@konneal/engine 0.2.164 → 0.2.166

package/dist/ask-XJOOMAXU.js

@@ -0,0 +1,16 @@
+import {
+  handleAsk
+} from "./chunk-2FME6R6J.js";
+import "./chunk-ECOYO7EX.js";
+import "./chunk-AP73VIHB.js";
+import "./chunk-5AME76UW.js";
+import "./chunk-5V2A3CAH.js";
+import "./chunk-TTEJQ5XA.js";
+import "./chunk-ETLSWSZ7.js";
+import "./chunk-OLO45N6D.js";
+import "./chunk-FQYHT4VC.js";
+import "./chunk-3FYJM7LH.js";
+import "./chunk-LSZTJI7V.js";
+export {
+  handleAsk
+};

package/dist/{chunk-GVCOGEJC.js → chunk-2FME6R6J.js}

@@ -32,7 +32,7 @@ import {
   tableRetyped,
   toolNote,
   verdictNote
-} from "./chunk-K5WL3DKH.js";
+} from "./chunk-AP73VIHB.js";
 import {
   NO_CONTEXT,
   STAGE_NAMES,
@@ -75,18 +75,18 @@ import {
   tokenBudget,
   understandQuery,
   usageTotal
-} from "./chunk-E74MNLBC.js";
+} from "./chunk-5AME76UW.js";
 import {
   embed,
   generateOnce
-} from "./chunk-FHRGC6TP.js";
+} from "./chunk-5V2A3CAH.js";
 import {
   corsHeaders,
   err,
   json,
   readJson,
   validateQuery
-} from "./chunk-GJATVGBS.js";
+} from "./chunk-TTEJQ5XA.js";
 import {
   cacheableAnswer,
   canonicalRefusal,
@@ -99,7 +99,7 @@ import {
   requestSalt,
   resolveRequestScope,
   standardKeysFrom
-} from "./chunk-D6UUAL7I.js";
+} from "./chunk-OLO45N6D.js";
 import {
   LIMITS,
   MODELS,
@@ -110,7 +110,7 @@ import {
   requestEffort,
   roleModel,
   sha256Hex
-} from "./chunk-ZD6TYLET.js";
+} from "./chunk-FQYHT4VC.js";
 import {
   P
 } from "./chunk-3FYJM7LH.js";

package/dist/{chunk-E74MNLBC.js → chunk-5AME76UW.js}

@@ -1,11 +1,11 @@
 import {
   embed,
   rerank
-} from "./chunk-FHRGC6TP.js";
+} from "./chunk-5V2A3CAH.js";
 import {
   bubbleConfirmPage,
   isAllowedBubbleOrigin
-} from "./chunk-GJATVGBS.js";
+} from "./chunk-TTEJQ5XA.js";
 import {
   DATASETS,
   LIMITS,
@@ -15,12 +15,13 @@ import {
   processExpansion,
   sha256Hex,
   today
-} from "./chunk-ZD6TYLET.js";
+} from "./chunk-FQYHT4VC.js";
 import {
   P
 } from "./chunk-3FYJM7LH.js";
 import {
   hasLane,
+  portBlobs,
   portIndex,
   portModelRunner,
   portStore
@@ -22163,6 +22164,11 @@ var dense = {
       matches = matches.filter((m) => matchesDocScope(m.metadata ?? {}, filters.doc_number));
       if (filters.edition) matches = matches.filter((m) => !m.metadata?.edition || String(m.metadata.edition) === filters.edition);
       console.log("dense scope (in code):", before, "\u2192", matches.length, "in doc", filters.doc_number, filters.edition ? `@${filters.edition}` : "");
+      if (!matches.length) {
+        const wide = await env.VECTORIZE.query(vector, { ...q, topK: THRESHOLDS.denseStarveK });
+        matches = (wide.matches ?? []).filter((m) => matchesDocScope(m.metadata ?? {}, filters.doc_number));
+        console.log("dense scope starvation \u2192 deep re-query:", matches.length, "doc members in the top", THRESHOLDS.denseStarveK);
+      }
       if (filters && filters.edition && matches.length < 3) {
         const docOnly = allMatches.filter((m) => matchesDocScope(m.metadata ?? {}, filters.doc_number));
         if (docOnly.length > matches.length) {
@@ -22921,6 +22927,69 @@ var lexicalRrf = {
   }
 };
 
