polymath-society 0.2.13 → 0.2.15

package/dist/cli.js

@@ -1194,6 +1194,11 @@ var init_cliAdapter = __esm({
 
 // src/grade/codexAdapter.ts
 import { spawn as spawn3 } from "child_process";
+function mapCodexTier(model, env = process.env) {
+  if (model && !/^(claude|haiku|sonnet|opus)/i.test(model)) return { model };
+  if (model && /^haiku/i.test(model)) return { model: env.POLYMATH_CODEX_MODEL_MINI || CODEX_MINI_MODEL };
+  return { model: env.POLYMATH_CODEX_MODEL || CODEX_DEFAULT_MODEL };
+}
 function composePrompt(inv, roots) {
   return [
     inv.systemPrompt ? `<system>
@@ -1204,12 +1209,14 @@ ${roots.map((r) => `  - ${r}`).join("\n")}` : "",
     inv.prompt
   ].filter(Boolean).join("\n\n");
 }
-var codexAdapter;
+var CODEX_DEFAULT_MODEL, CODEX_MINI_MODEL, codexAdapter;
 var init_codexAdapter = __esm({
   "src/grade/codexAdapter.ts"() {
     "use strict";
     init_adapter();
     init_backends();
+    CODEX_DEFAULT_MODEL = "gpt-5.5";
+    CODEX_MINI_MODEL = "gpt-5.4-mini";
     codexAdapter = {
       name: "codex",
       async run(inv) {
@@ -1227,7 +1234,7 @@ var init_codexAdapter = __esm({
           "read-only",
           "--json"
         ];
-        if (inv.model && !/^claude/i.test(inv.model)) args.push("--model", inv.model);
+        args.push("--model", mapCodexTier(inv.model).model);
         args.push(composePrompt(inv, roots));
         const started = Date.now();
         const toolCalls = [];
@@ -2213,7 +2220,7 @@ async function sayExportLinks(say, haveKinds) {
     say(`      ${bold(l.label)}${have ? green("  \u2713 already have one") : ""}  ${osc82(l.requestUrl, l.requestUrl)}`);
     say(`        ${l.buttonPath} \xB7 ${l.eta}`);
   }
-  say(`    ${dim("Run polymath-society again once a download lands \u2014 it finds the zip on its own.")}`);
+  say(`    Run polymath-society again once a download lands \u2014 it finds the zip on its own.`);
   say();
 }
 function describeExport(f, i, included, hadPrevious) {
@@ -2252,31 +2259,39 @@ async function runWizard() {
     }
   });
   const ask = async (q) => {
-    process.stderr.write(q);
+    say(q);
     if (lineQueue.length) {
       const l = lineQueue.shift();
-      process.stderr.write(l + "\n");
+      process.stderr.write(`> ${l}
+`);
       return l.trim();
     }
     if (ended) {
-      process.stderr.write("\n");
+      process.stderr.write("> \n");
       return "";
     }
+    rl.setPrompt("> ");
+    rl.prompt();
     return (await new Promise((res) => {
       waiting = res;
     })).trim();
   };
   const yes = async (q, def = true) => {
-    const a = (await ask(`${q} ${dim(def ? "(Y/n)" : "(y/N)")} `)).toLowerCase();
-    return a === "" ? def : a.startsWith("y");
+    for (; ; ) {
+      const a = (await ask(`${q} (y/n)`)).toLowerCase();
+      if (a.startsWith("y")) return true;
+      if (a.startsWith("n")) return false;
+      if (ended) return def;
+      say(`    Type y or n (Enter does nothing here).`);
+    }
   };
   const pick2 = async (defaults, hadPrevious) => {
     const hint = "(Enter = take everything into account \xB7 numbers to exclude, e.g. '2 4' \xB7 'all' = exclude everything)";
     for (; ; ) {
-      const a = await ask(`    Exclude any? ${dim(hint)} `);
+      const a = await ask(`    Exclude any? ${hint}`);
       const r = parsePicker(a, defaults);
       if (r) return r;
-      say(dim(`    Didn't understand "${a}" \u2014 numbers between 1 and ${defaults.length}, or all/none.`));
+      say(`    Didn't understand "${a}" \u2014 numbers between 1 and ${defaults.length}, or all/none.`);
     }
   };
   try {
@@ -4185,7 +4200,10 @@ async function getAccessToken(dataDir) {
       headers: { "content-type": "application/json", apikey: supabaseAnonKey },
       body: JSON.stringify({ refresh_token: id.refreshToken })
     });
-    if (!r.ok) return null;
+    if (!r.ok) {
+      if (r.status >= 400 && r.status < 500) await signOut(dataDir);
+      return null;
+    }
     const fresh = await persistSession(dataDir, await r.json());
     return { token: fresh.accessToken, identity: fresh };
   } catch {
@@ -4195,7 +4213,7 @@ async function getAccessToken(dataDir) {
 
