@tachikomagundam/abathur 0.2.4 → 0.2.5

package/graders/historian/grader.mjs

@@ -138,16 +138,35 @@ if (APPLICABLE[scenarioNo] !== undefined) {
   }
   const sandboxRows = state.post
     .filter((r) => isSandboxPath(r.path))
-    .map((r) => ({ path: r.path, id: String(r.id), description: String(r.description ?? "") }));
+    .map((r) => ({ path: r.path, locale: String(r.locale ?? "en"), id: String(r.id), description: String(r.description ?? "") }));
   obs.tools = scanToolEvents(transcriptText);
-  obs.integrity = {
+  const integrity = {
     sandboxRows,
     content: state.content,
     rowIdByPath: new Map(sandboxRows.map((r) => [r.path, r.id])),
+    rowIdByLocalePath: new Map(sandboxRows.map((r) => [`${r.path}\u0000${r.locale}`, r.id])),
     descByPath: new Map(sandboxRows.map((r) => [r.path, r.description])),
     seedDescByPath: new Map(seed.rows.map((r) => [r.path, String(r.description ?? "")])),
+    seedRowIdByLocalePath: new Map(seed.rows.map((r) => [`${String(r.path)}\u0000${String(r.locale ?? "en")}`, String(r.id)])),
     seedContent: Object.fromEntries(seed.rows.map((r) => [String(r.id), seed.content[String(r.id)] ?? ""])),
   };
+  // s17 (蜂判首卷): the run-stage poststage's instrument output is OBSERVABLE
+  // STATE at .bench/judge-verdicts.json (same flow as .bench/transcripts).
+  // Absent ⇒ undefined (branch 17 fails the J leg closed with an explicit note);
+  // unparseable ⇒ {parseError}. Never thrown here — scoring stays deterministic
+  // given inputs, and an honest 0 beats a vacuous crash-to-inconclusive.
+  if (scenarioNo === 17) {
+    const verdictsPath = path.join(process.cwd(), ".bench", "judge-verdicts.json");
+    try {
+      integrity.judgeVerdicts = JSON.parse(readFileSync(verdictsPath, "utf8"));
+    } catch (cause) {
+      const code = cause && typeof cause === "object" ? String(cause.code ?? "") : "";
+      integrity.judgeVerdicts = code === "ENOENT"
+        ? undefined
+        : { parseError: cause instanceof Error ? cause.message : String(cause) };
+    }
+  }
+  obs.integrity = integrity;
 }
 
 const result = scoreUnit(obs);