+// workers/worker_public/src/stages/vlmFigureRerank.ts
+var FIGURE_Q = /\bfig(ure|s)?\b/i;
+function parsePick(text, kept) {
+  const t = text;
+  const s8 = typeof text === "string" ? text : typeof t?.response === "string" ? t.response : t?.choices?.[0]?.message?.content ?? "";
+  const pick = Number(String(s8).match(/"pick"\s*:\s*(\d+)/)?.[1] ?? 0);
+  return pick >= 1 && pick <= kept ? pick : 0;
+}
+var vlmFigureRerank = {
+  name: "vlm-figure-rerank",
+  failure: "additive",
+  when: (c) => FIGURE_Q.test(c.query) && c.finalHits.some((h) => h.metadata.block === "figure" && h.metadata.unit_id),
+  run: async (c) => {
+    const figures = c.finalHits.filter((h) => h.metadata.block === "figure" && h.metadata.unit_id).slice(0, 3);
+    if (!figures.length) return;
+    try {
+      const ids = figures.map((h) => String(h.metadata.unit_id));
+      const ph = ids.map((_, i) => `?${i + 1}`).join(",");
+      const rows = await portStore(c.env).prepare(
+        `SELECT unit_id, json_extract(payload,'$.uri') AS uri FROM unit_payloads WHERE unit_id IN (${ph})`
+      ).bind(...ids).all();
+      const uris = new Map((rows.results ?? []).map((r) => [r.unit_id, r.uri]));
+      const content = [
+        {
+          type: "text",
+          text: `Question: ${c.query}
+
+Which of these figures best answers the question? Reply with ONLY: {"pick": N} using the image's number, or {"pick": 0} if none is relevant.`
+        }
+      ];
+      const kept = [];
+      let n = 0;
+      for (const h of figures) {
+        const uri = String(uris.get(String(h.metadata.unit_id)) ?? "");
+        const m = uri.match(/^\/assets\/(.+)/);
+        if (!m) continue;
+        const obj = await portBlobs(c.env).get(m[1]);
+        if (!obj?.body) continue;
+        const buf = new Uint8Array(await new Response(obj.body).arrayBuffer());
+        let binary = "";
+        for (let i = 0; i < buf.length; i += 8192) binary += String.fromCharCode(...buf.subarray(i, i + 8192));
+        n += 1;
+        content.push({ type: "text", text: `Image ${n}:` });
+        content.push({ type: "image_url", image_url: { url: `data:image/png;base64,${btoa(binary)}` } });
+        kept.push({ h, label: n });
+      }
+      if (!kept.length) return;
+      const raced = await Promise.race([
+        portModelRunner(c.env).run({ model: MODELS.vlmFigure, messages: [{ role: "user", content }], maxTokens: 256, effort: "low" }),
+        new Promise((r) => setTimeout(() => r(null), 15e3))
+      ]);
+      const pick = parsePick(raced?.text ?? raced, kept.length);
+      const chosen = kept.find((k) => k.label === pick);
+      if (!chosen) return;
+      const top = Math.max(...c.finalHits.map((h) => h.rerank_score ?? h.score ?? 0));
+      chosen.h.rerank_score = top + 1;
+      c.finalHits = [chosen.h, ...c.finalHits.filter((h) => h !== chosen.h)];
+      console.log("vlm-figure-rerank: promoted", chosen.h.metadata.unit_id, "over", c.finalHits.length - 1, "text-ranked hits");
+    } catch {
+    }
+  }
+};
+
 // workers/worker_public/src/stages/termNudge.ts
 var termNudge = {
   name: "term-nudge",
@@ -23419,6 +23488,7 @@ var STAGES = [
   typedPin,
   sectionDescent,
   dedup,
+  vlmFigureRerank,
   windowFloor
 ];
 var STAGE_NAMES = STAGES.map((s8) => s8.name);

package/dist/{chunk-FHRGC6TP.js → chunk-5V2A3CAH.js}

@@ -1,7 +1,7 @@
 import {
   answerEffort,
   effortBudget
-} from "./chunk-ZD6TYLET.js";
+} from "./chunk-FQYHT4VC.js";
 
