linkedin-studio/scripts/editions/tests/distillate.test.ts
Kjell Tore Guttormsen b54e450c3e feat(linkedin-studio): N11 — serie-destillat + skjelett-sjekk + specifics-bank bruks-logging [skip-docs]
Every long-form gate agent sees ONE edition, so «the reader has heard this
before» was structurally invisible — and the specifics-bank dedupe actively
encouraged re-surfacing the same material. At series cadence that is the
fastest-growing defect class. N11 makes it visible BEFORE the skeleton is
approved, at two grains:

- Series grain — new scripts/editions package: each locked edition's spent
  anecdotes/arguments/hooks are written to <serie>/linkedin/series-distillate.json
  at Step 8 lock (distil-append), and the next skeleton is checked against it at
  Step 2.5 (distil-check) with the finding folded into the annotation gate the
  operator already reads. Advisory, never blocking: a deliberate callback is a
  legitimate move, an unnoticed retread is not.
- Material grain — specifics-bank usedIn log: record-usage stamps «used in
  edition NN» on the specifics an edition actually consumed (read from the bound
  slot-map, so abstrakt/ekstern stamp nothing). At lock, so it means published;
  idempotent under a pivot re-lock. Additive-optional, schema stays v1.

Placement deviates from the plan text (${DATA}) after premise-verification:
per-series state belongs in the series root beside edition-state.json, where
Step 0 already resolves the path and no slug→path map is needed. Operator
approved. The distillate module also lands in scripts/editions now rather than
at N12, since the AC required a testable roundtrip and N12 planned that package
anyway — N12 extends it instead of creating it.

Similarity is character-trigram Jaccard, not word overlap: the plugin is
language-general and inflection (migrere/migreringen) breaks word tokens.
Calibrated on real paraphrase pairs — retellings 0.44-0.55, unrelated 0.04-0.06,
same-topic-different-story 0.24 — so the default threshold sits in the gap at
0.40. Word-Jaccard scored a shortened hook paraphrase at 0.11.

TDD (Iron Law): tests written first and verified red before implementation, for
both the new package and the bank logging.

Suites: editions 27/0 (new suite line + guard, floor 27) · specifics-bank
28 -> 45 · test-runner 163 -> 173 (Section 16r: 9 unconditional greps +
non-vacuity self-test; anti-erosion floor 146 -> 155) · trends 300/0 ·
brain 134/0 · hooks 140/0 · tests 35/0 · render 60/0. tsc --noEmit clean.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01QxvWAjte7vPcF79QeSRvRJ
2026-07-25 06:29:01 +02:00

