tickmarkr 1.43.0 → 1.44.0

package/dist/cli/commands/status.js

@@ -1,9 +1,10 @@
 import { BANNER } from "../../brand.js";
 import { blockedTasks, loadGraph, pendingTasks } from "../../graph/graph.js";
 import { GATE_NAMES } from "../../graph/schema.js";
-import { Journal } from "../../run/journal.js";
+import { Journal, journalComparable } from "../../run/journal.js";
 // ponytail: fixed 2s refresh; promote to config.visibility.* only when an operator asks.
 const REFRESH_MS = 2000;
+const NOT_COMPARABLE_NOTICE = "graph recompiled since this run — task states not comparable; run `tickmarkr run` to execute";
 // The timer must keep the process ALIVE: an unref'd timer here let the event loop drain after the
 // first frame, so a live `--watch` printed once and exited 0 (OBS-11). Never unref this.
 const defaultSleep = (ms) => new Promise((r) => setTimeout(r, ms));
@@ -40,6 +41,7 @@ const gateBox = (state, unicode) => {
         return state === "pass" ? "✓" : state === "fail" ? "✗" : state === "skip" ? "·" : "☐";
     return state === "pass" ? "[x]" : state === "fail" ? "[!]" : state === "skip" ? "." : "[ ]";
 };
+const defaultGateStates = (task) => GATE_NAMES.map((gate) => task.gates.includes(gate) ? "open" : "skip");
 const gateStates = (task, events) => {
     const outcomes = new Map();
     const start = attemptStartIdx(events, task.id);
@@ -121,18 +123,23 @@ const renderFrame = (cwd) => {
     let replayed = null;
     let events = [];
     const contexts = new Map();
+    let comparable = false;
     if (runId) {
         const j = Journal.open(cwd, runId);
         events = j.read();
-        replayed = j.replayStatuses();
-        for (const e of events) {
-            if (e.event === "task-dispatch" && e.taskId) {
-                const a = e.data.assignment;
-                if (typeof a.adapter === "string" && typeof a.model === "string")
-                    assignments.set(e.taskId, `${a.adapter}:${a.model}`);
-            }
-            if (e.event === "context-sample" && e.taskId && typeof e.data.tokens === "number" && Number.isFinite(e.data.tokens)) {
-                contexts.set(e.taskId, e.data.tokens); // last write wins
+        // OBS-52: refuse journal join when the run's recorded graphHash ≠ loaded graph — bare task-id replay lied after recompile.
+        comparable = journalComparable(events, g.spec.hash);
+        if (comparable) {
+            replayed = j.replayStatuses();
+            for (const e of events) {
+                if (e.event === "task-dispatch" && e.taskId) {
+                    const a = e.data.assignment;
+                    if (typeof a.adapter === "string" && typeof a.model === "string")
+                        assignments.set(e.taskId, `${a.adapter}:${a.model}`);
+                }
+                if (e.event === "context-sample" && e.taskId && typeof e.data.tokens === "number" && Number.isFinite(e.data.tokens)) {
+                    contexts.set(e.taskId, e.data.tokens); // last write wins
+                }
             }
         }
     }
@@ -148,7 +155,7 @@ const renderFrame = (cwd) => {
         const label = starved.has(t.id) ? " starved" : waiting.has(t.id) ? " dep-waiting" : "";
         const channel = assignments.get(t.id) ?? "-";
         const assignCol = contexts.has(t.id) ? `${channel}${divider}ctx ${contexts.get(t.id)}` : channel;
-        return { t, st, label, assignCol, states: gateStates(t, events) };
+        return { t, st, label, assignCol, states: comparable ? gateStates(t, events) : defaultGateStates(t) };
     });
     const statusColor = (st) => st === "done" ? 32 : st === "failed" ? 31 : st === "human" ? 33 : 36;
     if (!unicode) {
@@ -160,7 +167,7 @@ const renderFrame = (cwd) => {
             return `${prefix}${shortGoal(t.goal, Math.max(0, width - prefix.length - suffix.length))}${suffix}`;
         });
         const header = runId
-            ? `tickmarkr status${divider}run ${runId}${divider}${liveness(events).replaceAll(" · ", divider)}${divider}${done}/${g.tasks.length} done`
+            ? `tickmarkr status${divider}run ${runId}${!comparable ? `${divider}${NOT_COMPARABLE_NOTICE}` : ""}${divider}${liveness(events).replaceAll(" · ", divider)}${divider}${done}/${g.tasks.length} done`
             : `tickmarkr status${divider}no runs yet${divider}${done}/${g.tasks.length} done`;
         const legend = `  gates: ${GATE_NAMES.map((gate) => `${GATE_KEYS[gate]} ${gate}`).join(divider)}`;
         return [header, legend, ...rows].join("\n");
@@ -181,7 +188,9 @@ const renderFrame = (cwd) => {
         .replaceAll(" · ", dot);
     const tally = `${done}/${g.tasks.length} done`;
     const header = ` ${color("✓", 32, true)} ${color("tickmarkr", 1, true)}${dot}` +
-        (runId ? `run ${runId}${dot}${live}${dot}` : `no runs yet${dot}`) +
+        (runId
+            ? `run ${runId}${dot}${!comparable ? `${NOT_COMPARABLE_NOTICE}${dot}` : ""}${live}${dot}`
+            : `no runs yet${dot}`) +
         `${gauge} ${done === g.tasks.length && g.tasks.length > 0 ? color(tally, 32, true) : tally}`;
     const rule = dim("─".repeat(Math.min(width, 100)));
     const legend = dim(`   gates: ${GATE_NAMES.map((gate) => `${GATE_KEYS[gate]} ${gate}`).join(" · ")}`);