 // ../../lib/calibration/index.ts
 var DEFAULT_CALIBRATION = {
-  version: "2026-07-15.1",
+  version: "2026-07-16.2",
   // = RUBRIC_FINGERPRINT of the shipped npm rubric (packages/coding-analyzer/
   // src/grade/criteria.ts). A unit test fails loud when the rubric changes
   // without this being updated (and re-pushed to the server).
@@ -4286,6 +4304,7 @@ var DEFAULT_CALIBRATION = {
   codingTiers: {
     push: { median: 300, sigma: 0.91, tag: "estimated \u2014 log-normal fit", source: "Anthropic 2026 Claude Code study (400k sessions): median engaged user ~8 prompts/day at ~35 words \u2014 ~300 directed words/day; tail: a 1-in-1,000 user AVERAGES ~5k directed words/day (near-daily heavy dictation, clearly below the ~10k single-day record). AVERAGE over substantial days, cumulative not spiky." },
     focus: { median: 0.08, sigma: 0.559, tag: "estimated \u2014 proxy-anchored", source: "median: Anthropic 20h/week engaged-user figure implies ~1h/day of true rapid exchange against an 8h working day; ceiling anchored to the ~4h/day deep-work limit (45% share = 1-in-1,000). AVERAGE share, cumulative not spiky." },
+    hours: { median: 3, sigma: 0.4, tag: "estimated \u2014 proxy-anchored", source: "median: RescueTime-measured ~2h48m of real productive time/day (same source as the flow benchmark) \u2248 3h of active building; shape: sustaining ~7.5h/day \u2248 top 1%, a ~12h/day average \u2248 1-in-3,000. AVERAGE work-day length, cumulative not spiky." },
     orchestration: { tag: "estimated \u2014 behavior bands", source: "no published concurrency distribution exists; ceiling anchored to the Claude Code lead's documented 10-15 parallel sessions" }
   },
   codingBenchmarks: {
@@ -23773,6 +23792,8 @@ async function compileCodingProfile() {
   const avgWords = aggregate?.throughput?.avgWordsPerDay ?? null;
   const pushPctl = lognPct(avgWords, tiers?.push);
   const focusPctl = lognPct(flowPctDay, tiers?.focus);
+  const workDayHrs = aggregate?.flow?.flow?.workDayHours ? Math.round(aggregate.flow.flow.workDayHours * 10) / 10 : null;
+  const hoursPctl = lognPct(workDayHrs, tiers?.hours);
   const hoursAt = (min) => Object.entries(pHist).reduce((a, [k, v]) => Number(k) >= min ? a + v : a, 0) / 60;
   const orchPctl = hoursAt(5) >= 5 || hoursAt(3) >= 10 && parallelism.queueOps >= 1e3 ? 99.5 : hoursAt(3) >= 10 ? 99 : hoursAt(2) >= 0.1 * (soloMin + parMin) / 60 ? 97 : hoursAt(2) >= 2 ? 90 : 70;
   const stackUp = [
@@ -23782,14 +23803,25 @@ async function compileCodingProfile() {
       `you peak at ~${peakWords.toLocaleString()} words of pure direction in a single day \u2014 the very top sustain ~${BEST.throughput.topAvgWordsPerDay.toLocaleString()} every day, week after week; the wildest single days on record push ~${BEST.throughput.topPeakWordsPerDay.toLocaleString()}`
     ),
     sRow(
-      "How well you use AI",
+      "Hours you put in",
+      null,
+      workDayHrs != null ? `~${workDayHrs}h of active building on a typical work day \u2014 engaged builders average ~3h, and sustaining 7h+ day after day, week after week, is top-1% territory` : "not enough dated activity yet to measure a typical work day"
+    ),
+    sRow(
+      "Your AI parallelism",
       ceilOf("delegation"),
       soloPct != null ? `~${soloPct}% of your hours you drive a single agent (${Math.round(soloMin / 60)}h solo vs ${Math.round(parMin / 60)}h with 2+ live) \u2014 the Claude Code lead keeps 10\u201315 going at once, so this is the single biggest speed-up still on the table` : "you brief the AI sharply, but mostly drive one agent at a time \u2014 the top run 10\u201315 in parallel, a several-fold speed-up"
     ),
+    // The evidence line cites the RANKING metric (flow as a share of the work
+    // day), never absolute hours — a friend with MORE flow hours over a longer
+    // day landed a WORSE percentile and the hours-based line read as a bug
+    // (2026-07-16: "he has more hours but less percentile??"). The label says
+    // "when you work" ON PURPOSE: this chip ranks quality of the hours, and
+    // the volume of hours is its own chip above.
     sRow(
-      "How well you focus",
+      "Focus quality when you work",
       focusScore,
-      flowHrsDay != null ? `only ~${flowHrsDay}h of your day is true deep flow \u2014 the best hold ~4h in one unbroken block; yours fragments across parallel sessions, so the gain is directing your attention at one thread for longer` : "your attention scatters across parallel sessions \u2014 directing it at one thread for longer is the unlock"
+      flowPctDay != null ? `~${Math.round(flowPctDay * 100)}% of your work day is true deep flow${flowHrsDay != null ? ` (~${flowHrsDay}h)` : ""} \u2014 the very best hold ~${Math.round(BEST.flow.topPctOfDay * 100)}% of theirs in one unbroken block; yours fragments across parallel sessions, so the gain is directing your attention at one thread for longer` : "your attention scatters across parallel sessions \u2014 directing it at one thread for longer is the unlock"
     ),
     // The four estimates LEAD with a concrete moment from their own sessions, then an
     // HONEST caveat — how many graded sessions back the read. No invented weakness, no
@@ -23822,12 +23854,17 @@ async function compileCodingProfile() {
       r.metric = avgWords;
       r.curve = "push";
     }
-    if (r.label === "How well you focus" && focusPctl != null) {
+    if (r.label === "Hours you put in" && hoursPctl != null) {
+      r.percentile = hoursPctl;
+      r.metric = workDayHrs;
+      r.curve = "hours";
+    }
+    if (r.label === "Focus quality when you work" && focusPctl != null) {
       r.percentile = focusPctl;
       r.metric = flowPctDay;
       r.curve = "focus";
     }
-    if (r.label === "How well you use AI") r.percentile = orchPctl;
+    if (r.label === "Your AI parallelism") r.percentile = orchPctl;
   }
   return { generatedAt: (/* @__PURE__ */ new Date()).toISOString(), gradedSessions: grades.length, agglomeration, outcomeCadence, dayDigest, dayChart, timeline, walkthroughs, rubric: rubric2, wordsByDay, dayThroughput, throughputCeiling: tCeiling, stackUp, criteria, parallelism, throughput, workstyle, flow, focus, expertiseMap, frontierArsenal, frontierDetail, coaching, gap, aggregate, delegationRollup, gapDist, projects };
 }
@@ -23933,7 +23970,7 @@ async function buildSkeleton(codingDir, stackUp) {
       ...thr.avgWordsPerDay ? [{ big: String(Math.round(thr.avgWordsPerDay).toLocaleString()), lab: "typed/dictated words per day, averaged (~300 for the median active user)" }] : [],
       ...thr.substantialDays ? [{ big: `${thr.substantialDays}/${thr.totalCodingDays}`, lab: "coding days were substantial (3h+ span)" }] : []
     ], moments: [] },
-    { key: "ai", label: "How they use AI", topPct: pct2("use ai"), sub: "orchestration, measured", claim: "", read: "", numbers: [
+    { key: "ai", label: "How they use AI", topPct: pct2("parallelism"), sub: "orchestration, measured", claim: "", read: "", numbers: [
       ...par.max ? [{ big: String(par.max), lab: "agents at once, peak" }] : [],
       ...par.hours2plus ? [{ big: `${par.hours2plus}h`, lab: "working 2+ live agents" }] : [],
       ...par.queueOps ? [{ big: par.queueOps.toLocaleString(), lab: "instructions queued ahead" }] : []
@@ -24353,12 +24390,15 @@ async function handlePublicCodingPageShare(dataDir, body) {
     return { status: 409, json: { error: "nothing to share yet \u2014 generate your page first" } };
   }
   const chatReport = body.includeChat === false ? null : await localChatReport();
+  const prior = await readOwnReportContent(cfg, auth.token, auth.identity.userId);
   try {
-    const payload = JSON.stringify({ format: "coding-pr1", page, ...chatReport ? { chatReport } : {} });
+    const merged = { ...prior, format: "coding-pr1", page };
+    if (chatReport) merged.chatReport = chatReport;
+    else delete merged.chatReport;
     const up = await fetch(`${cfg.supabaseUrl}/rest/v1/rpc/submit_report`, {
       method: "POST",
       headers: { apikey: cfg.supabaseAnonKey, authorization: `Bearer ${auth.token}`, "content-type": "application/json" },
-      body: JSON.stringify({ p_session_id: null, p_content: payload })
+      body: JSON.stringify({ p_session_id: null, p_content: JSON.stringify(merged) })
     });
     if (!up.ok) return { status: 502, json: { error: `share failed (HTTP ${up.status})` } };
     const verification = String(await up.json().catch(() => "unverified"));
@@ -24367,6 +24407,49 @@ async function handlePublicCodingPageShare(dataDir, body) {
     return { status: 502, json: { error: e.message.slice(0, 200) } };
   }
 }
+async function readOwnReportContent(cfg, token, userId) {
+  try {
+    const r = await fetch(`${cfg.supabaseUrl}/rest/v1/reports?user_id=eq.${encodeURIComponent(userId)}&select=content`, {
+      headers: { apikey: cfg.supabaseAnonKey, authorization: `Bearer ${token}` }
+    });
+    if (!r.ok) return {};
+    const rows = await r.json();
+    const c = rows[0]?.content;
+    return c && typeof c === "object" ? c : {};
+  } catch {
+    return {};
+  }
+}
+async function handleAnalysisShare(dataDir, kind, full) {
+  const cfg = centralConfig();
+  if (!cfg.configured) return { status: 503, json: { error: "central services are disabled in this build" } };
+  const auth = await getAccessToken(dataDir);
+  if (!auth) return { status: 401, json: { error: "sign in with Google (top right) first" } };
+  const slim = (p) => {
+    if (!p || typeof p !== "object") return p;
+    const { timeline: _t, walkthroughs: _w, dayDigest: _d, dayChart: _c, ...rest } = p;
+    return rest;
+  };
+  const value = kind === "coding" ? slim(full.coding()) : await full.chat().catch(() => null);
+  if (!value) return { status: 409, json: { error: "nothing to share yet \u2014 run the analysis first" } };
+  const prior = await readOwnReportContent(cfg, auth.token, auth.identity.userId);
+  try {
+    const merged = {
+      format: "coding-pr1",
+      ...prior,
+      [kind === "coding" ? "fullCoding" : "fullChat"]: value
+    };
+    const up = await fetch(`${cfg.supabaseUrl}/rest/v1/rpc/submit_report`, {
+      method: "POST",
+      headers: { apikey: cfg.supabaseAnonKey, authorization: `Bearer ${auth.token}`, "content-type": "application/json" },
+      body: JSON.stringify({ p_session_id: null, p_content: JSON.stringify(merged) })
+    });
+    if (!up.ok) return { status: 502, json: { error: `share failed (HTTP ${up.status})` } };
+    return { status: 200, json: { ok: true, kind } };
+  } catch (e) {
+    return { status: 502, json: { error: e.message.slice(0, 200) } };
+  }
+}
 