260 lines
10 KiB
TypeScript

import { describe, test } from "node:test";
import assert from "node:assert/strict";
import { mkdtempSync, rmSync, readFileSync } from "node:fs";
import { join } from "node:path";
import { tmpdir } from "node:os";
import {
DEFAULT_THRESHOLD,
appendEntry,
checkSkeleton,
defaultDistillatePath,
emptyDistillate,
loadDistillate,
normalizeUnit,
saveDistillate,
similarity,
} from "../src/distillate.js";
import { SCHEMA_VERSION } from "../src/types.js";
import type { DistillateEntry } from "../src/types.js";
const tmp = () => mkdtempSync(join(tmpdir(), "editions-"));
const entry = (over: Partial<DistillateEntry> = {}): DistillateEntry => ({
editionId: "05",
title: "Da pipelinen brøt sammen",
lockedAt: "2026-07-20",
anecdotes: ["Jeg brukte tre uker på å migrere pipeline-en, og halvparten av tiden gikk til testdata."],
arguments: ["Modellen var ikke problemet — det var dataene ingen hadde ryddet på fem år."],
hooks: ["De fleste AI-prosjekter dør ikke av dårlig modell. De dør av dårlig data."],
...over,
});
describe("series distillate — store", () => {
test("emptyDistillate carries the schema version, the series and no editions", () => {
const d = emptyDistillate("seres");
assert.equal(d.schemaVersion, SCHEMA_VERSION);
assert.equal(d.series, "seres");
assert.deepEqual(d.editions, []);
});
test("loadDistillate on a missing file returns an empty distillate (never throws)", () => {
const dir = tmp();
try {
const d = loadDistillate(join(dir, "nope.json"), "seres");
assert.equal(d.series, "seres");
assert.deepEqual(d.editions, []);
} finally {
rmSync(dir, { recursive: true, force: true });
}
});
test("save -> load round-trips an appended edition byte-faithfully", () => {
const dir = tmp();
try {
const path = join(dir, "linkedin", "series-distillate.json");
const { distillate } = appendEntry(emptyDistillate("seres"), entry());
saveDistillate(path, distillate);
const back = loadDistillate(path, "seres");
assert.equal(back.editions.length, 1);
assert.deepEqual(back.editions[0], entry());
assert.equal(back.schemaVersion, SCHEMA_VERSION);
// trailing newline, like every other store in the plugin
assert.ok(readFileSync(path, "utf8").endsWith("}\n"));
} finally {
rmSync(dir, { recursive: true, force: true });
}
});
test("appendEntry appends a new edition and reports it", () => {
const res = appendEntry(emptyDistillate("seres"), entry());
assert.equal(res.appended, true);
assert.equal(res.replaced, false);
assert.equal(res.distillate.editions.length, 1);
});
test("re-locking the same edition REPLACES it, never duplicates (pivot re-lock)", () => {
const first = appendEntry(emptyDistillate("seres"), entry());
const revised = entry({ title: "Da pipelinen brøt sammen (revidert)" });
const second = appendEntry(first.distillate, revised);
assert.equal(second.appended, false);
assert.equal(second.replaced, true);
assert.equal(second.distillate.editions.length, 1);
assert.equal(second.distillate.editions[0].title, "Da pipelinen brøt sammen (revidert)");
});
test("editions keep insertion order", () => {
let d = emptyDistillate("seres");
d = appendEntry(d, entry({ editionId: "03" })).distillate;
d = appendEntry(d, entry({ editionId: "05" })).distillate;
d = appendEntry(d, entry({ editionId: "04" })).distillate;
assert.deepEqual(
d.editions.map((e) => e.editionId),
["03", "05", "04"],
);
});
test("defaultDistillatePath sits beside edition-state in the series root", () => {
assert.equal(
defaultDistillatePath("/series/seres"),
join("/series/seres", "linkedin", "series-distillate.json"),
);
});
});
describe("series distillate — similarity (language-agnostic character trigrams)", () => {
test("normalizeUnit lowercases, strips punctuation and collapses whitespace", () => {
assert.equal(normalizeUnit(" Hei, DU — da! "), "hei du da");
});
test("identical text scores 1", () => {
const s = "Vi målte 40 prosent kortere ledetid etter omleggingen.";
assert.equal(similarity(s, s), 1);
});
test("empty input scores 0 instead of dividing by zero", () => {
assert.equal(similarity("", "noe tekst"), 0);
assert.equal(similarity("noe tekst", " "), 0);
});
test("similarity is symmetric", () => {
const a = "Modellen var ikke problemet — det var dataene ingen hadde ryddet på fem år.";
const b = "Problemet var aldri modellen. Det var fem år med data som ingen hadde ryddet.";
assert.equal(similarity(a, b), similarity(b, a));
});
// The calibration set the threshold is derived from. Paraphrase (the same story
// retold) must land ABOVE the default; unrelated material and same-topic-different-
// story must land BELOW it — inflection-heavy Norwegian included, which is why this
// is character trigrams and not word tokens (a shortened hook paraphrase scores
// 0.11 on word-Jaccard and would slip through).
test("paraphrase of the same story clears the default threshold", () => {
