@konneal/engine 0.2.175 → 0.2.176

package/dist/ablate.d.ts

@@ -13,11 +13,16 @@ export interface AblateConfig {
      *  diversified subset + one strong verifier. Off in serving until the
      *  gate promotes it; the grid arms it per ask. */
     speculative: boolean;
-    /** Extractive prompt compression (TODO.sota/02 row 5): each passage
-     *  body is query-weighted sentence selection at a fraction of its
-     *  budget before the answer model reads it. Off in serving; the grid
-     *  arms it per ask (the row's accuracy-at-token-budget gate). */
+    /** Extractive prompt compression (TODO.sota/02 row 5): PROMOTED — the
+     *  passage bodies are query-weighted sentence selection at a fraction
+     *  of their budget in serving (the row's gate: identical accuracy at
+     *  ~1/20th the prompt, the whole-document baseline measured it).
+     *  ablate_compress survives for the grid's explicit arm;
+     *  ablate_no_compress is the measurement stand-down. */
     compress: boolean;
+    /** Per-request stand-down for the promoted compressor (the
+     *  measurement arm; the serving default is on). */
+    noCompress: boolean;
     /** The decision pass (TODO.decisions/01): per-request stand-down for
      *  the measurement — the serving switch is DECISIONS_ENABLED. */
     noDecisions: boolean;

package/dist/{ask-7TNQCRD6.js → ask-ZWZFZBKZ.js}

@@ -1,6 +1,6 @@
 import {
   handleAsk
-} from "./chunk-ZZLUGHZ2.js";
+} from "./chunk-T6T2QUO6.js";
 import "./chunk-ECOYO7EX.js";
 import "./chunk-B2MU4ZWM.js";
 import "./chunk-CLDMP5YH.js";

package/dist/{chunk-ZZLUGHZ2.js → chunk-T6T2QUO6.js}

@@ -1404,10 +1404,11 @@ function parseAblate(body, adminToken, presentedToken, knownStages) {
   const rawRoute = body?.ablate_route;
   const speculative = body?.ablate_speculative === true;
   const compress = body?.ablate_compress === true;
+  const noCompress = body?.ablate_no_compress === true;
   const noDecisions = body?.ablate_no_decisions === true;
   const namesAblation = rawStages !== void 0 && rawStages !== null;
   const routeAblation = rawRoute !== void 0 && rawRoute !== null && rawRoute !== "adaptive";
-  if (!namesAblation && !noVerdict && !routeAblation && !speculative && !compress && !noDecisions) return { ok: true, config: null };
+  if (!namesAblation && !noVerdict && !routeAblation && !speculative && !compress && !noCompress && !noDecisions) return { ok: true, config: null };
   if (!adminToken || presentedToken !== adminToken) {
     return {
       ok: false,
@@ -1438,7 +1439,7 @@ function parseAblate(body, adminToken, presentedToken, knownStages) {
     }
     stages = [...new Set(rawStages)];
   }
-  return { ok: true, config: { stages, noVerdict, route, speculative, compress, noDecisions } };
+  return { ok: true, config: { stages, noVerdict, route, speculative, compress, noCompress, noDecisions } };
 }
 
 // workers/worker_public/src/decisions.ts
@@ -2837,7 +2838,7 @@ Answer account questions from these records ONLY: name the record when you use i
       [ops.note, processNote, eNote, regNote, contextNote(declaredCtx, docScope), accountNote, modelNote, vocabNote, memNote, machineNote, conditionNote, aggregationNote, boundaryNote, licenseNote, ...retrieved.notes ?? []].filter(Boolean).join("\n") || void 0,
       summary,
       budget,
-      ablate?.compress === true
+      ablate?.noCompress !== true
     );
     const decisionPromise = decisionsArmed && decisionsGatePromise ? decisionsGatePromise.then((g) => g != null && g >= THRESHOLDS.decisionsGateMin ? runDecisionPass(env, {
       question: q2.query,
@@ -3175,7 +3176,7 @@ Answer account questions from these records ONLY: name the record when you use i
           ...routeSubset ? { stageSubset: routeSubset ?? void 0 } : {}
         });
         if (retryRetrieve.hits.length > 0) {
-          const { messages: retryMessages, usedHits: retryUsed } = buildMessages(q2.query, retryRetrieve.hits, q2.lang, keptHistory, void 0, summary, budget, ablate?.compress === true);
+          const { messages: retryMessages, usedHits: retryUsed } = buildMessages(q2.query, retryRetrieve.hits, q2.lang, keptHistory, void 0, summary, budget, ablate?.noCompress !== true);
           const retryAnswer = await generateOnce(env, model, retryMessages, routeEffort(routed.route));
           if (retryAnswer) {
             answer = canonicalRefusal(retryAnswer, refusalScope(answerCites(retryAnswer)));