 // workers/worker_public/src/ai.ts
 var delay = (ms) => new Promise((r) => setTimeout(r, ms));

package/dist/{chunk-K5WL3DKH.js → chunk-AP73VIHB.js}

@@ -1,12 +1,12 @@
 import {
   embed
-} from "./chunk-FHRGC6TP.js";
+} from "./chunk-5V2A3CAH.js";
 import {
   standardKeysFrom
-} from "./chunk-D6UUAL7I.js";
+} from "./chunk-OLO45N6D.js";
 import {
   THRESHOLDS
-} from "./chunk-ZD6TYLET.js";
+} from "./chunk-FQYHT4VC.js";
 import {
   P
 } from "./chunk-3FYJM7LH.js";

package/dist/{chunk-ZD6TYLET.js → chunk-FQYHT4VC.js}

@@ -11,6 +11,10 @@ var MODELS = {
   // Vision image-parts land with answer contract v2; text answers work now.
   anon: "@cf/zai-org/glm-5.3-flash",
   member: "@cf/zai-org/glm-5.3-flash",
+  // the VLM figure rerank (TODO.sota/18.3): the newly-landed
+  // vision-capable MoE scores figure IMAGES against figure-seeking
+  // questions — one bounded call per figure question, ablatable stage
+  vlmFigure: "@cf/deepseek-ai/deepseek-v4.1-flash-1017",
   // hot-path understanding/summarize stays on the cheap Qwen (cost-first lane)
   understand: "@cf/qwen/qwen3-30b-a3b-fp8",
   // the api_call act's intent-to-act mapping (TODO.ai-platform/09) — the
@@ -84,6 +88,11 @@ var THRESHOLDS = {
   // the extractive compression arm (TODO.sota/02 row 5): when ablated
   // on, each passage body is compressed to this share of its budget
   compressShare: 0.6,
+  // the dense scope's starvation fallback (the R 134-2 figure question):
+  // when the in-code doc scope cuts the vector-level top-K to zero, the
+  // re-query runs at this depth and scopes in code (the native filter is
+  // dead — measured 2026-10-05)
+  denseStarveK: 1e3,
   entailmentGate: true,
   entailmentBudgetMs: 6e3,
   entailmentSupportedFloor: 0.9,

package/dist/{chunk-D6UUAL7I.js → chunk-OLO45N6D.js}

@@ -1,7 +1,7 @@
 import {
   DATASETS,
   datasetAllowed
-} from "./chunk-ZD6TYLET.js";
+} from "./chunk-FQYHT4VC.js";
 import {
   P
 } from "./chunk-3FYJM7LH.js";

package/dist/{chunk-K7UGBTBZ.js → chunk-SIT4XSD5.js}

@@ -6,24 +6,24 @@ import {
   sessionFrom,
   telemetry,
   understandQuery
-} from "./chunk-E74MNLBC.js";
+} from "./chunk-5AME76UW.js";
 import {
   corsHeaders,
   err,
   json,
   readJson,
   validateQuery
-} from "./chunk-GJATVGBS.js";
+} from "./chunk-TTEJQ5XA.js";
 import {
   entitlementScope,
   standardKeysFrom
-} from "./chunk-D6UUAL7I.js";
+} from "./chunk-OLO45N6D.js";
 import {
   LIMITS,
   MODELS,
   num,
   sha256Hex
-} from "./chunk-ZD6TYLET.js";
+} from "./chunk-FQYHT4VC.js";
 import {
   portModelRunner
 } from "./chunk-LSZTJI7V.js";

package/dist/{chunk-GJATVGBS.js → chunk-TTEJQ5XA.js}

@@ -1,7 +1,7 @@
 import {
   LIMITS,
   sha256Hex
-} from "./chunk-ZD6TYLET.js";
+} from "./chunk-FQYHT4VC.js";
 import {
   P
 } from "./chunk-3FYJM7LH.js";

package/dist/config.d.ts

@@ -3,6 +3,7 @@ export declare const MODELS: {
     readonly rerank: "@cf/baai/bge-reranker-base";
     readonly anon: "@cf/zai-org/glm-5.3-flash";
     readonly member: "@cf/zai-org/glm-5.3-flash";
+    readonly vlmFigure: "@cf/deepseek-ai/deepseek-v4.1-flash-1017";
     readonly understand: "@cf/qwen/qwen3-30b-a3b-fp8";
     readonly acts: "@cf/qwen/qwen3-30b-a3b-fp8";
     readonly fallback: "@cf/qwen/qwen3-30b-a3b-fp8";
@@ -50,6 +51,7 @@ export declare const THRESHOLDS: {
      *  not only the source rung. Fails open (a slow or unparseable check
      *  leaves the source-rung note unchanged). */
     readonly compressShare: 0.6;
+    readonly denseStarveK: 1000;
     readonly entailmentGate: true;
     readonly entailmentBudgetMs: 6000;
     readonly entailmentSupportedFloor: 0.9;