 // src/server.ts
 var __dirname = path33.dirname(fileURLToPath2(import.meta.url));
@@ -24694,6 +24777,18 @@ function startServer(opts) {
         json(res, out.status, out.json);
         return;
       }
+      if (req.method === "POST" && route === "/api/analysis/share") {
+        const body = JSON.parse(await readBody(req) || "{}");
+        const out = await handleAnalysisShare(dataDir, body.kind === "chat" ? "chat" : "coding", {
+          coding: () => liveProfile,
+          chat: async () => {
+            const p = await handlePerson();
+            return p.status === 200 ? p.json : null;
+          }
+        });
+        json(res, out.status, out.json);
+        return;
+      }
       if (route.startsWith("/api/")) {
         json(res, 404, { error: `this viewer doesn't serve ${route}` });
         return;
@@ -24715,7 +24810,7 @@ function startServer(opts) {
       serverUrl = `http://${host}:${port}`;
       void getCalibration().then((cal) => {
         if (cal.source === "central" && cal.doc.rubricVersion !== RUBRIC_FINGERPRINT2) {
-          console.warn(`[polymath-society] a newer grading rubric is published (${cal.doc.rubricVersion}; this install: ${RUBRIC_FINGERPRINT2}) \u2014 update the package and re-grade to stay comparable.`);
+          console.warn(`[polymath-society] you're not on the latest version \u2014 run: npm install -g polymath-society@latest`);
         }
       }).catch(() => {
       });
@@ -25046,15 +25141,19 @@ async function evalUpToDate(codingDir, threshold, regrade) {
       (await fs31.readdir(path36.join(codingDir, "grades")).catch(() => [])).filter((f) => f.endsWith(".json")).map((f) => f.slice(0, -5))
     );
     const agg = JSON.parse(await fs31.readFile(path36.join(codingDir, "aggregate.json"), "utf8").catch(() => "{}"));
-    return upToDateSkip({
-      gradableIds: gradable.map((s) => s.sessionId),
-      gradedIds: graded,
-      hasBigPicture: typeof agg.bigPicture === "string" && agg.bigPicture.trim().length > 0,
-      threshold,
-      regrade
-    });
+    const hasPriorRun = typeof agg.bigPicture === "string" && agg.bigPicture.trim().length > 0;
+    return {
+      ...upToDateSkip({
+        gradableIds: gradable.map((s) => s.sessionId),
+        gradedIds: graded,
+        hasBigPicture: hasPriorRun,
+        threshold,
+        regrade
+      }),
+      hasPriorRun
+    };
   } catch {
-    return { skip: false, pending: -1 };
+    return { skip: false, pending: -1, hasPriorRun: false };
   }
 }
 function startKeepAwake(o = {}) {
@@ -25163,6 +25262,10 @@ async function runPipeline(o) {
         o.onLine?.(
           `\u2713 report is up to date \u2014 ${gate2.pending} new session(s) since the last full run (below the ${threshold}-session threshold). Skipping the AI stages; free metrics still refresh. They'll be analyzed once ${threshold}+ accumulate (POLYMATH_MIN_NEW_SESSIONS=0 or --regrade to force now).`
         );
+      } else if (gate2.hasPriorRun) {
+        o.onLine?.(
+          `\u2713 your finished report from the last run is safe and already serving at the link below \u2014 nothing restarts from scratch. ${gate2.pending >= 0 ? `${gate2.pending} pending session(s)` : "New work"} plus any stages added by an update run on top of it, and sections refresh in place as they finish.`
+        );
       }
     }
   }

