@konneal/engine 0.2.72 → 0.2.73

package/dist/{ask-UOTYZ5DC.js → ask-MFFOHVQ2.js}

@@ -1,7 +1,7 @@
 import {
   handleAsk
-} from "./chunk-UW5M2AXR.js";
-import "./chunk-TPV5GPY2.js";
+} from "./chunk-V7PEVNNO.js";
+import "./chunk-LL3LWBZH.js";
 import "./chunk-SPKPH54J.js";
 import "./chunk-A3QHHUN5.js";
 import "./chunk-LVZTWUVJ.js";

package/dist/{chunk-TPV5GPY2.js → chunk-LL3LWBZH.js}

@@ -9,6 +9,7 @@ import {
   THRESHOLDS,
   answerEffort,
   effortBudget,
+  num,
   processExpansion,
   sha256Hex,
   today
@@ -817,8 +818,8 @@ var BaseUrnGenerator = class {
     const ed = this.maybe("edition");
     if (ed === void 0 || ed === null)
       return void 0;
-    const num = typeof ed === "object" && "number" in ed ? ed.number : ed;
-    return num === void 0 || num === null || num === "" ? void 0 : `ed.${String(num)}`;
+    const num2 = typeof ed === "object" && "number" in ed ? ed.number : ed;
+    return num2 === void 0 || num2 === null || num2 === "" ? void 0 : `ed.${String(num2)}`;
   }
   urnLanguage() {
     const langs = this.maybe("languages");
@@ -7139,9 +7140,9 @@ var IsoIdentifier = class _IsoIdentifier extends BaseIdentifier {
         }
       }
     }
-    const num = this.numberPortion(true);
-    if (num !== "")
-      result += result === "" ? num : ` ${num}`;
+    const num2 = this.numberPortion(true);
+    if (num2 !== "")
+      result += result === "" ? num2 : ` ${num2}`;
     result += this.languagePortion(this.languageSingle());
     if (this.all_parts === true)
       result += " (all parts)";
@@ -7306,9 +7307,9 @@ var IsoSingleIdentifier = class extends IsoIdentifier {
         result += `${sep}${stage}`;
       }
     }
-    const num = this.numberPortion(true);
-    if (num !== "")
-      result += result === "" ? num : ` ${num}`;
+    const num2 = this.numberPortion(true);
+    if (num2 !== "")
+      result += result === "" ? num2 : ` ${num2}`;
     result += this.languagePortion(this.languageSingle());
     if (this.all_parts === true)
       result += " (all parts)";
@@ -7611,11 +7612,11 @@ var IsoSupplementIdentifier = class extends IsoSingleIdentifier {
     const ts = this.typedStageState;
     const stage = ts === void 0 ? "" : stageAbbr(ts.entry, long, ts.originalAbbr);
     let result = `${baseStr}/${stage}`;
-    const num = this.numberPortion(true);
-    if (num !== "") {
-      if (!num.startsWith(":") && !stage.endsWith("."))
+    const num2 = this.numberPortion(true);
+    if (num2 !== "") {
+      if (!num2.startsWith(":") && !stage.endsWith("."))
         result += " ";
-      result += num;
+      result += num2;
     }
     result += this.languagePortion(this.languageSingle());
     return result;
@@ -7670,20 +7671,20 @@ var IsoDirectives = class extends IsoSingleIdentifier {
     const abbr = this.typedStageState === void 0 ? "" : stageAbbr(this.typedStageState.entry, false);
     if (abbr !== "")
       head += ` ${abbr}`;
-    let num = this.number ?? "";
+    let num2 = this.number ?? "";
     if (this.part !== void 0)
-      num += ` ${this.part}`;
+      num2 += ` ${this.part}`;
     if (this.subpart !== void 0)
-      num += `-${this.subpart}`;
+      num2 += `-${this.subpart}`;
     if (this.stage_iteration !== void 0)
-      num += `.${this.stage_iteration.render() ?? ""}`;
+      num2 += `.${this.stage_iteration.render() ?? ""}`;
     if (this.date !== void 0 && this.date.present())
-      num += `:${this.date.render()}`;
-    if (num === "")
+      num2 += `:${this.date.render()}`;
+    if (num2 === "")
       return head;
-    if (num.startsWith(":"))