package/dist/pipeline.d.ts

@@ -51,9 +51,10 @@ export declare function splitHistory(history: HistoryTurn[], budgetTokens: numbe
  *  incoming order (timeout/parse failure never blocks serving). */
 export declare function listwiseRerank(env: any, model: string, query: string, hits: Hit[]): Promise<Hit[] | null>;
 export declare function buildMessages(query: string, hits: Hit[], lang?: string, history?: HistoryTurn[], retrievalNote?: string, conversationSummary?: string, budgetTokens?: number, 
-/** the extractive compression arm (TODO.sota/02 row 5): ablate_compress
- *  arms it per ask — each passage body becomes query-weighted sentence
- *  selection at compressShare of its budget. Off in serving. */
+/** the extractive compression (TODO.sota/02 row 5): PROMOTED — on in
+ *  serving; ablate_no_compress stands it down per ask for the
+ *  measurement. Each passage body becomes query-weighted sentence
+ *  selection at compressShare of its budget. */
 compress?: boolean): BuiltMessages;
 export declare function citations(hits: Hit[]): {
     doc_id: string;

package/dist/worker_public/src/index.js

@@ -4,7 +4,7 @@ import {
   handleMemories,
   scoreFaithfulness,
   scoreJudge
-} from "../../chunk-ZZLUGHZ2.js";
+} from "../../chunk-T6T2QUO6.js";
 import "../../chunk-ECOYO7EX.js";
 import {
   TOOLS_REGISTRY,
@@ -1181,7 +1181,7 @@ async function callToolOnce(env, ctx, _req, tier, key, name, args, t0) {
       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 ${name} returned nothing for the given arguments` } };
     } else {
-      const res = name === "ask" ? await (await import("../../ask-7TNQCRD6.js")).handleAsk(env, ctx, inner, tier, key) : await (await import("../../search-KNPPTHMG.js")).handleSearch(env, ctx, inner, tier, key);
+      const res = name === "ask" ? await (await import("../../ask-ZWZFZBKZ.js")).handleAsk(env, ctx, inner, tier, key) : await (await import("../../search-KNPPTHMG.js")).handleSearch(env, ctx, inner, tier, key);
       payload = await res.json().catch(() => ({ error: { message: "tool transport failed", status: res.status } }));
     }
     telemetry(env, ctx, tier, `mcp:${name}`, null, !payload?.error, JSON.stringify(payload).length, await sha256Hex(`${name}:${JSON.stringify(args ?? {})}`), void 0, void 0, { durationMs: Date.now() - t0 });

package/package.json

@@ -27,7 +27,7 @@
     "playwright": "^1.62.1",
     "yaml": "^2.9.1"
   },
-  "version": "0.2.175",
+  "version": "0.2.176",
   "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/ablate.ts