package/dist/engine/ingest-export.js

@@ -661,7 +661,7 @@ var GIFTS = [
 
 // ../../lib/calibration/index.ts
 var DEFAULT_CALIBRATION = {
-  version: "2026-07-15.1",
+  version: "2026-07-16.2",
   // = RUBRIC_FINGERPRINT of the shipped npm rubric (packages/coding-analyzer/
   // src/grade/criteria.ts). A unit test fails loud when the rubric changes
   // without this being updated (and re-pushed to the server).
@@ -752,6 +752,7 @@ var DEFAULT_CALIBRATION = {
   codingTiers: {
     push: { median: 300, sigma: 0.91, tag: "estimated \u2014 log-normal fit", source: "Anthropic 2026 Claude Code study (400k sessions): median engaged user ~8 prompts/day at ~35 words \u2014 ~300 directed words/day; tail: a 1-in-1,000 user AVERAGES ~5k directed words/day (near-daily heavy dictation, clearly below the ~10k single-day record). AVERAGE over substantial days, cumulative not spiky." },
     focus: { median: 0.08, sigma: 0.559, tag: "estimated \u2014 proxy-anchored", source: "median: Anthropic 20h/week engaged-user figure implies ~1h/day of true rapid exchange against an 8h working day; ceiling anchored to the ~4h/day deep-work limit (45% share = 1-in-1,000). AVERAGE share, cumulative not spiky." },
+    hours: { median: 3, sigma: 0.4, tag: "estimated \u2014 proxy-anchored", source: "median: RescueTime-measured ~2h48m of real productive time/day (same source as the flow benchmark) \u2248 3h of active building; shape: sustaining ~7.5h/day \u2248 top 1%, a ~12h/day average \u2248 1-in-3,000. AVERAGE work-day length, cumulative not spiky." },
     orchestration: { tag: "estimated \u2014 behavior bands", source: "no published concurrency distribution exists; ceiling anchored to the Claude Code lead's documented 10-15 parallel sessions" }
   },
   codingBenchmarks: {

package/dist/engine/public-report.js

@@ -15656,7 +15656,7 @@ var ENGINE_DATA = process.env.POLYMATH_DATA_DIR ? path10.resolve(process.env.POL
 