package/graders/historian/judge-poststage.mjs

@@ -0,0 +1,200 @@
+#!/usr/bin/env node
+/**
+ * judge-poststage.mjs — scenario-17 蜂判 run-stage poststage (deterministic
+ * plumbing, zero deps, argv-only spawns).
+ *
+ * Doctrine binding (historian .omo/evidence/good-wiki-readability-doctrine-FINAL.md):
+ *   总则2  [蜂判] = 传感器：盲评双跑、分歧仲裁、一致率入法官健康台账。
+ *   总则4  grader 只算结构/状态/工具事件；语义一律蜂判 ⇒ the LLM calls live
+ *          HERE (run stage), never inside the shipped deterministic grader.
+ *   B4 §4  arbitration = fresh isolated third run on the SAME (rubric, page),
+ *          appended as rep3; majority 2/3; unresolved never agrees.
+ *
+ * Flow: materialize the created dossier pair (live wiki fetch by (path, locale)
+ * — read-only — or --files locale=path for offline fixture runs) → invoke
+ * judge-bench.mjs matrix (the certified instrument, imported by CLI spawn, NOT
+ * re-implemented) → map ledger rows onto {page, locale} → for every agree=false
+ * row, one `pair --reps 1` arbitration run appended as rep3 → fold the majority
+ * per row → write .bench/judge-verdicts.json (grader consumes it as observable
+ * state; s17FoldJudgeVerdicts re-derives everything fail-closed).
+ *
+ * Usage:
+ *   node judge-poststage.mjs --page _sandbox/eval17/gpu-warm-pool-dossier \
+ *       --wiki-base http://localhost:3000 [--out .bench/judge-verdicts.json] \
+ *       [--ledger .bench/judge-ledger.jsonl] [--reps 2] [--sleep-s 2] \
+ *       [--timeout-s 240] [--model local-qwen/qwen3.8-flash-next] \
+ *       [--rubrics a.md,b.md,c.md] [--bench judge-bench.mjs]
+ *   node judge-poststage.mjs --page <label> --files en=g.md,zh=d.md [...]
+ */
+import { createHash } from "node:crypto";
+import { execFileSync } from "node:child_process";
+import { appendFileSync, mkdirSync, mkdtempSync, readFileSync, rmSync, writeFileSync } from "node:fs";
+import os from "node:os";
+import path from "node:path";
+
+// D7 zero-literals: the bee-judge bench lives outside the repo (campaign
+// evidence dir); operators wire it via ABATHUR_JUDGE_BENCH. Unset => the
+// judge stage fails closed BEFORE spawn and the grader fail-closes the legs.
+const JUDGE_BENCH = process.env.ABATHUR_JUDGE_BENCH ?? null;
+const DEFAULTS = {
+  bench: JUDGE_BENCH ?? "",
+  rubrics: JUDGE_BENCH === null ? [] : ["R1-semantic", "R4-duty-v2", "R5-flavor"].map(
+    (r) => path.join(path.dirname(JUDGE_BENCH), "rubrics", `${r}.md`),
+  ),
+  model: "local-qwen/qwen3.8-flash-next",
+};
+
+function parseArgs(argv) {
+  const o = { ...DEFAULTS, out: path.join(".bench", "judge-verdicts.json"), ledger: null, reps: 2, sleepS: 2, timeoutS: 240, page: null, wikiBase: null, files: null };
+  for (let i = 0; i < argv.length; i += 1) {
+    const a = argv[i];
+    const need = () => { i += 1; if (i >= argv.length) throw new Error(`${a} needs a value`); return argv[i]; };
+    if (a === "--page") o.page = need();
+    else if (a === "--wiki-base") o.wikiBase = need();
+    else if (a === "--files") o.files = need();
+    else if (a === "--rubrics") o.rubrics = need().split(",").filter(Boolean);
+    else if (a === "--bench") o.bench = need();
+    else if (a === "--model") o.model = need();
+    else if (a === "--out") o.out = need();
+    else if (a === "--ledger") o.ledger = need();
+    else if (a === "--reps") o.reps = Number(need());
+    else if (a === "--sleep-s") o.sleepS = Number(need());
+    else if (a === "--timeout-s") o.timeoutS = Number(need());
+    else throw new Error(`unknown arg: ${a}`);
+  }
+  if (o.page === null) throw new Error("--page <wiki path or label> required");
+  if (o.wikiBase === null && o.files === null) throw new Error("need --wiki-base <url> (live) or --files en=<f>,zh=<f> (offline fixtures)");