-      return head + num;
-    return `${head} ${num}`;
+    if (num2.startsWith(":"))
+      return head + num2;
+    return `${head} ${num2}`;
   }
 };
 registerType(IsoDirectives);
@@ -8606,9 +8607,9 @@ function preprocessNist(input) {
   cleaned = cleaned.replace(/(?<!e)(\d)(rev\d{4})/g, "$1 $2");
   cleaned = cleaned.replace(/(\d+e\d+)\.([A-Za-z]{3,9}\d{4})/g, "$1rev$2");
   if (/\bIR\b/.test(cleaned) && !cleaned.includes("CIRC")) {
-    cleaned = cleaned.replace(/(\d)r(\d{1,2})\/(\d{2,4})/g, (_m, num, mon, yr) => {
+    cleaned = cleaned.replace(/(\d)r(\d{1,2})\/(\d{2,4})/g, (_m, num2, mon, yr) => {
       const yyyy = yr.length === 2 ? `19${yr}` : yr;
-      return `${num}/Upd1-${yyyy}${mon.padStart(2, "0")}`;
+      return `${num2}/Upd1-${yyyy}${mon.padStart(2, "0")}`;
     });
   }
   if (!cleaned.includes("LCIRC") && !cleaned.includes("CIRC")) {
@@ -11321,17 +11322,17 @@ function normalizeJointStageSpellings(input) {
   let cleaned = input;
   const pubs = "((?:ISO/IEC/IEEE|IEEE/ISO/IEC|IEEE/IEC/ISO|ISO/IEEE|IEC/IEEE|IEEE/IEC|ISO/IEC)(?:/ ?| ))";
   const stage = "(FDIS|FCD|CDV|DIS\\d?|CD\\d?|WD|PWI|NP)";
-  const num = "(P?\\d+(?:[.-]\\d+)?)";
+  const num2 = "(P?\\d+(?:[.-]\\d+)?)";
   const tail = "(,? (?:Jan|Feb|Mar|Apr|May|Jun|Jul|Aug|Sep|Sept|Oct|Nov|Dec)[a-z]* \\d{4}|$|-\\d{4}(?:-\\d\\d)?)";
   cleaned = cleaned.replace(/(\d)_D(\d)/g, "$1/D$2");
   cleaned = cleaned.replace(/,(?=[A-Za-z])/, ", ");
-  cleaned = cleaned.replace(new RegExp(`^${pubs}${num}/(D[\\d.]+)/ ?${stage}\\b`), "$1$4 $2/$3");
-  cleaned = cleaned.replace(new RegExp(`^${pubs}${num}/ ${stage}${tail}`), "$1$3 $2$4");
-  cleaned = cleaned.replace(new RegExp(`^${pubs}${num}/${stage} (\\d{4})\\b`), "$1$3 $2-$4");
-  cleaned = cleaned.replace(new RegExp(`^${pubs}${num} ${stage}${tail}`), "$1$3 $2$4");
-  cleaned = cleaned.replace(new RegExp(`^${pubs}${num} ${stage} (D\\d+)\\b`), "$1$3 $2/$4");
-  cleaned = cleaned.replace(new RegExp(`^${pubs}${num}/ ?${stage}(-(?:19|20)\\d\\d(?:-\\d\\d)?)`), "$1$3 $2$4");
-  cleaned = cleaned.replace(new RegExp(`^${pubs}${stage} ${num}/? ?(D\\d+)\\b`), "$1$2 $3/$4");
+  cleaned = cleaned.replace(new RegExp(`^${pubs}${num2}/(D[\\d.]+)/ ?${stage}\\b`), "$1$4 $2/$3");
+  cleaned = cleaned.replace(new RegExp(`^${pubs}${num2}/ ${stage}${tail}`), "$1$3 $2$4");
+  cleaned = cleaned.replace(new RegExp(`^${pubs}${num2}/${stage} (\\d{4})\\b`), "$1$3 $2-$4");
+  cleaned = cleaned.replace(new RegExp(`^${pubs}${num2} ${stage}${tail}`), "$1$3 $2$4");
+  cleaned = cleaned.replace(new RegExp(`^${pubs}${num2} ${stage} (D\\d+)\\b`), "$1$3 $2/$4");
+  cleaned = cleaned.replace(new RegExp(`^${pubs}${num2}/ ?${stage}(-(?:19|20)\\d\\d(?:-\\d\\d)?)`), "$1$3 $2$4");
+  cleaned = cleaned.replace(new RegExp(`^${pubs}${stage} ${num2}/? ?(D\\d+)\\b`), "$1$2 $3/$4");
   return cleaned;
 }
 function preprocessIeee(input) {
@@ -14556,8 +14557,8 @@ var CenBuilder = class {
     const numStr = (str3(d["number"]) ?? "").trim();
     if (!/^\d{1,6}$/.test(numStr))
       return void 0;
-    const num = Number.parseInt(numStr, 10);
-    if (!(num >= 6e4 && num <= 79999))
+    const num2 = Number.parseInt(numStr, 10);
+    if (!(num2 >= 6e4 && num2 <= 79999))
       return void 0;
     const impl = grammarImplementation("iec");
     if (impl === void 0)
@@ -19006,9 +19007,9 @@ var CsaAdoptedClass = (() => {
           baseStr = baseStr.replace(`:${year}`, `:${year.slice(2)}`);
         }
       }
-      baseStr = baseStr.replace(/\/Amd\s+(\d+)([:/-])(\d{2,4})\b/g, (_m, num, sep, year) => {
+      baseStr = baseStr.replace(/\/Amd\s+(\d+)([:/-])(\d{2,4})\b/g, (_m, num2, sep, year) => {