 // ../../lib/calibration/index.ts
 var DEFAULT_CALIBRATION = {
-  version: "2026-07-15.1",
+  version: "2026-07-16.2",
   // = RUBRIC_FINGERPRINT of the shipped npm rubric (packages/coding-analyzer/
   // src/grade/criteria.ts). A unit test fails loud when the rubric changes
   // without this being updated (and re-pushed to the server).
@@ -15747,6 +15747,7 @@ var DEFAULT_CALIBRATION = {
   codingTiers: {
     push: { median: 300, sigma: 0.91, tag: "estimated \u2014 log-normal fit", source: "Anthropic 2026 Claude Code study (400k sessions): median engaged user ~8 prompts/day at ~35 words \u2014 ~300 directed words/day; tail: a 1-in-1,000 user AVERAGES ~5k directed words/day (near-daily heavy dictation, clearly below the ~10k single-day record). AVERAGE over substantial days, cumulative not spiky." },
     focus: { median: 0.08, sigma: 0.559, tag: "estimated \u2014 proxy-anchored", source: "median: Anthropic 20h/week engaged-user figure implies ~1h/day of true rapid exchange against an 8h working day; ceiling anchored to the ~4h/day deep-work limit (45% share = 1-in-1,000). AVERAGE share, cumulative not spiky." },
+    hours: { median: 3, sigma: 0.4, tag: "estimated \u2014 proxy-anchored", source: "median: RescueTime-measured ~2h48m of real productive time/day (same source as the flow benchmark) \u2248 3h of active building; shape: sustaining ~7.5h/day \u2248 top 1%, a ~12h/day average \u2248 1-in-3,000. AVERAGE work-day length, cumulative not spiky." },
     orchestration: { tag: "estimated \u2014 behavior bands", source: "no published concurrency distribution exists; ceiling anchored to the Claude Code lead's documented 10-15 parallel sessions" }
   },
   codingBenchmarks: {

package/dist/engine/run-analysis.js

@@ -15768,7 +15768,7 @@ async function materializeCorpus(importedDir, docsDir, opts = {}) {
 // ../../lib/agents/engine/explore.ts
 import { promises as fs7 } from "fs";
 async function runCorpusAgent(opts) {
-  const run2 = await invokeAgent({
+  const run2 = await (opts.invoke ?? invokeAgent)({
     prompt: opts.prompt,
     systemPrompt: opts.systemPrompt,
     addDir: opts.docsDir,
@@ -15780,7 +15780,12 @@ async function runCorpusAgent(opts) {
     logPath: opts.logPath ?? opts.outFile?.replace(/\.json$/, ".toolcalls.log")
   });
   if (opts.outFile) {
-    const result = { generatedAt: (/* @__PURE__ */ new Date()).toISOString(), dimension: opts.dimension, raw: run2.json ?? run2.raw, costUsd: run2.costUsd };
+    if (!run2.json || typeof run2.json !== "object") {
+      const debris = { generatedAt: (/* @__PURE__ */ new Date()).toISOString(), dimension: opts.dimension, raw: run2.raw, costUsd: run2.costUsd };
+      await fs7.writeFile(opts.outFile.replace(/\.json$/, ".failed.json"), JSON.stringify(debris, null, 2), "utf-8");
+      throw new Error(`[${opts.dimension}] agent produced no JSON analysis \u2014 kept existing ${opts.outFile}, debris in .failed.json`);
+    }
+    const result = { generatedAt: (/* @__PURE__ */ new Date()).toISOString(), dimension: opts.dimension, raw: run2.json, costUsd: run2.costUsd };
     await fs7.writeFile(opts.outFile, JSON.stringify(result, null, 2), "utf-8");
   }
   return run2.json;

package/dist/index.js

@@ -2064,6 +2064,13 @@ var cliAdapter = {
 
 // src/grade/codexAdapter.ts
 import { spawn as spawn2 } from "child_process";
+var CODEX_DEFAULT_MODEL = "gpt-5.5";
+var CODEX_MINI_MODEL = "gpt-5.4-mini";
+function mapCodexTier(model, env = process.env) {
+  if (model && !/^(claude|haiku|sonnet|opus)/i.test(model)) return { model };
+  if (model && /^haiku/i.test(model)) return { model: env.POLYMATH_CODEX_MODEL_MINI || CODEX_MINI_MODEL };
+  return { model: env.POLYMATH_CODEX_MODEL || CODEX_DEFAULT_MODEL };
+}
 function composePrompt(inv, roots) {
   return [
     inv.systemPrompt ? `<system>
@@ -2091,7 +2098,7 @@ var codexAdapter = {
       "read-only",
       "--json"
     ];
-    if (inv.model && !/^claude/i.test(inv.model)) args.push("--model", inv.model);
+    args.push("--model", mapCodexTier(inv.model).model);
     args.push(composePrompt(inv, roots));
     const started = Date.now();
     const toolCalls = [];
@@ -2671,7 +2678,7 @@ var LocalGradeProvider = class {
 