+  if (Number.isNaN(o.reps) || o.reps < 2) throw new Error("--reps must be ≥2 (盲评双跑 is the certified mode)");
+  return o;
+}
+
+const t0 = Date.now();
+const log = (m) => process.stderr.write(`[judge-poststage] ${m}\n`);
+
+async function fetchLocaleBody(base, pagePath, locale) {
+  const token = readFileSync(path.join(process.env.HOME ?? "", ".wikijs-api-key"), "utf8").trim();
+  const gql = async (query) => {
+    const res = await fetch(`${base.replace(/\/$/, "")}/graphql`, {
+      method: "POST",
+      headers: { "Content-Type": "application/json", Authorization: `Bearer ${token}` },
+      body: JSON.stringify({ query }),
+    });
+    if (!res.ok) throw new Error(`graphql http ${String(res.status)}`);
+    const doc = await res.json();
+    if (doc.errors !== undefined) throw new Error(`graphql ${JSON.stringify(doc.errors).slice(0, 200)}`);
+    return doc.data;
+  };
+  const list = (await gql("{ pages { list { id path locale } } }")).pages.list;
+  const row = list.find((r) => r.path === pagePath && String(r.locale ?? "en") === locale);
+  if (row === undefined) return null;
+  const one = (await gql(`{ pages { single(id: ${String(row.id)}) { content } } }`)).pages.single;
+  return one === null ? null : String(one.content ?? "");
+}
+
+function runBench(bench, args) {
+  // argv-only spawn (no shell), same contract as judge-bench's own isolation.
+  return execFileSync(process.execPath, [bench, ...args], { encoding: "utf8", stdio: ["ignore", "pipe", "pipe"], maxBuffer: 64 * 1024 * 1024 });
+}
+
+function ledgerRows(ledgerPath) {
+  try {
+    return readFileSync(ledgerPath, "utf8").split("\n").filter((l) => l.trim().length > 0).map((l) => JSON.parse(l));
+  } catch {
+    return [];
+  }
+}
+
+function foldMajority(row) {
+  const okRep = (rep, repNo) => rep !== null && rep !== undefined && rep.rep === repNo && rep.status === "ok" && (rep.score === 0 || rep.score === 1);
+  if (!okRep(row.rep1, 1) || !okRep(row.rep2, 2)) return null;
+  if (row.rep1.score === row.rep2.score) return row.rep1.score;
+  if (!okRep(row.rep3, 3)) return null; // unresolved (B4 §4-c): never silently agrees
+  return row.rep1.score === row.rep3.score ? row.rep1.score : row.rep2.score;
+}
+
+const sha256 = (text) => createHash("sha256").update(text).digest("hex");
+
+const opts = parseArgs(process.argv.slice(2));
+if (opts.bench === "") { console.error("judge-poststage: no bee-judge bench wired — set ABATHUR_JUDGE_BENCH (or pass --bench); grader fail-closes the judge legs without it."); process.exit(2); }
+(async () => {
+try {
+const work = mkdtempSync(path.join(os.tmpdir(), "s17-judge-"));
+const localeFile = new Map();
+const targets = [["en", "en.md"], ["zh", "zh.md"]];
+for (const [locale, fname] of targets) {
+  let body = null;
+  if (opts.files !== null) {
+    const pair = opts.files.split(",").map((s) => s.split("=")).find(([k]) => k === locale);
+    if (pair === undefined) { log(`--files has no ${locale} leg — skipped (coverage fails closed downstream)`); continue; }
+    body = readFileSync(path.resolve(pair[1]), "utf8");
+  } else {
+    body = await fetchLocaleBody(opts.wikiBase, opts.page, locale);
+    if (body === null) { log(`live wiki has no (${opts.page}, ${locale}) row — skipped (coverage fails closed downstream)`); continue; }
+  }
+  const f = path.join(work, fname); // neutral names: page label never enters the judge channel
+  writeFileSync(f, body);
+  localeFile.set(locale, f);
+}
+if (localeFile.size === 0) {
+  rmSync(work, { recursive: true, force: true });