         const short = year.length === 4 && (year.startsWith("20") || year.startsWith("19")) ? year.slice(2) : year;
-        return `/A${num}${sep}${short}`;
+        return `/A${num2}${sep}${short}`;
       });
       let result;
       if (this.publisher_prefix !== void 0 && this.publisher_prefix.endsWith("-")) {
@@ -19635,9 +19636,9 @@ function parseCsa(input) {
       const full = Number(two) < 50 ? `20${two}` : `19${two}`;
       wrapped = wrapped.replace(new RegExp(`:${two}\\b`), `:${full}`);
     }
-    wrapped = wrapped.replace(/\/A(\d+)([:/-])(\d{2,4})\b/g, (_m, num, sep, year) => {
+    wrapped = wrapped.replace(/\/A(\d+)([:/-])(\d{2,4})\b/g, (_m, num2, sep, year) => {
       const full = year.length === 2 ? Number(year) < 50 ? `20${year}` : `19${year}` : year;
-      return `/Amd ${num}${sep}${full}`;
+      return `/Amd ${num2}${sep}${full}`;
     });
     const base = parseExternalStandard(wrapped);
     if (base === void 0) {
@@ -21908,13 +21909,13 @@ function parseOimlSpine(display) {
     const kind = String(h._type ?? "").split(":").pop() ?? "";
     const letter = TYPE_LETTER[kind] ?? "";
     if (!letter) return null;
-    const num = String(Number(h.number));
+    const num2 = String(Number(h.number));
     const part = h.part !== void 0 ? String(h.part) : void 0;
     const ed = h.year !== void 0 ? String(h.year) : h.edition !== void 0 ? String(h.edition) : void 0;
     return {
-      doc_number: num,
+      doc_number: num2,
       ...ed ? { edition: ed } : {},
-      label: `OIML ${letter} ${num}${part ? `-${part}` : ""}${ed ? `:${ed}` : ""}`
+      label: `OIML ${letter} ${num2}${part ? `-${part}` : ""}${ed ? `:${ed}` : ""}`
     };
   } catch {
     return null;
@@ -21949,9 +21950,9 @@ var oimlPubid = {
     for (const m of query.matchAll(re)) {
       const [, oimlPrefix, letter, gap, digits, edition] = m;
       if (digits.length === 1 && !oimlPrefix && !gap) continue;
-      const num = String(Number(digits));
+      const num2 = String(Number(digits));
       const type = letter.toUpperCase();
-      return { doc_number: num, ...edition ? { edition } : {}, label: `OIML ${type} ${num}${edition ? `:${edition}` : ""}` };
+      return { doc_number: num2, ...edition ? { edition } : {}, label: `OIML ${type} ${num2}${edition ? `:${edition}` : ""}` };
     }
     return null;
   },
@@ -24578,6 +24579,32 @@ function telemetry(env, ctx, tier, route, model, ok, answerChars, queryHash, lan
     ])
   );
 }
+function estimateTokens(chars) {
+  return Math.ceil(chars / 4);
+}
+function tokenTier(bucketId) {
+  return bucketId.startsWith("key:") ? "key" : bucketId.startsWith("sub:") ? "member" : "anon";
+}
+function tokenLimit(env, bucketId) {
+  const tier = tokenTier(bucketId);
+  return tier === "key" ? num(env, "KEY_DAY_TOKENS", 4e6) : tier === "member" ? num(env, "MEMBER_DAY_TOKENS", 15e5) : num(env, "ANON_DAY_TOKENS", 3e5);
+}
+async function tokenBudget(env, bucketId) {
+  const used = Number(await env.CACHE.get(`t:${today()}:ask:${await sha256Hex(bucketId)}`) ?? "0");
+  return { used, limit: tokenLimit(env, bucketId) };
+}
+async function chargeTokens(env, bucketId, tokens) {
+  if (!(tokens > 0)) return;
+  await kvIncr(env.CACHE, `t:${today()}:ask:${await sha256Hex(bucketId)}`, tokens);
+}
+function usageTotal(usage) {
+  if (!usage || typeof usage !== "object") return null;
+  const u = usage;
+  const p = Number(u.prompt_tokens ?? 0);
+  const c = Number(u.completion_tokens ?? 0);
+  const total = p + c;
+  return Number.isFinite(total) && total > 0 ? total : null;
+}
 