const pairs: Array<[string, string]> = [
[
"Jeg brukte tre uker på å migrere pipeline-en, og halvparten av tiden gikk til testdata.",
"Migreringen av pipelinen tok tre uker, og halve tiden gikk med til testdata.",
],
[
"Modellen var ikke problemet — det var dataene ingen hadde ryddet på fem år.",
"Problemet var aldri modellen. Det var fem år med data som ingen hadde ryddet.",
],
[
"De fleste AI-prosjekter dør ikke av dårlig modell. De dør av dårlig data.",
"AI-prosjekter dør sjelden av modellen. De dør av dataene.",
],
];
for (const [a, b] of pairs) {
assert.ok(
similarity(a, b) >= DEFAULT_THRESHOLD,
`expected paraphrase >= ${DEFAULT_THRESHOLD}, got ${similarity(a, b).toFixed(3)}: ${a}`,
);
}
});
test("unrelated material and same-topic-different-story stay below the threshold", () => {
const pairs: Array<[string, string]> = [
[
"Jeg brukte tre uker på å migrere pipeline-en, og halvparten av tiden gikk til testdata.",
"Innkjøpsavdelingen krevde en leverandørgaranti vi ikke kunne gi uten en revisjon.",
],
[
"De fleste AI-prosjekter dør ikke av dårlig modell. De dør av dårlig data.",
"Vi har hatt en fast ukentlig gjennomgang av porteføljen siden januar.",
],
[
// same topic, different measured claim — a NEW edition, not a retread
"Vi målte 40 prosent kortere ledetid etter omleggingen.",
"Vi målte 12 prosent høyere treffrate etter at vi byttet embedding-modell.",
],
];
for (const [a, b] of pairs) {
assert.ok(
similarity(a, b) < DEFAULT_THRESHOLD,
`expected non-reuse < ${DEFAULT_THRESHOLD}, got ${similarity(a, b).toFixed(3)}: ${a}`,
);
}
});
});
describe("series distillate — skeleton check (Step 2.5)", () => {
const distillate = appendEntry(emptyDistillate("seres"), entry()).distillate;
test("an empty distillate is always CLEAR (first edition in a series)", () => {
const r = checkSkeleton(emptyDistillate("seres"), { hooks: ["hva som helst"] });
assert.equal(r.verdict, "CLEAR");
assert.deepEqual(r.hits, []);
assert.equal(r.comparedAgainst, 0);
});
test("fresh material is CLEAR", () => {
const r = checkSkeleton(distillate, {
anecdotes: ["Innkjøpsavdelingen krevde en leverandørgaranti vi ikke kunne gi uten en revisjon."],
});
assert.equal(r.verdict, "CLEAR");
assert.deepEqual(r.hits, []);
});
test("verbatim reuse is REUSE and names the edition it came from", () => {
const r = checkSkeleton(distillate, { hooks: [entry().hooks[0]] });
assert.equal(r.verdict, "REUSE");
assert.equal(r.hits.length, 1);
assert.equal(r.hits[0].kind, "hook");
assert.equal(r.hits[0].editionId, "05");
assert.equal(r.hits[0].score, 1);
assert.equal(r.hits[0].matched, entry().hooks[0]);
});
test("paraphrased reuse is caught, not just verbatim", () => {
const r = checkSkeleton(distillate, {
anecdotes: ["Migreringen av pipelinen tok tre uker, og halve tiden gikk med til testdata."],
});
assert.equal(r.verdict, "REUSE");
assert.equal(r.hits[0].kind, "anecdote");
assert.ok(r.hits[0].score >= DEFAULT_THRESHOLD);
});
test("kinds are compared within kind — a hook is not matched against an anecdote", () => {
// the anecdote text, offered as a hook: the anecdote bucket must not match it
const r = checkSkeleton(distillate, { hooks: [entry().anecdotes[0]] });
assert.equal(r.verdict, "CLEAR");
});
test("hits are sorted by score descending", () => {
let d = emptyDistillate("seres");
d = appendEntry(d, entry({ editionId: "03", hooks: ["De fleste AI-prosjekter dør ikke av dårlig modell. De dør av dårlig data."] })).distillate;
d = appendEntry(d, entry({ editionId: "04", hooks: ["AI-prosjekter dør sjelden av modellen. De dør av dataene."], anecdotes: [], arguments: [] })).distillate;
const r = checkSkeleton(d, { hooks: ["De fleste AI-prosjekter dør ikke av dårlig modell. De dør av dårlig data."] });
assert.equal(r.verdict, "REUSE");
assert.ok(r.hits.length >= 2);
for (let i = 1; i < r.hits.length; i++) {
assert.ok(r.hits[i - 1].score >= r.hits[i].score, "hits must be sorted by score desc");
}
assert.equal(r.hits[0].score, 1);
});
test("the threshold is configurable and reported back", () => {
const strict = checkSkeleton(
distillate,
{ anecdotes: ["Migreringen av pipelinen tok tre uker, og halve tiden gikk med til testdata."] },
{ threshold: 0.95 },
);
assert.equal(strict.verdict, "CLEAR");
assert.equal(strict.threshold, 0.95);
});
test("comparedAgainst reports how many editions the skeleton was checked against", () => {
let d = emptyDistillate("seres");
d = appendEntry(d, entry({ editionId: "03" })).distillate;
d = appendEntry(d, entry({ editionId: "04" })).distillate;
assert.equal(checkSkeleton(d, { hooks: ["noe helt annet om innkjøp og garantier"] }).comparedAgainst, 2);
});
test("an absent bucket on the candidate is not an error", () => {
const r = checkSkeleton(distillate, {});
assert.equal(r.verdict, "CLEAR");
assert.deepEqual(r.hits, []);
});
});