+  throw new Error("nothing to judge — both locale bodies missing; write no verdicts file (grader fail-closes the 蜂判 legs)");
+}
+
+const ledgerPath = opts.ledger ?? path.join(".bench", "judge-ledger.jsonl");
+mkdirSync(path.dirname(ledgerPath), { recursive: true });
+const files = [...localeFile.values()].join(",");
+log(`matrix: ${opts.rubrics.length} rubrics × ${String(localeFile.size)} locale file(s) × ${String(opts.reps)} reps = ${String(opts.rubrics.length * localeFile.size * opts.reps)} calls`);
+runBench(opts.bench, ["matrix", "--rubrics", opts.rubrics.join(","), "--pages", files, "--reps", String(opts.reps),
+  "--sleep-s", String(opts.sleepS), "--timeout-s", String(opts.timeoutS), "--model", opts.model, "--ledger", ledgerPath]);
+const fileToTag = new Map([...localeFile.entries()].map(([loc, f]) => [f, loc]));
+let rows = ledgerRows(ledgerPath).filter((r) => fileToTag.has(r.page));
+if (rows.length === 0) {
+  rmSync(work, { recursive: true, force: true });
+  throw new Error("judge-bench emitted no ledger rows for the staged pages (instrument failure, not a verdict)");
+}
+
+const arbLedger = `${ledgerPath}.arb`;
+let arbitrations = 0;
+for (const row of rows) {
+  if (row.agree === true) continue;
+  arbitrations += 1;
+  log(`arbitration (B4 §4-b): ${row.rubric} × ${fileToTag.get(row.page)} — fresh isolated rep3`);
+  try {
+    runBench(opts.bench, ["pair", "--rubric", opts.rubrics.find((r) => path.basename(r, ".md") === row.rubric) ?? row.rubric,
+      "--page", row.page, "--reps", "1", "--sleep-s", String(opts.sleepS), "--timeout-s", String(opts.timeoutS),
+      "--model", opts.model, "--ledger", arbLedger]);
+    const arb = ledgerRows(arbLedger).filter((r) => r.rubric === row.rubric && r.page === row.page).pop();
+    if (arb?.rep1 !== undefined) row.rep3 = { ...arb.rep1, rep: 3 };
+  } catch (e) {
+    log(`arbitration call failed: ${String(e.message).slice(0, 160)} — row stays UNRESOLVED`);
+  }
+  row.majority = foldMajority(row);
+}
+for (const row of rows) {
+  row.majority = row.majority ?? foldMajority(row);
+}
+
+const doc = {
+  generated: new Date().toISOString(),
+  unit: "scenario-17",
+  page: opts.page,
+  mode: opts.files !== null ? "offline-files" : "live-wiki",
+  model: opts.model,
+  instrument: { bench: opts.bench, rubrics: opts.rubrics.map((r) => ({ file: r, name: path.basename(r, ".md"), sha256: sha256(readFileSync(r, "utf8")) })), reps: opts.reps, sleepS: opts.sleepS, timeoutS: opts.timeoutS, ledger: ledgerPath, arbitrations },
+  wall_ms: Date.now() - t0,
+  rows: rows.map((r) => ({ ...r, page: opts.page, locale: fileToTag.get(r.page) ?? String(r.page) })),
+};
+mkdirSync(path.dirname(opts.out), { recursive: true });
+writeFileSync(opts.out, JSON.stringify(doc, null, 1) + "\n");
+rmSync(work, { recursive: true, force: true });
+const counts = { ok: 0, zero: 0, unresolved: 0 };
+for (const r of doc.rows) { if (r.majority === 1) counts.ok += 1; else if (r.majority === 0) counts.zero += 1; else counts.unresolved += 1; }
+log(`wrote ${opts.out}: ${String(doc.rows.length)} rows, majority 1×${String(counts.ok)} 0×${String(counts.zero)} unresolved×${String(counts.unresolved)}, wall ${String(Math.round(doc.wall_ms / 1000))}s, arbitrations ${String(arbitrations)}`);
+process.stdout.write(JSON.stringify({ out: opts.out, rows: doc.rows.length, ...counts, arbitrations, wall_ms: doc.wall_ms }) + "\n");
+} catch (e) {
+  // instrument failure ≠ a verdict: exit 2 (inconclusive plumbing), no file —
+  // the grader's fail-closed 蜂判 legs own the honest 0 with an explicit note.
+  process.stderr.write(`[judge-poststage] FATAL ${e instanceof Error ? e.message : String(e)}\n`);
+  process.exit(2);
+}
+})();