 // workers/worker_public/src/graph.ts
 function docNumberOf(nodeId) {
@@ -24662,6 +24689,10 @@ export {
   clientIp,
   checkQuota,
   telemetry,
+  estimateTokens,
+  tokenBudget,
+  chargeTokens,
+  usageTotal,
   graphExpand,
   editionNote
 };

package/dist/{chunk-FLNX4O7C.js → chunk-OXTRNMTZ.js}

@@ -7,7 +7,7 @@ import {
   sessionFrom,
   telemetry,
   understandQuery
-} from "./chunk-TPV5GPY2.js";
+} from "./chunk-LL3LWBZH.js";
 import {
   corsHeaders,
   err,

package/dist/{chunk-UW5M2AXR.js → chunk-V7PEVNNO.js}

@@ -3,12 +3,14 @@ import {
   answerQuality,
   appliedContext,
   buildMessages,
+  chargeTokens,
   checkQuota,
   citations,
   clientIp,
   contextNote,
   editionNote,
   embed,
+  estimateTokens,
   experimentalSourceLabels,
   generateOnce,
   graphExpand,
@@ -34,8 +36,10 @@ import {
   splitHistory,
   syntheticUnderstanding,
   telemetry,
-  understandQuery
-} from "./chunk-TPV5GPY2.js";
+  tokenBudget,
+  understandQuery,
+  usageTotal
+} from "./chunk-LL3LWBZH.js";
 import {
   corsHeaders,
   err,
@@ -2166,7 +2170,23 @@ async function summarizeHistory(env, model, turns) {
     return null;
   }
 }
-async function* sseTokens(stream) {
+function messageChars(messages) {
+  let n = 0;
+  for (const m of messages) {
+    if (typeof m.content === "string") n += m.content.length;
+    else if (Array.isArray(m.content)) {
+      for (const part of m.content) if (part?.type === "text" && typeof part.text === "string") n += part.text.length;
+    }
+  }
+  return n;
+}
+async function chargeAnswerTokens(env, bucketId, stats, promptChars, answerChars) {
+  const total = usageTotal(stats?.usage) ?? estimateTokens(promptChars + answerChars);
+  await chargeTokens(env, bucketId, total);
+  const budget = await tokenBudget(env, bucketId);
+  return { tokens_used: budget.used, token_limit: budget.limit };
+}
+async function* sseTokens(stream, stats) {
   const reader = stream.getReader();
   const decoder = new TextDecoder();
   let buf = "";
@@ -2184,6 +2204,7 @@ async function* sseTokens(stream) {
       try {
         const evt = JSON.parse(payload);
         const tok = typeof evt?.response === "string" ? evt.response : evt?.choices?.[0]?.delta?.content;
+        if (evt.usage && stats && stats.usage === void 0) stats.usage = evt.usage;
         if (tok) yield tok;
       } catch {
       }
@@ -2227,6 +2248,12 @@ async function handleAsk(env, ctx, req, tier, key) {
   if (!quota.ok) {
     return err(429, "quota_exceeded", `Daily question limit reached (${quota.limit}). Try again tomorrow.`);
   }
+  const tokBudget = await tokenBudget(env, bucketId);
+  if (tokBudget.used >= tokBudget.limit) {
+    return err(429, "quota_exceeded", `Daily token budget reached (${tokBudget.limit} tokens). Try again tomorrow.`);
+  }
+  quota.tokens_used = tokBudget.used;
+  quota.token_limit = tokBudget.limit;
   const hardCap = num(env, "ANON_DAY_HARD_CAP", 5e3);
   if (tier === "anon" && quota.used > hardCap) {
     return err(503, "generation_disabled", "Generation is temporarily paused; search remains available.");
@@ -2444,14 +2471,17 @@ ${summary}` }] : [],
 `));
             if (readAs()) send({ type: "read", read: readAs() });
             send({ type: "citations", citations: [], context_applied: NO_CONTEXT, quota });
+            const stats = {};
             let full = "";
             try {
-              for await (const tok of sseTokens(stream)) {
+              for await (const tok of sseTokens(stream, stats)) {
                 full += tok;
                 send({ type: "token", v: tok });
               }
             } catch {
             }
+            const tq2 = await chargeAnswerTokens(env, bucketId, stats, messageChars(messages2), full.length);
+            send({ type: "quota", quota: { ...quota, ...tq2 } });
             send({ type: "done", model, query_hash: queryHash2, context_applied: NO_CONTEXT, read: readAs() });
             telemetry(env, ctx, tier, "ask", model, true, full.length, queryHash2, q.lang, void 0, telemetryMeta());