package/dist/compile/native.js

@@ -144,6 +144,15 @@ export function compileNative(file) {
     if (plainCount > 0) {
         console.warn(`tickmarkr: ${plainCount} acceptance item${plainCount === 1 ? "" : "s"} in ${file} ${plainCount === 1 ? "is a plain string" : "are plain strings"} — compiled as judge oracle${plainCount === 1 ? "" : "s"}. Prefix with command:/test:/judge: to make the oracle explicit.`);
     }
+    // OBS-51: semicolon-joined judge criteria invite intermittent clause-split verdicts — warn per item.
+    for (const draft of drafts) {
+        for (const item of draft.acceptance) {
+            const text = typeof item === "string" ? item : item.oracle === "judge" ? item.text : null;
+            if (text?.includes(";")) {
+                console.warn(`tickmarkr: OBS-51: task ${draft.id} judge criterion contains semicolon-joined clauses — split into separate acceptance items: ${JSON.stringify(text)}`);
+            }
+        }
+    }
     return result;
 }
 // Commented native spec written to tickmarkr.spec.md by `tickmarkr init`. Documented via HTML comments (which the

package/dist/gates/acceptance.js

@@ -93,12 +93,18 @@ Respond with ONLY this JSON (no prose before or after):
             meta: { unparseable: true, judge: channelKey({ adapter: judge.adapter.id, model: judge.model }) } };
     }
     const inconsistencies = [];
-    if (v.criteria.length !== judgeItems.length) {
-        inconsistencies.push(`judge verdict inconsistent: criteria count mismatch — expected ${judgeItems.length}, received ${v.criteria.length}`);
+    const expected = judgeItems.length;
+    const received = v.criteria.length;
+    // OBS-51: judges sometimes split one semicolon-joined criterion into multiple verdict items.
+    const clauseSplitAllPass = received > expected
+        && v.criteria.every((c) => c && typeof c === "object" && c.met === true);
+    if (received !== expected && !clauseSplitAllPass) {
+        inconsistencies.push(`judge verdict inconsistent: criteria count mismatch — expected ${expected}, received ${received}`);
     }
     v.criteria.forEach((c, i) => {
         if (!c || typeof c !== "object" || c.met !== true) {
-            inconsistencies.push(`judge verdict inconsistent: criteria[${i}].met must be true`);
+            const unmet = c && typeof c === "object" && c.criterion ? ` — unmet clause: ${c.criterion}` : "";
+            inconsistencies.push(`judge verdict inconsistent: criteria[${i}].met must be true${unmet}`);
         }
     });
     const pass = v.pass === true && inconsistencies.length === 0;

package/dist/run/daemon.js

@@ -87,7 +87,7 @@ export async function runDaemon(repoRoot, opts = {}) {
             baseRef = await gitHead(repoRoot);
             baseline = await captureBaseline(repoRoot, commands);
             writeFileSync(join(journal.dir, "baseline.json"), JSON.stringify(baseline, null, 2));
-            journal.append("run-start", undefined, { pid: process.pid, baseRef, commands, channels: channels.map(channelKey), branch }); // pid: v1.13 (VIS-11) liveness
+            journal.append("run-start", undefined, { pid: process.pid, baseRef, commands, channels: channels.map(channelKey), branch, graphHash: graph.spec.hash }); // graphHash: OBS-52 status join guard; pid: v1.13 (VIS-11) liveness
         }
         // T6: open the narrator AFTER run-start/run-resume is journaled so the watch surface has a run to
         // show. driver.narrator is undefined on subprocess → no-op (subprocess spawns nothing). Swallowed:

package/dist/run/journal.d.ts

@@ -62,6 +62,8 @@ export declare const TelemetryRowSchema: z.ZodObject<{
     }>>;
 }, z.core.$strip>;
 export type TelemetryRow = z.infer<typeof TelemetryRowSchema>;
+export declare function recordedGraphHash(events: JournalEvent[]): string | undefined;
+export declare function journalComparable(events: JournalEvent[], loadedHash: string): boolean;
 export declare function newRunId(now?: Date): string;
 export declare function readAllTelemetry(repoRoot: string, lastK: number, opts?: {
     after?: string;

package/dist/run/journal.js

@@ -72,6 +72,18 @@ export const TelemetryRowSchema = z.object({
     // v1.29 additive: mode of the attempt represented by this row. Absent on old telemetry.
     retryMode: z.enum(RETRY_MODES).optional(),
 });
+// OBS-52: graph hash on run-start binds status replay to the compiled graph that started the run.
+export function recordedGraphHash(events) {
+    for (const e of events) {
+        if (e.event === "run-start" && typeof e.data.graphHash === "string")
+            return e.data.graphHash;
+    }
+    return undefined;
+}
+export function journalComparable(events, loadedHash) {
+    const recorded = recordedGraphHash(events);
+    return recorded !== undefined && recorded === loadedHash;
+}
 export function newRunId(now = new Date()) {
     const p = (n, w = 2) => String(n).padStart(w, "0");
     return `run-${now.getFullYear()}${p(now.getMonth() + 1)}${p(now.getDate())}-${p(now.getHours())}${p(now.getMinutes())}${p(now.getSeconds())}`;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "tickmarkr",
-  "version": "1.43.0",
+  "version": "1.44.0",
   "description": "Spec in, verified work out.",
   "type": "module",
   "license": "MIT",