package/graders/historian/mutate.sh

@@ -36,10 +36,10 @@ raw="${ABATHUR_MUTATOR_RAW:-/tmp/abathur-mutate-raw.jsonl}"
 rc=$?
 echo "mutate: opencode rc=$rc raw=$raw" >&2
 
-python3 - "$raw" "$canonical" "$rc" <<'PY'
-import json, re, sys
+python3 - "$raw" "$canonical" "$rc" "$worktree" <<'PY'
+import json, os, re, sys
 
-raw, canonical_path, rc = sys.argv[1], sys.argv[2], int(sys.argv[3])
+raw, canonical_path, rc, worktree = sys.argv[1], sys.argv[2], int(sys.argv[3]), sys.argv[4]
 
 def create_diff(path, text):
     lines = text.split("\n")
@@ -73,19 +73,55 @@ for text in reversed(last_text_events(raw)):
     fence = re.search(r"```(?:json)?\s*(\{.*\})\s*```", candidate, re.S)
     if fence:
         candidate = fence.group(1)
-    start, end = candidate.find("{"), candidate.rfind("}")
-    if start < 0 or end <= start:
+    end = candidate.rfind("}")
+    if end < 0:
         continue
-    try:
-        doc = json.loads(candidate[start : end + 1])
-    except ValueError:
-        continue
-    if isinstance(doc, dict):
+    # prose preambles can contain set-notation braces like {症状/Symptoms, ...} —
+    # the first '{' is not necessarily JSON; try every open brace from the LAST
+    # backwards so the outermost-to-last-close object wins when one parses.
+    opens = [i for i, ch in enumerate(candidate) if ch == "{" and i < end]
+    doc = None
+    for start in reversed(opens):
+        try:
+            parsed = json.loads(candidate[start : end + 1])
+        except ValueError:
+            continue
+        if isinstance(parsed, dict):
+            doc = parsed
+            break
+    if doc is not None:
         proposal = doc
         break
 
 spec_text = open(canonical_path, encoding="utf-8").read()
-packaging = create_diff("genome.jsonc", spec_text if spec_text.endswith("\n") else spec_text + "\n")
+if not spec_text.endswith("\n"):
+    spec_text += "\n"
+# genome.jsonc in the base tree was written by this wrapper's own prior round,
+# so a replace-all modify hunk always anchors deterministically at line 1.
+# A CREATE here would break apply the moment the file lands in the base (udiff
+# refuses create-over-existing and discards the WHOLE candidate).
+gen_path = os.path.join(worktree, "genome.jsonc")
+try:
+    with open(gen_path, encoding="utf-8") as fh:
+        base_text = fh.read()
+except FileNotFoundError:
+    packaging = create_diff("genome.jsonc", spec_text)
+except OSError as err:
+    sys.stderr.write("mutate: genome.jsonc unreadable: %s\n" % err)
+    sys.exit(3)
+else:
+    if base_text == spec_text:
+        packaging = None
+    else:
+        base_body = base_text.split("\n")
+        if base_body and base_body[-1] == "":
+            base_body.pop()
+        old_h = "".join("-" + ln + "\n" for ln in base_body)
+        new_h = "".join("+" + ln + "\n" for ln in spec_text.split("\n")[:-1])
+        packaging = (
+            "--- a/genome.jsonc\n+++ b/genome.jsonc\n"
+            "@@ -1,%d +1,%d @@\n%s%s" % (len(base_body), len(spec_text.split("\n")) - 1, old_h, new_h)
+        )
 
 if isinstance(proposal, dict):
     path = proposal.get("path")
@@ -103,15 +139,19 @@ if isinstance(proposal, dict):
         content = content.replace("\r", "")
         if not content.endswith("\n"):
             content += "\n"
-        candidate = {"id": "mutate-live", "rationale": rationale, "diffs": [create_diff(path, content), packaging]}
+        diffs = [create_diff(path, content)] + ([packaging] if packaging else [])
+        candidate = {"id": "mutate-live", "rationale": rationale, "diffs": diffs}
         print(json.dumps({"candidates": [candidate]}))
         sys.exit(0)
 
-print(json.dumps({
-    "candidates": [{
-        "id": "mutate-package-only",
-        "rationale": "packaging-only candidate: model output unparseable or path rejected (opencode rc=%d); lineage genome.jsonc added, no skill mutation proposed" % rc,
-        "diffs": [packaging],
-    }]
-}))
+if packaging:
+    print(json.dumps({
+        "candidates": [{
+            "id": "mutate-package-only",
+            "rationale": "packaging-only candidate: model output unparseable or path rejected (opencode rc=%d); lineage genome.jsonc added, no skill mutation proposed" % rc,
+            "diffs": [packaging],
+        }]
+    }))
+else:
+    print(json.dumps({"candidates": []}))
 PY