             controller.close();
@@ -2507,7 +2537,8 @@ ${summary}` }] : [],
         corsHeaders(req)
       );
     }
-    return json({ answer: verdict.answer, citations: citations2, model, query_hash: queryHash2, follow_ups: [], context_applied: draftCtxApplied, read: readAs(), ...draftPayload ? { draft: draftPayload } : {}, quota });
+    const tq2 = await chargeAnswerTokens(env, bucketId, void 0, q.query.length, verdict.answer.length);
+    return json({ answer: verdict.answer, citations: citations2, model, query_hash: queryHash2, follow_ups: [], context_applied: draftCtxApplied, read: readAs(), ...draftPayload ? { draft: draftPayload } : {}, quota: { ...quota, ...tq2 } });
   }
   if (apiCallIntent) {
     const callCtxApplied = declaredCtx ? appliedContext(declaredCtx, null) : NO_CONTEXT;
@@ -2534,7 +2565,8 @@ ${summary}` }] : [],
         corsHeaders(req)
       );
     }
-    return json({ answer: verdict.answer, citations: [], model, query_hash: queryHash2, follow_ups: [], context_applied: callCtxApplied, read: readAs(), ...draftPayload ? { draft: draftPayload } : {}, quota });
+    const tq2 = await chargeAnswerTokens(env, bucketId, void 0, q.query.length, verdict.answer.length);
+    return json({ answer: verdict.answer, citations: [], model, query_hash: queryHash2, follow_ups: [], context_applied: callCtxApplied, read: readAs(), ...draftPayload ? { draft: draftPayload } : {}, quota: { ...quota, ...tq2 } });
   }
   let liveRecords;
   let accountNote;
@@ -2834,14 +2866,17 @@ ${q.query}` }],
           send({ type: "read", read: readAs() });
           const sourceQuality = answerQuality(cites.map((c) => c.quality));
           send({ type: "citations", citations: cites, context_applied: ctxApplied, ...sourceQuality ? { source_quality: sourceQuality, confidence_note: qualityNote(sourceQuality), ...sourceQuality === "ocr" ? { experimental_sources: experimentalSourceLabels(cites) } : {} } : {}, ...liveRecords ? { records: liveRecords } : {}, quota });
+          const stats = {};
           let full = "";
           try {
-            for await (const tok of sseTokens(stream)) {
+            for await (const tok of sseTokens(stream, stats)) {
               full += tok;
               send({ type: "token", v: tok });
             }
           } catch {
           }
+          const tq2 = await chargeAnswerTokens(env, bucketId, stats, messageChars(messages), full.length);
+          send({ type: "quota", quota: { ...quota, ...tq2 } });
           const canonical0 = canonicalRefusal(full);
           const c2 = canonical0.includes(refusalAnswer()) ? { text: canonical0, blocks: [], dropped: [] } : await contractV2(env.DB, canonical0, usedHits);
           send({
@@ -3007,7 +3042,8 @@ ${q.query}` }],
     text: h.text.slice(0, 1200),
     ...h.metadata.table_selection ? { sel: h.metadata.table_selection } : {}
   }));
-  return json({ ...out, context: contextOut, read: readAs(), quota }, 200, { ...corsHeaders(req), "server-timing": serverTiming() });
+  const tq = await chargeAnswerTokens(env, bucketId, void 0, messageChars(messages), answer.length);
+  return json({ ...out, context: contextOut, read: readAs(), quota: { ...quota, ...tq } }, 200, { ...corsHeaders(req), "server-timing": serverTiming() });
 }
 function sseResponse(events, cors) {
   const encoder = new TextEncoder();

package/dist/quota.d.ts

@@ -15,3 +15,11 @@ export declare function telemetry(env: Env, ctx: Background, tier: string, route
     keyId?: string | null;
     retries?: number;
 }): void;
+export declare function estimateTokens(chars: number): number;
+export declare function tokenLimit(env: Env, bucketId: string): number;
+export declare function tokenBudget(env: Env, bucketId: string): Promise<{
+    used: number;
+    limit: number;
+}>;
+export declare function chargeTokens(env: Env, bucketId: string, tokens: number): Promise<void>;
+export declare function usageTotal(usage: unknown): number | null;