 // ../../lib/calibration/index.ts
 var DEFAULT_CALIBRATION = {
-  version: "2026-07-15.1",
+  version: "2026-07-16.2",
   // = RUBRIC_FINGERPRINT of the shipped npm rubric (packages/coding-analyzer/
   // src/grade/criteria.ts). A unit test fails loud when the rubric changes
   // without this being updated (and re-pushed to the server).
@@ -2762,6 +2769,7 @@ var DEFAULT_CALIBRATION = {
   codingTiers: {
     push: { median: 300, sigma: 0.91, tag: "estimated \u2014 log-normal fit", source: "Anthropic 2026 Claude Code study (400k sessions): median engaged user ~8 prompts/day at ~35 words \u2014 ~300 directed words/day; tail: a 1-in-1,000 user AVERAGES ~5k directed words/day (near-daily heavy dictation, clearly below the ~10k single-day record). AVERAGE over substantial days, cumulative not spiky." },
     focus: { median: 0.08, sigma: 0.559, tag: "estimated \u2014 proxy-anchored", source: "median: Anthropic 20h/week engaged-user figure implies ~1h/day of true rapid exchange against an 8h working day; ceiling anchored to the ~4h/day deep-work limit (45% share = 1-in-1,000). AVERAGE share, cumulative not spiky." },
+    hours: { median: 3, sigma: 0.4, tag: "estimated \u2014 proxy-anchored", source: "median: RescueTime-measured ~2h48m of real productive time/day (same source as the flow benchmark) \u2248 3h of active building; shape: sustaining ~7.5h/day \u2248 top 1%, a ~12h/day average \u2248 1-in-3,000. AVERAGE work-day length, cumulative not spiky." },
     orchestration: { tag: "estimated \u2014 behavior bands", source: "no published concurrency distribution exists; ceiling anchored to the Claude Code lead's documented 10-15 parallel sessions" }
   },
   codingBenchmarks: {
@@ -3135,7 +3143,10 @@ async function getAccessToken(dataDir) {
       headers: { "content-type": "application/json", apikey: supabaseAnonKey },
       body: JSON.stringify({ refresh_token: id.refreshToken })
     });
-    if (!r.ok) return null;
+    if (!r.ok) {
+      if (r.status >= 400 && r.status < 500) await signOut(dataDir);
+      return null;
+    }
     const fresh = await persistSession(dataDir, await r.json());
     return { token: fresh.accessToken, identity: fresh };
   } catch {
@@ -22609,6 +22620,8 @@ async function compileCodingProfile() {
   const avgWords = aggregate?.throughput?.avgWordsPerDay ?? null;
   const pushPctl = lognPct(avgWords, tiers?.push);
   const focusPctl = lognPct(flowPctDay, tiers?.focus);
+  const workDayHrs = aggregate?.flow?.flow?.workDayHours ? Math.round(aggregate.flow.flow.workDayHours * 10) / 10 : null;
+  const hoursPctl = lognPct(workDayHrs, tiers?.hours);
   const hoursAt = (min) => Object.entries(pHist).reduce((a, [k, v]) => Number(k) >= min ? a + v : a, 0) / 60;
   const orchPctl = hoursAt(5) >= 5 || hoursAt(3) >= 10 && parallelism.queueOps >= 1e3 ? 99.5 : hoursAt(3) >= 10 ? 99 : hoursAt(2) >= 0.1 * (soloMin + parMin) / 60 ? 97 : hoursAt(2) >= 2 ? 90 : 70;
   const stackUp = [
@@ -22618,14 +22631,25 @@ async function compileCodingProfile() {
       `you peak at ~${peakWords.toLocaleString()} words of pure direction in a single day \u2014 the very top sustain ~${BEST.throughput.topAvgWordsPerDay.toLocaleString()} every day, week after week; the wildest single days on record push ~${BEST.throughput.topPeakWordsPerDay.toLocaleString()}`
     ),
     sRow(
-      "How well you use AI",
+      "Hours you put in",
+      null,
+      workDayHrs != null ? `~${workDayHrs}h of active building on a typical work day \u2014 engaged builders average ~3h, and sustaining 7h+ day after day, week after week, is top-1% territory` : "not enough dated activity yet to measure a typical work day"
+    ),
+    sRow(
+      "Your AI parallelism",
       ceilOf("delegation"),
       soloPct != null ? `~${soloPct}% of your hours you drive a single agent (${Math.round(soloMin / 60)}h solo vs ${Math.round(parMin / 60)}h with 2+ live) \u2014 the Claude Code lead keeps 10\u201315 going at once, so this is the single biggest speed-up still on the table` : "you brief the AI sharply, but mostly drive one agent at a time \u2014 the top run 10\u201315 in parallel, a several-fold speed-up"
     ),
+    // The evidence line cites the RANKING metric (flow as a share of the work
+    // day), never absolute hours — a friend with MORE flow hours over a longer
+    // day landed a WORSE percentile and the hours-based line read as a bug
+    // (2026-07-16: "he has more hours but less percentile??"). The label says
+    // "when you work" ON PURPOSE: this chip ranks quality of the hours, and
+    // the volume of hours is its own chip above.
     sRow(
-      "How well you focus",
+      "Focus quality when you work",
       focusScore,
-      flowHrsDay != null ? `only ~${flowHrsDay}h of your day is true deep flow \u2014 the best hold ~4h in one unbroken block; yours fragments across parallel sessions, so the gain is directing your attention at one thread for longer` : "your attention scatters across parallel sessions \u2014 directing it at one thread for longer is the unlock"
+      flowPctDay != null ? `~${Math.round(flowPctDay * 100)}% of your work day is true deep flow${flowHrsDay != null ? ` (~${flowHrsDay}h)` : ""} \u2014 the very best hold ~${Math.round(BEST.flow.topPctOfDay * 100)}% of theirs in one unbroken block; yours fragments across parallel sessions, so the gain is directing your attention at one thread for longer` : "your attention scatters across parallel sessions \u2014 directing it at one thread for longer is the unlock"
     ),
     // The four estimates LEAD with a concrete moment from their own sessions, then an
     // HONEST caveat — how many graded sessions back the read. No invented weakness, no
@@ -22658,12 +22682,17 @@ async function compileCodingProfile() {
       r.metric = avgWords;
       r.curve = "push";
     }
-    if (r.label === "How well you focus" && focusPctl != null) {
+    if (r.label === "Hours you put in" && hoursPctl != null) {
+      r.percentile = hoursPctl;
+      r.metric = workDayHrs;
+      r.curve = "hours";
+    }
+    if (r.label === "Focus quality when you work" && focusPctl != null) {
       r.percentile = focusPctl;
       r.metric = flowPctDay;
       r.curve = "focus";
     }
-    if (r.label === "How well you use AI") r.percentile = orchPctl;
+    if (r.label === "Your AI parallelism") r.percentile = orchPctl;
   }
   return { generatedAt: (/* @__PURE__ */ new Date()).toISOString(), gradedSessions: grades.length, agglomeration, outcomeCadence, dayDigest, dayChart, timeline, walkthroughs, rubric: rubric2, wordsByDay, dayThroughput, throughputCeiling: tCeiling, stackUp, criteria, parallelism, throughput, workstyle, flow, focus, expertiseMap, frontierArsenal, frontierDetail, coaching, gap, aggregate, delegationRollup, gapDist, projects };
 }
@@ -22769,7 +22798,7 @@ async function buildSkeleton(codingDir, stackUp) {
       ...thr.avgWordsPerDay ? [{ big: String(Math.round(thr.avgWordsPerDay).toLocaleString()), lab: "typed/dictated words per day, averaged (~300 for the median active user)" }] : [],
       ...thr.substantialDays ? [{ big: `${thr.substantialDays}/${thr.totalCodingDays}`, lab: "coding days were substantial (3h+ span)" }] : []
     ], moments: [] },
-    { key: "ai", label: "How they use AI", topPct: pct2("use ai"), sub: "orchestration, measured", claim: "", read: "", numbers: [
+    { key: "ai", label: "How they use AI", topPct: pct2("parallelism"), sub: "orchestration, measured", claim: "", read: "", numbers: [
       ...par.max ? [{ big: String(par.max), lab: "agents at once, peak" }] : [],
       ...par.hours2plus ? [{ big: `${par.hours2plus}h`, lab: "working 2+ live agents" }] : [],
       ...par.queueOps ? [{ big: par.queueOps.toLocaleString(), lab: "instructions queued ahead" }] : []
@@ -23189,12 +23218,15 @@ async function handlePublicCodingPageShare(dataDir, body) {
     return { status: 409, json: { error: "nothing to share yet \u2014 generate your page first" } };
   }
   const chatReport = body.includeChat === false ? null : await localChatReport();
+  const prior = await readOwnReportContent(cfg, auth.token, auth.identity.userId);
   try {
-    const payload = JSON.stringify({ format: "coding-pr1", page, ...chatReport ? { chatReport } : {} });
+    const merged = { ...prior, format: "coding-pr1", page };
+    if (chatReport) merged.chatReport = chatReport;
+    else delete merged.chatReport;
     const up = await fetch(`${cfg.supabaseUrl}/rest/v1/rpc/submit_report`, {
       method: "POST",
       headers: { apikey: cfg.supabaseAnonKey, authorization: `Bearer ${auth.token}`, "content-type": "application/json" },
-      body: JSON.stringify({ p_session_id: null, p_content: payload })
+      body: JSON.stringify({ p_session_id: null, p_content: JSON.stringify(merged) })
     });
     if (!up.ok) return { status: 502, json: { error: `share failed (HTTP ${up.status})` } };
     const verification = String(await up.json().catch(() => "unverified"));
@@ -23203,6 +23235,49 @@ async function handlePublicCodingPageShare(dataDir, body) {
     return { status: 502, json: { error: e.message.slice(0, 200) } };
   }
 }
+async function readOwnReportContent(cfg, token, userId) {
+  try {
+    const r = await fetch(`${cfg.supabaseUrl}/rest/v1/reports?user_id=eq.${encodeURIComponent(userId)}&select=content`, {
+      headers: { apikey: cfg.supabaseAnonKey, authorization: `Bearer ${token}` }
+    });
+    if (!r.ok) return {};
+    const rows = await r.json();
+    const c = rows[0]?.content;
+    return c && typeof c === "object" ? c : {};
+  } catch {
+    return {};
+  }
+}
+async function handleAnalysisShare(dataDir, kind, full) {
+  const cfg = centralConfig();
+  if (!cfg.configured) return { status: 503, json: { error: "central services are disabled in this build" } };
+  const auth = await getAccessToken(dataDir);
+  if (!auth) return { status: 401, json: { error: "sign in with Google (top right) first" } };
+  const slim = (p) => {
+    if (!p || typeof p !== "object") return p;
+    const { timeline: _t, walkthroughs: _w, dayDigest: _d, dayChart: _c, ...rest } = p;
+    return rest;
+  };
+  const value = kind === "coding" ? slim(full.coding()) : await full.chat().catch(() => null);
+  if (!value) return { status: 409, json: { error: "nothing to share yet \u2014 run the analysis first" } };
+  const prior = await readOwnReportContent(cfg, auth.token, auth.identity.userId);
+  try {
+    const merged = {
+      format: "coding-pr1",
+      ...prior,
+      [kind === "coding" ? "fullCoding" : "fullChat"]: value
+    };
+    const up = await fetch(`${cfg.supabaseUrl}/rest/v1/rpc/submit_report`, {
+      method: "POST",
+      headers: { apikey: cfg.supabaseAnonKey, authorization: `Bearer ${auth.token}`, "content-type": "application/json" },
+      body: JSON.stringify({ p_session_id: null, p_content: JSON.stringify(merged) })
+    });
+    if (!up.ok) return { status: 502, json: { error: `share failed (HTTP ${up.status})` } };
+    return { status: 200, json: { ok: true, kind } };
+  } catch (e) {
+    return { status: 502, json: { error: e.message.slice(0, 200) } };
+  }
+}
 
