=%d", res.Flagged) } rs, _ := r.Store.GetRetrievalState("test-book", 1, 0) if rs == nil || rs.NPostcheckMiss != 1 { t.Fatalf("post-check miss not recorded: %+v", rs) } if !strings.Contains(rs.PostcheckDetail, "鈴木") { t.Errorf("post-check detail should name the missed src: %q", rs.PostcheckDetail) } } func TestRunnerMemoryPostcheckGateFlags(t *testing.T) { rec := &reqRec{} srv := newJSONProvider(rec, func(body string) (string, string) { if isEditBody(body) { return "ОТРЕДАКТИРОВАННЫЙ.", "stop" } return "Некто пошёл в библиотеку.", "stop" // drops the name }) defer srv.Close() // Opt-in hard gate. bookPath := setupProjectOpts(t, srv.URL, projectOpts{regenerate: 1, source: suzukiSource, glossarySeed: suzukiSeed, postcheckGate: true}) r := newRunner(t, bookPath) defer r.Close() res, err := r.TranslateBook(context.Background()) if err != nil { t.Fatal(err) } if res.Flagged != 1 { t.Fatalf("gate mode must flag the missed chunk: flagged=%d", res.Flagged) } if res.Chunks[0].FlagReason != FlagGlossaryMiss { t.Errorf("flag reason = %q, want glossary_miss", res.Chunks[0].FlagReason) } // The post-check runs on the FINAL exported output, so both stages ran (each ok in // chunk_status), but the CHUNK is flagged and its text is not exported. if res.Chunks[0].FinalText != "" { t.Errorf("a glossary-flagged chunk must not export text, got %q", res.Chunks[0].FinalText) } var editorRan bool for _, st := range res.Chunks[0].Stages { if st.Role == "editor" && st.Disposition == DispOK { editorRan = true } } if !editorRan { t.Error("editor stage should have run (post-check validates the final/editor output, not the draft)") } } func TestRunnerMemoryResnapshotOnApprovedChange(t *testing.T) { rec := &reqRec{} srv := newJSONProvider(rec, func(body string) (string, string) { if isEditBody(body) { return "ОТРЕДАКТИРОВАННЫЙ про Судзуки.", "stop" } return "Судзуки пошёл в библиотеку.", "stop" }) defer srv.Close() bookPath := setupProjectOpts(t, srv.URL, projectOpts{regenerate: 1, source: suzukiSource, glossarySeed: suzukiSeed}) ctx := context.Background() r1 := newRunner(t, bookPath) if _, err := r1.TranslateBook(ctx); err != nil { t.Fatal(err) } r1.Close() callsAfterRun1 := rec.count() // Change the APPROVED dst in the seed → the frozen materialization changes → the // snapshot changes → a resume without --resnapshot must fail loud (F1: no silent // divergent re-pay). seedPath := filepath.Join(filepath.Dir(bookPath), "glossary-seed.yaml") if err := os.WriteFile(seedPath, []byte(strings.Replace(suzukiSeed, "Судзуки", "Сузуки", -1)), 0o644); err != nil { t.Fatal(err) } r2 := newRunner(t, bookPath) _, err := r2.TranslateBook(ctx) r2.Close() if err == nil || !strings.Contains(err.Error(), "resnapshot") { t.Fatalf("changed approved glossary must fail loud mentioning --resnapshot, got: %v", err) } if rec.count() != callsAfterRun1 { t.Fatalf("denied resume must not call the provider: calls=%d (after run1=%d)", rec.count(), callsAfterRun1) } // With --resnapshot the book re-translates under the new approved glossary. r3 := newRunner(t, bookPath) defer r3.Close() r3.Resnapshot = true if _, err := r3.TranslateBook(ctx); err != nil { t.Fatal(err) } if rec.count() <= callsAfterRun1 { t.Fatal("resnapshot run must re-call the provider for the re-translation") } }