package/dist/search-6KM4P2E5.js

@@ -0,0 +1,14 @@
+import {
+  handleSearch
+} from "./chunk-SIT4XSD5.js";
+import "./chunk-5AME76UW.js";
+import "./chunk-5V2A3CAH.js";
+import "./chunk-TTEJQ5XA.js";
+import "./chunk-ETLSWSZ7.js";
+import "./chunk-OLO45N6D.js";
+import "./chunk-FQYHT4VC.js";
+import "./chunk-3FYJM7LH.js";
+import "./chunk-LSZTJI7V.js";
+export {
+  handleSearch
+};

package/dist/{tools-JZY6GSYC.js → tools-PR7PFAV6.js}

@@ -4,10 +4,10 @@ import {
   parseToolCall,
   runTool,
   toolNote
-} from "./chunk-K5WL3DKH.js";
-import "./chunk-FHRGC6TP.js";
-import "./chunk-D6UUAL7I.js";
-import "./chunk-ZD6TYLET.js";
+} from "./chunk-AP73VIHB.js";
+import "./chunk-5V2A3CAH.js";
+import "./chunk-OLO45N6D.js";
+import "./chunk-FQYHT4VC.js";
 import "./chunk-3FYJM7LH.js";
 import "./chunk-LSZTJI7V.js";
 export {

package/dist/worker_mcp/src/index.js

@@ -1,7 +1,7 @@
 import {
   authenticate
-} from "../../chunk-GJATVGBS.js";
-import "../../chunk-ZD6TYLET.js";
+} from "../../chunk-TTEJQ5XA.js";
+import "../../chunk-FQYHT4VC.js";
 import {
   P,
   setProfile

package/dist/worker_public/src/config.js

@@ -16,7 +16,7 @@ import {
   roleModel,
   sha256Hex,
   today
-} from "../../chunk-ZD6TYLET.js";
+} from "../../chunk-FQYHT4VC.js";
 import "../../chunk-3FYJM7LH.js";
 export {
   DATASETS,