@@ -34,11 +34,16 @@ export interface AblateConfig {
    *  diversified subset + one strong verifier. Off in serving until the
    *  gate promotes it; the grid arms it per ask. */
   speculative: boolean;
-  /** Extractive prompt compression (TODO.sota/02 row 5): each passage
-   *  body is query-weighted sentence selection at a fraction of its
-   *  budget before the answer model reads it. Off in serving; the grid
-   *  arms it per ask (the row's accuracy-at-token-budget gate). */
+  /** Extractive prompt compression (TODO.sota/02 row 5): PROMOTED — the
+   *  passage bodies are query-weighted sentence selection at a fraction
+   *  of their budget in serving (the row's gate: identical accuracy at
+   *  ~1/20th the prompt, the whole-document baseline measured it).
+   *  ablate_compress survives for the grid's explicit arm;
+   *  ablate_no_compress is the measurement stand-down. */
   compress: boolean;
+  /** Per-request stand-down for the promoted compressor (the
+   *  measurement arm; the serving default is on). */
+  noCompress: boolean;
   /** The decision pass (TODO.decisions/01): per-request stand-down for
    *  the measurement — the serving switch is DECISIONS_ENABLED. */
   noDecisions: boolean;
@@ -64,11 +69,12 @@ export function parseAblate(
   const rawRoute = body?.ablate_route;
   const speculative = body?.ablate_speculative === true;
   const compress = body?.ablate_compress === true;
+  const noCompress = body?.ablate_no_compress === true;
   const noDecisions = body?.ablate_no_decisions === true;
   const namesAblation = rawStages !== undefined && rawStages !== null;
   const routeAblation = rawRoute !== undefined && rawRoute !== null && rawRoute !== "adaptive";
 
-  if (!namesAblation && !noVerdict && !routeAblation && !speculative && !compress && !noDecisions) return { ok: true, config: null };
+  if (!namesAblation && !noVerdict && !routeAblation && !speculative && !compress && !noCompress && !noDecisions) return { ok: true, config: null };
 
   if (!adminToken || presentedToken !== adminToken) {
     return {
@@ -103,5 +109,5 @@ export function parseAblate(
     stages = [...new Set(rawStages as string[])];
   }
 
-  return { ok: true, config: { stages, noVerdict, route, speculative, compress, noDecisions } };
+  return { ok: true, config: { stages, noVerdict, route, speculative, compress, noCompress, noDecisions } };
 }

package/workers/worker_public/src/ask.ts

@@ -1507,7 +1507,7 @@ let convEntities: Array<{ entity: string; kind: string }> = [];
     [ops.note, processNote, eNote, regNote, contextNote(declaredCtx, docScope), accountNote, modelNote, vocabNote, memNote, machineNote, conditionNote, aggregationNote, boundaryNote, licenseNote, ...(retrieved.notes ?? [])].filter(Boolean).join("\n") || undefined,
     summary,
     budget,
-    ablate?.compress === true,
+    ablate?.noCompress !== true,
   );
   // the decision pass (TODO.decisions/01): starts here so its calls run
   // concurrently with generation — the checklist section arrives the
@@ -2005,7 +2005,7 @@ let convEntities: Array<{ entity: string; kind: string }> = [];
         ...(routeSubset ? { stageSubset: routeSubset ?? undefined } : {}),
       });
       if (retryRetrieve.hits.length > 0) {
-        const { messages: retryMessages, usedHits: retryUsed } = buildMessages(q.query, retryRetrieve.hits, q.lang, keptHistory, undefined, summary, budget, ablate?.compress === true);
+        const { messages: retryMessages, usedHits: retryUsed } = buildMessages(q.query, retryRetrieve.hits, q.lang, keptHistory, undefined, summary, budget, ablate?.noCompress !== true);
         const retryAnswer = await generateOnce(env, model, retryMessages, routeEffort(routed.route));
         // the answer now comes from the retry passages — citations must follow
         if (retryAnswer) {

package/workers/worker_public/src/pipeline.ts

@@ -288,9 +288,10 @@ export function buildMessages(
   retrievalNote?: string,
   conversationSummary?: string,
   budgetTokens: number = LIMITS.inputTokenBudget,
-  /** the extractive compression arm (TODO.sota/02 row 5): ablate_compress
-   *  arms it per ask — each passage body becomes query-weighted sentence
-   *  selection at compressShare of its budget. Off in serving. */
+  /** the extractive compression (TODO.sota/02 row 5): PROMOTED — on in
+   *  serving; ablate_no_compress stands it down per ask for the
+   *  measurement. Each passage body becomes query-weighted sentence
+   *  selection at compressShare of its budget. */
   compress?: boolean,
 ): BuiltMessages {
   // per-corpus guidance travels WITH the dataset (config.ts): every