 // src/server.ts
 var __dirname = path33.dirname(fileURLToPath2(import.meta.url));
@@ -23530,6 +23605,18 @@ function startServer(opts) {
         json(res, out.status, out.json);
         return;
       }
+      if (req.method === "POST" && route === "/api/analysis/share") {
+        const body = JSON.parse(await readBody(req) || "{}");
+        const out = await handleAnalysisShare(dataDir, body.kind === "chat" ? "chat" : "coding", {
+          coding: () => liveProfile,
+          chat: async () => {
+            const p = await handlePerson();
+            return p.status === 200 ? p.json : null;
+          }
+        });
+        json(res, out.status, out.json);
+        return;
+      }
       if (route.startsWith("/api/")) {
         json(res, 404, { error: `this viewer doesn't serve ${route}` });
         return;
@@ -23551,7 +23638,7 @@ function startServer(opts) {
       serverUrl = `http://${host}:${port}`;
       void getCalibration().then((cal) => {
         if (cal.source === "central" && cal.doc.rubricVersion !== RUBRIC_FINGERPRINT2) {
-          console.warn(`[polymath-society] a newer grading rubric is published (${cal.doc.rubricVersion}; this install: ${RUBRIC_FINGERPRINT2}) \u2014 update the package and re-grade to stay comparable.`);
+          console.warn(`[polymath-society] you're not on the latest version \u2014 run: npm install -g polymath-society@latest`);
         }
       }).catch(() => {
       });