package/dist/worker_public/src/index.js

@@ -4,16 +4,16 @@ import {
   handleMemories,
   scoreFaithfulness,
   scoreJudge
-} from "../../chunk-GVCOGEJC.js";
+} from "../../chunk-2FME6R6J.js";
 import "../../chunk-ECOYO7EX.js";
 import {
   TOOLS_REGISTRY,
   bindModelNode,
   modelGroundingBlock
-} from "../../chunk-K5WL3DKH.js";
+} from "../../chunk-AP73VIHB.js";
 import {
   handleSearch
-} from "../../chunk-K7UGBTBZ.js";
+} from "../../chunk-SIT4XSD5.js";
 import {
   buildMessages,
   citations,
@@ -38,11 +38,11 @@ import {
   sessionFrom,
   telemetry,
   understandQuery
-} from "../../chunk-E74MNLBC.js";
+} from "../../chunk-5AME76UW.js";
 import {
   embed,
   generateOnce
-} from "../../chunk-FHRGC6TP.js";
+} from "../../chunk-5V2A3CAH.js";
 import {
   authenticate,
   corsHeaders,
@@ -51,14 +51,14 @@ import {
   readJson,
   validateQuery,
   withCors
-} from "../../chunk-GJATVGBS.js";
+} from "../../chunk-TTEJQ5XA.js";
 import {
   canonicalRefusal
 } from "../../chunk-ETLSWSZ7.js";
 import {
   entitlementScope,
   standardKeysFrom
-} from "../../chunk-D6UUAL7I.js";
+} from "../../chunk-OLO45N6D.js";
 import {
   LIMITS,
   MODELS,
@@ -69,7 +69,7 @@ import {
   roleModel,
   sha256Hex,
   today
-} from "../../chunk-ZD6TYLET.js";
+} from "../../chunk-FQYHT4VC.js";
 import {
   P,
   setProfile
@@ -1119,13 +1119,13 @@ async function callToolOnce(env, ctx, _req, tier, key, name, args, t0) {
       // stream:false forces the JSON lane (anon defaults to SSE)
       body: JSON.stringify({ ...args, stream: false })
     });
-    const { runTool, TOOLS_REGISTRY: TOOLS_REGISTRY2 } = await import("../../tools-JZY6GSYC.js");
+    const { runTool, TOOLS_REGISTRY: TOOLS_REGISTRY2 } = await import("../../tools-PR7PFAV6.js");
     let payload;
     if (TOOLS_REGISTRY2.some((t) => t.name === name && t.audiences.includes("mcp"))) {
       const r = await runTool({ ...env, DB: env.DB }, { name, args }, "mcp");
       payload = r ? { tool: r.name, query: r.query, result: r.output } : { error: { message: `tool ${name} returned nothing for the given arguments` } };
     } else {
-      const res = name === "ask" ? await (await import("../../ask-B7XXKYZP.js")).handleAsk(env, ctx, inner, tier, key) : await (await import("../../search-IXNVNOOY.js")).handleSearch(env, ctx, inner, tier, key);
+      const res = name === "ask" ? await (await import("../../ask-XJOOMAXU.js")).handleAsk(env, ctx, inner, tier, key) : await (await import("../../search-6KM4P2E5.js")).handleSearch(env, ctx, inner, tier, key);
       payload = await res.json().catch(() => ({ error: { message: "tool transport failed", status: res.status } }));
     }
     telemetry(env, ctx, tier, `mcp:${name}`, null, !payload?.error, JSON.stringify(payload).length, await sha256Hex(`${name}:${JSON.stringify(args ?? {})}`), void 0, void 0, { durationMs: Date.now() - t0 });

package/dist/worker_public/src/requestScope.js

@@ -4,8 +4,8 @@ import {
   requestSalt,
   resolveRequestScope,
   standardKeysFrom
-} from "../../chunk-D6UUAL7I.js";
-import "../../chunk-ZD6TYLET.js";
+} from "../../chunk-OLO45N6D.js";
+import "../../chunk-FQYHT4VC.js";
 import "../../chunk-3FYJM7LH.js";
 export {
   entitlementScope,

package/package.json

@@ -27,7 +27,7 @@
     "playwright": "^1.62.1",
     "yaml": "^2.9.1"
   },
-  "version": "0.2.164",
+  "version": "0.2.166",
   "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/config.ts

@@ -7,6 +7,10 @@ export const MODELS = {
   // Vision image-parts land with answer contract v2; text answers work now.
   anon: "@cf/zai-org/glm-5.3-flash",
   member: "@cf/zai-org/glm-5.3-flash",
+  // the VLM figure rerank (TODO.sota/18.3): the newly-landed
+  // vision-capable MoE scores figure IMAGES against figure-seeking
+  // questions — one bounded call per figure question, ablatable stage
+  vlmFigure: "@cf/deepseek-ai/deepseek-v4.1-flash-1017",
   // hot-path understanding/summarize stays on the cheap Qwen (cost-first lane)
   understand: "@cf/qwen/qwen3-30b-a3b-fp8",
   // the api_call act's intent-to-act mapping (TODO.ai-platform/09) — the
@@ -81,6 +85,11 @@ export const THRESHOLDS = {
   // the extractive compression arm (TODO.sota/02 row 5): when ablated
   // on, each passage body is compressed to this share of its budget
   compressShare: 0.6,
+  // the dense scope's starvation fallback (the R 134-2 figure question):
+  // when the in-code doc scope cuts the vector-level top-K to zero, the
+  // re-query runs at this depth and scopes in code (the native filter is
+  // dead — measured 2026-10-05)
+  denseStarveK: 1000,
   entailmentGate: true,
   entailmentBudgetMs: 6000,
   entailmentSupportedFloor: 0.9,

package/workers/worker_public/src/stages/dense.ts

@@ -3,7 +3,7 @@
 // query, no filter — bit-for-bit same candidates, one serial round-trip
 // saved), the doc/edition-filtered query with its guessed-pin drop and
 // sparse-filter widen, and the plain unfiltered query.
-import { LIMITS } from "../config.ts";
+import { LIMITS, THRESHOLDS } from "../config.ts";
 import { matchesDocScope } from "../context.ts";
 import type { Stage } from "./types.ts";
 