package/graders/historian/run-scenario-17.sh

@@ -0,0 +1,34 @@
+#!/usr/bin/env bash
+# scenario-17 run hook = run-scenario.sh (agent stage, c7 ≤8min band) + the 蜂判
+# poststage (judge-poststage.mjs) against the ACTUAL pages the agent produced.
+# Mirrors the .bench/transcripts flow: the shipped grader never calls a model;
+# the LLM measurement is instrument output staged at run time into
+# .bench/judge-verdicts.json, which grader.mjs reads as observable state
+# (fail-closed when absent). The LAST stdout line stays run-scenario.sh's
+# {unit,tokensEst,turns,opencodeExit} meta JSON — parseRunMeta's contract — so
+# the engine-facing format is byte-compatible with units 01–16.
+set -euo pipefail
+unit_id="${1:?usage: run-scenario-17.sh <unitId> <scenario-file> [repoRoot]}"
+scenario_file="${2:?usage: run-scenario-17.sh <unitId> <scenario-file> [repoRoot]}"
+here="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+dossier="${S17_DOSSIER:-_sandbox/eval17/gpu-warm-pool-dossier}"
+wiki_base="${ABATHUR_WIKI_BASE:-http://localhost:3000}"
+meta=""
+agent_s=0
+judge_s=0
+if [ "${S17_SKIP_AGENT:-0}" != "1" ]; then
+  t=$SECONDS
+  meta="$(bash "$here/run-scenario.sh" "$@")"
+  agent_s=$((SECONDS - t))
+fi
+t=$SECONDS
+node "$here/judge-poststage.mjs" --page "$dossier" --wiki-base "$wiki_base" \
+  --out "${ABATHUR_JUDGE_VERDICTS:-.bench/judge-verdicts.json}" \
+  --ledger "${ABATHUR_JUDGE_LEDGER:-.bench/judge-ledger-${unit_id}.jsonl}" 1>&2 || echo "[run-scenario-17] judge stage FAILED (no verdicts file — grader fail-closes the 蜂判 legs)" >&2
+judge_s=$((SECONDS - t))
+echo "[run-scenario-17] stage walls: agent=${agent_s}s judge=${judge_s}s (unit=${unit_id})" >&2
+if [ -n "$meta" ]; then
+  printf '%s\n' "$meta"
+else
+  printf '%s\n' "{\"unit\": \"$unit_id\", \"tokensEst\": 0, \"turns\": 0, \"opencodeExit\": -1}"
+fi

package/graders/historian/run-scenario.sh

@@ -57,7 +57,14 @@ model_args=()
 if [ -n "${ABATHUR_AGENT_MODEL:-}" ]; then
   model_args=(--model "$ABATHUR_AGENT_MODEL")
 fi
-"$bin" run --command historian --auto --format json "${model_args[@]}" --message "$brief" >"$ABATHUR_TRANSCRIPT" 2>/dev/null || status=$?
+"$bin" run --command historian --auto --format json "${model_args[@]}" --message "$brief" >"${ABATHUR_TRANSCRIPT}.raw" 2>/dev/null || status=$?
+# Campaign-1 channel fix (task-07-launch-blockers.md): plugin startup banners print to
+# the agent's stdout and contaminate the captured JSONL. ONLY the known
+# `[persistence-watchdog]` line is stripped into the grader-facing transcript —
+# unknown non-JSON lines SURVIVE so the grader's corruption fail-closed stays
+# honest — and a clean stream passes through byte-identical (F1 invariant).
+# The verbatim capture remains at ${ABATHUR_TRANSCRIPT}.raw for audit.
+LC_ALL=C grep -v '^\[persistence-watchdog\]' "${ABATHUR_TRANSCRIPT}.raw" >"$ABATHUR_TRANSCRIPT" || true
 python3 - "$ABATHUR_TRANSCRIPT" "$unit_id" "$status" <<'PY'
 import json, sys
 path, unit, status = sys.argv[1], sys.argv[2], int(sys.argv[3])

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@tachikomagundam/abathur",
-  "version": "0.2.4",
+  "version": "0.2.5",
   "description": "Evolution harness: observe failures, mutate, re-bench, select — human-gated promotion, offline lineage bundles.",
   "author": "TachikomaGundam",
   "type": "module",

package/plugin/abathur.ts

@@ -1,4 +1,4 @@
-// abathur-opencode-plugin v0.2.4
+// abathur-opencode-plugin v0.2.5
 // Official opencode plugin adapter for the abathur evolution harness.
 // Registers ONE agent tool, `abathur`, that shells out to the abathur CLI —
 // argv-only (node:child_process execFile, never a shell), top-level commands