package/dist/pipeline/coding-agglomerate.js

@@ -16127,7 +16127,7 @@ var TRANSCRIPT_LINE_RE = new RegExp(`^\\s*(${NOTE})((,| and) ${NOTE})*\\s*$|^\\s
 
 // ../../lib/calibration/index.ts
 var DEFAULT_CALIBRATION = {
-  version: "2026-07-15.1",
+  version: "2026-07-16.2",
   // = RUBRIC_FINGERPRINT of the shipped npm rubric (packages/coding-analyzer/
   // src/grade/criteria.ts). A unit test fails loud when the rubric changes
   // without this being updated (and re-pushed to the server).
@@ -16218,6 +16218,7 @@ var DEFAULT_CALIBRATION = {
   codingTiers: {
     push: { median: 300, sigma: 0.91, tag: "estimated \u2014 log-normal fit", source: "Anthropic 2026 Claude Code study (400k sessions): median engaged user ~8 prompts/day at ~35 words \u2014 ~300 directed words/day; tail: a 1-in-1,000 user AVERAGES ~5k directed words/day (near-daily heavy dictation, clearly below the ~10k single-day record). AVERAGE over substantial days, cumulative not spiky." },
     focus: { median: 0.08, sigma: 0.559, tag: "estimated \u2014 proxy-anchored", source: "median: Anthropic 20h/week engaged-user figure implies ~1h/day of true rapid exchange against an 8h working day; ceiling anchored to the ~4h/day deep-work limit (45% share = 1-in-1,000). AVERAGE share, cumulative not spiky." },
+    hours: { median: 3, sigma: 0.4, tag: "estimated \u2014 proxy-anchored", source: "median: RescueTime-measured ~2h48m of real productive time/day (same source as the flow benchmark) \u2248 3h of active building; shape: sustaining ~7.5h/day \u2248 top 1%, a ~12h/day average \u2248 1-in-3,000. AVERAGE work-day length, cumulative not spiky." },
     orchestration: { tag: "estimated \u2014 behavior bands", source: "no published concurrency distribution exists; ceiling anchored to the Claude Code lead's documented 10-15 parallel sessions" }
   },
   codingBenchmarks: {

package/dist/pipeline/coding-aggregate.js

@@ -431,7 +431,7 @@ var IDLE_CAP_MS = 12 * 6e4;
 
 // ../../lib/calibration/index.ts
 var DEFAULT_CALIBRATION = {
-  version: "2026-07-15.1",
+  version: "2026-07-16.2",
   // = RUBRIC_FINGERPRINT of the shipped npm rubric (packages/coding-analyzer/
   // src/grade/criteria.ts). A unit test fails loud when the rubric changes
   // without this being updated (and re-pushed to the server).
@@ -522,6 +522,7 @@ var DEFAULT_CALIBRATION = {
   codingTiers: {
     push: { median: 300, sigma: 0.91, tag: "estimated \u2014 log-normal fit", source: "Anthropic 2026 Claude Code study (400k sessions): median engaged user ~8 prompts/day at ~35 words \u2014 ~300 directed words/day; tail: a 1-in-1,000 user AVERAGES ~5k directed words/day (near-daily heavy dictation, clearly below the ~10k single-day record). AVERAGE over substantial days, cumulative not spiky." },
     focus: { median: 0.08, sigma: 0.559, tag: "estimated \u2014 proxy-anchored", source: "median: Anthropic 20h/week engaged-user figure implies ~1h/day of true rapid exchange against an 8h working day; ceiling anchored to the ~4h/day deep-work limit (45% share = 1-in-1,000). AVERAGE share, cumulative not spiky." },
+    hours: { median: 3, sigma: 0.4, tag: "estimated \u2014 proxy-anchored", source: "median: RescueTime-measured ~2h48m of real productive time/day (same source as the flow benchmark) \u2248 3h of active building; shape: sustaining ~7.5h/day \u2248 top 1%, a ~12h/day average \u2248 1-in-3,000. AVERAGE work-day length, cumulative not spiky." },
     orchestration: { tag: "estimated \u2014 behavior bands", source: "no published concurrency distribution exists; ceiling anchored to the Claude Code lead's documented 10-15 parallel sessions" }
   },
   codingBenchmarks: {