package/dist/{search-PBOODCMQ.js → search-LN5Q7OSZ.js}

@@ -1,7 +1,7 @@
 import {
   handleSearch
-} from "./chunk-FLNX4O7C.js";
-import "./chunk-TPV5GPY2.js";
+} from "./chunk-OXTRNMTZ.js";
+import "./chunk-LL3LWBZH.js";
 import "./chunk-SPKPH54J.js";
 import "./chunk-A3QHHUN5.js";
 import "./chunk-LVZTWUVJ.js";

package/dist/worker_public/src/index.js

@@ -6,10 +6,10 @@ import {
   modelGroundingBlock,
   scoreFaithfulness,
   scoreJudge
-} from "../../chunk-UW5M2AXR.js";
+} from "../../chunk-V7PEVNNO.js";
 import {
   handleSearch
-} from "../../chunk-FLNX4O7C.js";
+} from "../../chunk-OXTRNMTZ.js";
 import {
   buildMessages,
   cfBlobs,
@@ -35,7 +35,7 @@ import {
   sessionFrom,
   telemetry,
   understandQuery
-} from "../../chunk-TPV5GPY2.js";
+} from "../../chunk-LL3LWBZH.js";
 import {
   authenticate,
   corsHeaders,
@@ -959,7 +959,7 @@ async function handleMcp(env, ctx, req, tier, key) {
       // stream:false forces the JSON lane (anon defaults to SSE)
       body: JSON.stringify({ ...args, stream: false })
     });
-    const res = name === "ask" ? await (await import("../../ask-UOTYZ5DC.js")).handleAsk(env, ctx, inner, tier, key) : await (await import("../../search-PBOODCMQ.js")).handleSearch(env, ctx, inner, tier, key);
+    const res = name === "ask" ? await (await import("../../ask-MFFOHVQ2.js")).handleAsk(env, ctx, inner, tier, key) : await (await import("../../search-LN5Q7OSZ.js")).handleSearch(env, ctx, inner, tier, key);
     return res.json().catch(() => ({ error: { message: "tool transport failed", status: res.status } }));
   });
   if (out.ok && "accepted" in out) return new Response(null, { status: 202 });

package/package.json

@@ -27,7 +27,7 @@
     "playwright": "^1.62.1",
     "yaml": "^2.9.1"
   },
-  "version": "0.2.72",
+  "version": "0.2.73",
   "description": "The Konneal engine: the publisher-agnostic build pipeline and API plane for standards intelligence (retrieval, answer contract, verdicts, evaluation).",
   "license": "BSD-3-Clause",
   "type": "module",

package/workers/worker_public/src/ask.ts

@@ -37,7 +37,7 @@ import { cacheKeyMaterial, corpusGen, exactCacheKey, freshRequested } from "./an
 import type { Env } from "./env";
 export type { Env };
 import { json, err, corsHeaders, readJson, validateQuery, type ApiKey } from "./lib/http";
-import { clientIp, checkQuota, telemetry } from "./quota";
+import { chargeTokens, checkQuota, clientIp, estimateTokens, telemetry, tokenBudget, usageTotal } from "./quota";
 import { graphExpand, editionNote } from "./graph";
 import { registerNote, searchRegister, certificateLinks } from "./certificates";
 import { TOOL_DECLARATION, parseToolCall, runTool, toolNote } from "./tools";
@@ -206,7 +206,28 @@ async function summarizeHistory(
   }
 }
 