@@ -41,6 +41,20 @@ export const dense: Stage = {
       matches = matches.filter((m) => matchesDocScope(m.metadata ?? {}, filters.doc_number!));
       if (filters.edition) matches = matches.filter((m) => !m.metadata?.edition || String(m.metadata.edition) === filters.edition);
       console.log("dense scope (in code):", before, "→", matches.length, "in doc", filters.doc_number, filters.edition ? `@${filters.edition}` : "");
+      // the in-code starvation fallback (observed live: the R 134-2
+      // figure question — the vector-level top-K held ZERO members of
+      // the asked document, so the scope cut to nothing and the answer
+      // ran on other lanes' leftovers): re-query at a deep topK and
+      // scope in code — the NATIVE metadata filter is dead on this
+      // platform (measured 2026-10-05: filters return empty for values
+      // proven present), so the document's own nearest neighbors are
+      // reachable only through a wider net. Fires only when the scope
+      // starves to zero — zero behavior when the scope matches.
+      if (!matches.length) {
+        const wide: any = await env.VECTORIZE.query(vector, { ...q, topK: THRESHOLDS.denseStarveK });
+        matches = (wide.matches ?? []).filter((m: any) => matchesDocScope(m.metadata ?? {}, filters.doc_number!));
+        console.log("dense scope starvation → deep re-query:", matches.length, "doc members in the top", THRESHOLDS.denseStarveK);
+      }
       // an edition pin corroborated by (almost) nothing means the pin
       // was a guess — understanding emits editions for families it
       // mixes up (observed: R 76 pinned @2021, an R 60 year; the corpus

package/workers/worker_public/src/stages/index.ts

@@ -33,6 +33,7 @@ import { stdRefNudge } from "./stdRefNudge.ts";
 import { overviewDemote } from "./overviewDemote.ts";
 import { familyBoost } from "./familyBoost.ts";
 import { rerankStage, lexicalRrf } from "./rerank.ts";
+import { vlmFigureRerank } from "./vlmFigureRerank.ts";
 import { termNudge } from "./termNudge.ts";
 import { conceptSteer } from "./conceptSteer.ts";
 import { editionSteer } from "./editionSteer.ts";
@@ -75,6 +76,7 @@ export const STAGES: Stage[] = [
   typedPin,
   sectionDescent,
   dedup,
+  vlmFigureRerank,
   windowFloor,
 ];
 