-async function* sseTokens(stream: ReadableStream<Uint8Array>): AsyncGenerator<string> {
+/** The prompt's character weight — text parts only, image data URLs
+ *  excluded (a photo must not read as half a million characters). */
+function messageChars(messages: any[]): number {
+  let n = 0;
+  for (const m of messages) {
+    if (typeof m.content === "string") n += m.content.length;
+    else if (Array.isArray(m.content)) for (const part of m.content) if (part?.type === "text" && typeof part.text === "string") n += part.text.length;
+  }
+  return n;
+}
+
+/** Post-generation true-up: the model's own usage when the stream
+ *  reported it, the character estimate otherwise — charged, then the
+ *  budget's fresh state returned for the quota payload. */
+async function chargeAnswerTokens(env: Env, bucketId: string, stats: { usage?: unknown } | undefined, promptChars: number, answerChars: number): Promise<{ tokens_used: number; token_limit: number }> {
+  const total = usageTotal(stats?.usage) ?? estimateTokens(promptChars + answerChars);
+  await chargeTokens(env, bucketId, total);
+  const budget = await tokenBudget(env, bucketId);
+  return { tokens_used: budget.used, token_limit: budget.limit };
+}
+
+async function* sseTokens(stream: ReadableStream<Uint8Array>, stats?: { usage?: unknown }): AsyncGenerator<string> {
   const reader = stream.getReader();
   const decoder = new TextDecoder();
   let buf = "";
@@ -227,6 +248,7 @@ async function* sseTokens(stream: ReadableStream<Uint8Array>): AsyncGenerator<st
           typeof evt?.response === "string"
             ? evt.response
             : evt?.choices?.[0]?.delta?.content;
+        if (evt.usage && stats && stats.usage === undefined) stats.usage = evt.usage;
         if (tok) yield tok;
       } catch {
         // partial JSON in line splitting — ignore
@@ -320,6 +342,14 @@ async function handleAsk(
   if (!quota.ok) {
     return err(429, "quota_exceeded", `Daily question limit reached (${quota.limit}). Try again tomorrow.`);
   }
+  // the real allowance is the day's TOKENS (the 2026-09-29 shift): the
+  // request cap above is the anti-abuse backstop, this is the budget
+  const tokBudget = await tokenBudget(env, bucketId);
+  if (tokBudget.used >= tokBudget.limit) {
+    return err(429, "quota_exceeded", `Daily token budget reached (${tokBudget.limit} tokens). Try again tomorrow.`);
+  }
+  (quota as any).tokens_used = tokBudget.used;
+  (quota as any).token_limit = tokBudget.limit;
 
   const hardCap = num(env as any, "ANON_DAY_HARD_CAP", 5000);
   if (tier === "anon" && quota.used > hardCap) {
@@ -668,15 +698,18 @@ let convEntities: Array<{ entity: string; kind: string }> = [];
             // the reading arrives first — a conversational turn is still a reading
             if (readAs()) send({ type: "read", read: readAs() });
             send({ type: "citations", citations: [], context_applied: NO_CONTEXT, quota, });
+            const stats: { usage?: unknown } = {};
             let full = "";
             try {
-              for await (const tok of sseTokens(stream)) {
+              for await (const tok of sseTokens(stream, stats)) {
                 full += tok;
                 send({ type: "token", v: tok });
               }
             } catch {
               // stream ended prematurely — deliver what we have
             }
+            const tq = await chargeAnswerTokens(env, bucketId, stats, messageChars(messages), full.length);
+            send({ type: "quota", quota: { ...quota, ...tq } });
             send({ type: "done", model, query_hash: queryHash, context_applied: NO_CONTEXT, read: readAs() });
             telemetry(env, ctx, tier, "ask", model, true, full.length, queryHash, q.lang, undefined, telemetryMeta());
             controller.close();
@@ -748,7 +781,8 @@ let convEntities: Array<{ entity: string; kind: string }> = [];
         corsHeaders(req),
       );
     }
-    return json({ answer: verdict.answer, citations, model, query_hash: queryHash, follow_ups: [], context_applied: draftCtxApplied, read: readAs(), ...(draftPayload ? { draft: draftPayload } : {}), quota, });
+    const tq = await chargeAnswerTokens(env, bucketId, undefined, q.query.length, verdict.answer.length);
+    return json({ answer: verdict.answer, citations, model, query_hash: queryHash, follow_ups: [], context_applied: draftCtxApplied, read: readAs(), ...(draftPayload ? { draft: draftPayload } : {}), quota: { ...quota, ...tq } });
   }
 
   // ── The api_call draft (TODO.ai-platform/09) — the operations
@@ -785,7 +819,8 @@ let convEntities: Array<{ entity: string; kind: string }> = [];
         corsHeaders(req),
       );
     }
-    return json({ answer: verdict.answer, citations: [], model, query_hash: queryHash, follow_ups: [], context_applied: callCtxApplied, read: readAs(), ...(draftPayload ? { draft: draftPayload } : {}), quota, });
+    const tq = await chargeAnswerTokens(env, bucketId, undefined, q.query.length, verdict.answer.length);
+    return json({ answer: verdict.answer, citations: [], model, query_hash: queryHash, follow_ups: [], context_applied: callCtxApplied, read: readAs(), ...(draftPayload ? { draft: draftPayload } : {}), quota: { ...quota, ...tq } });
   }
 
   // (declared before the retrieval try: the account block, the refusal
@@ -1227,15 +1262,18 @@ let convEntities: Array<{ entity: string; kind: string }> = [];
           send({ type: "read", read: readAs() });
           const sourceQuality = answerQuality(cites.map((c: any) => c.quality));
           send({ type: "citations", citations: cites, context_applied: ctxApplied, ...(sourceQuality ? { source_quality: sourceQuality, confidence_note: qualityNote(sourceQuality), ...(sourceQuality === "ocr" ? { experimental_sources: experimentalSourceLabels(cites) } : {}) } : {}), ...(liveRecords ? { records: liveRecords } : {}), quota, });
+          const stats: { usage?: unknown } = {};
           let full = "";
           try {
-            for await (const tok of sseTokens(stream)) {
+            for await (const tok of sseTokens(stream, stats)) {
               full += tok;
               send({ type: "token", v: tok });
             }
           } catch {
             // stream ended prematurely — deliver what we have
           }
+          const tq = await chargeAnswerTokens(env, bucketId, stats, messageChars(messages), full.length);
+          send({ type: "quota", quota: { ...quota, ...tq } });
           const canonical0 = canonicalRefusal(full);
           // answer contract v2: validate [[u:]] refs, resolve typed blocks
           const c2 = canonical0.includes(refusalAnswer())
@@ -1475,7 +1513,8 @@ let convEntities: Array<{ entity: string; kind: string }> = [];
     text: h.text.slice(0, 1200),
     ...((h.metadata as any).table_selection ? { sel: (h.metadata as any).table_selection } : {}),
   }));
-  return json({ ...out, context: contextOut, read: readAs(), quota }, 200, { ...corsHeaders(req), "server-timing": serverTiming() });
+  const tq = await chargeAnswerTokens(env, bucketId, undefined, messageChars(messages), answer.length);
+  return json({ ...out, context: contextOut, read: readAs(), quota: { ...quota, ...tq } }, 200, { ...corsHeaders(req), "server-timing": serverTiming() });
 }
 
 function sseResponse(events: unknown[], cors: Record<string, string>): Response {

package/workers/worker_public/src/quota.ts

@@ -1,6 +1,6 @@
 // Quota + telemetry: per-bucket daily counters (KV) and the D1
 // queries/spend ledger writes (TODO.impl/23).
-import { sha256Hex, today } from "./config";
+import { num, sha256Hex, today } from "./config.ts";
 import type { Env } from "./env";
 import type { Background } from "./ports/runtime.ts";
 
@@ -54,3 +54,44 @@ export function telemetry(
     ]),
   );
 }
+
+// ── the token budget (2026-09-29): the daily allowance is TOKENS, not
+// questions — an elevated-effort answer over a wide retrieval context
+// costs orders of magnitude more than a cached refusal, and one number
+// should say so. Charged after generation from the model's own usage
+// when the stream reports it, estimated by characters otherwise.
+export function estimateTokens(chars: number): number {
+  return Math.ceil(chars / 4);
+}
+
+function tokenTier(bucketId: string): "anon" | "member" | "key" {
+  return bucketId.startsWith("key:") ? "key" : bucketId.startsWith("sub:") ? "member" : "anon";
+}
+
+export function tokenLimit(env: Env, bucketId: string): number {
+  const tier = tokenTier(bucketId);
+  return tier === "key"
+    ? num(env as any, "KEY_DAY_TOKENS", 4_000_000)
+    : tier === "member"
+      ? num(env as any, "MEMBER_DAY_TOKENS", 1_500_000)
+      : num(env as any, "ANON_DAY_TOKENS", 300_000);
+}
+
+export async function tokenBudget(env: Env, bucketId: string): Promise<{ used: number; limit: number }> {
+  const used = Number((await env.CACHE.get(`t:${today()}:ask:${await sha256Hex(bucketId)}`)) ?? "0");
+  return { used, limit: tokenLimit(env, bucketId) };
+}
+
+export async function chargeTokens(env: Env, bucketId: string, tokens: number): Promise<void> {
+  if (!(tokens > 0)) return;
+  await kvIncr(env.CACHE, `t:${today()}:ask:${await sha256Hex(bucketId)}`, tokens);
+}
+
+export function usageTotal(usage: unknown): number | null {
+  if (!usage || typeof usage !== "object") return null;
+  const u = usage as any;
+  const p = Number(u.prompt_tokens ?? 0);
+  const c = Number(u.completion_tokens ?? 0);
+  const total = p + c;
+  return Number.isFinite(total) && total > 0 ? total : null;
+}