package/workers/worker_public/src/stages/vlmFigureRerank.ts

@@ -0,0 +1,80 @@
+// VLM figure rerank (TODO.sota/18.3 — the hedge that sees past
+// embeddings): a figure-SEEKING question's figure candidates are scored
+// by a vision model on their actual images — the MKO figure chunks'
+// text is thin (a breadcrumb line + a caption), so text ranking loses
+// them to sibling tables no matter what the caption-augmented embed
+// carries. One bounded vision call, the winner promoted to the front of
+// the window with a rerank score above the incumbent top; every failure
+// mode is additive (the text ranking stands). Off via the registry
+// projection like any stage (ablate_stages drops it).
+import { MODELS } from "../config.ts";
+import { portBlobs, portModelRunner, portStore } from "../env.ts";
+import type { Stage } from "./types.ts";
+import type { Hit } from "../../../shared/chunk";
+
+export const FIGURE_Q = /\bfig(ure|s)?\b/i;
+
+export function parsePick(text: unknown, kept: number): number {
+  const t = text as any;
+  const s = typeof text === "string" ? text : typeof t?.response === "string" ? t.response : t?.choices?.[0]?.message?.content ?? "";
+  const pick = Number(String(s).match(/"pick"\s*:\s*(\d+)/)?.[1] ?? 0);
+  return pick >= 1 && pick <= kept ? pick : 0;
+}
+
+export const vlmFigureRerank: Stage = {
+  name: "vlm-figure-rerank",
+  failure: "additive",
+  when: (c) => FIGURE_Q.test(c.query) && c.finalHits.some((h: Hit) => (h.metadata as any).block === "figure" && (h.metadata as any).unit_id),
+  run: async (c) => {
+    const figures = c.finalHits.filter((h: Hit) => (h.metadata as any).block === "figure" && (h.metadata as any).unit_id).slice(0, 3);
+    if (!figures.length) return;
+    try {
+      const ids = figures.map((h: Hit) => String((h.metadata as any).unit_id));
+      const ph = ids.map((_, i) => `?${i + 1}`).join(",");
+      const rows = await portStore(c.env).prepare(
+        `SELECT unit_id, json_extract(payload,'$.uri') AS uri FROM unit_payloads WHERE unit_id IN (${ph})`,
+      )
+        .bind(...ids)
+        .all();
+      const uris = new Map(((rows.results ?? []) as any[]).map((r: any) => [r.unit_id as string, r.uri as string | null]));
+      // the multimodal gotcha (2026-09-09): image parts ride a SHORT
+      // message — the question and the labels, nothing else
+      const content: any[] = [
+        {
+          type: "text",
+          text: `Question: ${c.query}\n\nWhich of these figures best answers the question? Reply with ONLY: {"pick": N} using the image's number, or {"pick": 0} if none is relevant.`,
+        },
+      ];
+      const kept: { h: Hit; label: number }[] = [];
+      let n = 0;
+      for (const h of figures) {
+        const uri = String(uris.get(String((h.metadata as any).unit_id)) ?? "");
+        const m = uri.match(/^\/assets\/(.+)/);
+        if (!m) continue;
+        const obj = await portBlobs(c.env).get(m[1]);
+        if (!obj?.body) continue;
+        const buf = new Uint8Array(await new Response(obj.body).arrayBuffer());
+        let binary = "";
+        for (let i = 0; i < buf.length; i += 8192) binary += String.fromCharCode(...buf.subarray(i, i + 8192));
+        n += 1;
+        content.push({ type: "text", text: `Image ${n}:` });
+        content.push({ type: "image_url", image_url: { url: `data:image/png;base64,${btoa(binary)}` } });
+        kept.push({ h, label: n });
+      }
+      if (!kept.length) return;
+      const raced: any = await Promise.race([
+        portModelRunner(c.env).run({ model: MODELS.vlmFigure, messages: [{ role: "user", content }], maxTokens: 256, effort: "low" }),
+        new Promise<null>((r) => setTimeout(() => r(null), 15000)),
+      ]);
+      const pick = parsePick(raced?.text ?? raced, kept.length);
+      const chosen = kept.find((k) => k.label === pick);
+      if (!chosen) return;
+      const top = Math.max(...c.finalHits.map((h: Hit) => h.rerank_score ?? h.score ?? 0));
+      chosen.h.rerank_score = top + 1;
+      c.finalHits = [chosen.h, ...c.finalHits.filter((h: Hit) => h !== chosen.h)];
+      console.log("vlm-figure-rerank: promoted", (chosen.h.metadata as any).unit_id, "over", c.finalHits.length - 1, "text-ranked hits");
+    } catch {
+      // additive — the text ranking stands
+    }
+  },
+};

package/dist/ask-B7XXKYZP.js

@@ -1,16 +0,0 @@
-import {
-  handleAsk
-} from "./chunk-GVCOGEJC.js";
-import "./chunk-ECOYO7EX.js";
-import "./chunk-K5WL3DKH.js";
-import "./chunk-E74MNLBC.js";
-import "./chunk-FHRGC6TP.js";
-import "./chunk-GJATVGBS.js";
-import "./chunk-ETLSWSZ7.js";
-import "./chunk-D6UUAL7I.js";
-import "./chunk-ZD6TYLET.js";
-import "./chunk-3FYJM7LH.js";
-import "./chunk-LSZTJI7V.js";
-export {
-  handleAsk
-};

package/dist/search-IXNVNOOY.js

@@ -1,14 +0,0 @@
-import {
-  handleSearch
-} from "./chunk-K7UGBTBZ.js";
-import "./chunk-E74MNLBC.js";
-import "./chunk-FHRGC6TP.js";
-import "./chunk-GJATVGBS.js";
-import "./chunk-ETLSWSZ7.js";
-import "./chunk-D6UUAL7I.js";
-import "./chunk-ZD6TYLET.js";
-import "./chunk-3FYJM7LH.js";
-import "./chunk-LSZTJI7V.js";
-export {
-  handleSearch
-};