@nijaru/pi-provider-compaction 0.0.6 → 0.0.8

package/README.md

@@ -1,6 +1,6 @@
 # pi-provider-compaction
 
-Experimental provider-native compaction for [Pi](https://github.com/earendil-works/pi), with a meaningful portable summary saved alongside each native checkpoint. Minimum runtime: Pi 0.87.0. Development checks target Pi 1.2.0.
+Experimental provider-native compaction for [Pi](https://github.com/earendil-works/pi), with a meaningful portable summary saved alongside each native checkpoint. Minimum runtime: Pi 0.87.0. Development checks target Pi 1.3.0.
 
 **This experimental design is not production-validated. Do not enable it globally yet.** Real runner/adapter fixtures and a Codex live smoke test pass; broader backend and deployment validation remains incomplete.
 
@@ -73,7 +73,7 @@ pi --extension ./dist/index.js
 
 The manifest loads the built `dist/index.js`, including a bundled stateless Responses serializer validator; source or host dependency changes require a rebuild. `bun run check` rebuilds first and tests the artifact against Pi's actual host serializer, including function-call history replayed as grammar tools. Keep global installation disabled while testing. Publishing and global activation are separate approval steps.
 
-Tests include real Pi 1.2.0 runner/adapter fixtures for all three protocols, repeated prefix compaction, transformed suffix preservation, unchanged fallback, grammar tools, strict defaults, tool IDs, images and prompt updates. Set `PI_COMPACTOR_FIXTURE_PATH` to a `pi-compactor` entrypoint to check SDK flag precedence in both extension orders. The optional installed-fast-mode fixture (`PI_FAST_MODE_FIXTURE_PATH`) checks actual `pi-fast-mode` composition across `/reload`; it creates an isolated fast-mode configuration.
+Tests include real Pi 1.3.0 runner/adapter fixtures for all three protocols, repeated prefix compaction, transformed suffix preservation, unchanged fallback, grammar tools, strict defaults, tool IDs, images, prompt updates, cross-model thinking and mid-conversation effort changes. Set `PI_COMPACTOR_FIXTURE_PATH` to a `pi-compactor` entrypoint to check SDK flag precedence in both extension orders. The optional installed-fast-mode fixture (`PI_FAST_MODE_FIXTURE_PATH`) checks actual `pi-fast-mode` composition across `/reload`; it creates an isolated fast-mode configuration.
 
 Authenticated protocol smoke tests are separate and send three small requests:
 

package/dist/index.js

@@ -200,6 +200,8 @@ function installObserver(pi, ctx, observe) {
   const id = model.provider;
   const native = registry.getRegisteredNativeProvider(id);
   if (native) {
+    if (typeof native.getModels !== "function" || !native.getModels().some((entry) => entry.api === model.api))
+      return;
     const isInstalled = () => registry.getRegisteredNativeProvider(id) === wrapper;
     const guarded = (...args) => {
       if (!isInstalled())
@@ -637,19 +639,13 @@ function transformMessages(messages, model, normalizeToolCallId) {
       const isSameModel = assistantMsg.provider === model.provider && assistantMsg.api === model.api && assistantMsg.model === model.id;
       const transformedContent = assistantMsg.content.flatMap((block) => {
         if (block.type === "thinking") {
-          if (block.redacted) {
-            return isSameModel ? block : [];
-          }
-          if (isSameModel && block.thinkingSignature)
+          if (!isSameModel)
+            return [];
+          if (block.redacted || block.thinkingSignature)
             return block;
           if (!block.thinking || block.thinking.trim() === "")
             return [];
-          if (isSameModel)
-            return block;
-          return {
-            type: "text",
-            text: block.thinking
-          };
+          return block;
         }
         if (block.type === "text") {
           if (isSameModel)
@@ -840,6 +836,12 @@ function convertResponsesMessages(model, context, allowedToolCallProviders, opti
     });
   };
   const includeInitialSystemMessage = options?.includeSystemPrompt ?? true;
+  const pushConfigurationUpdate = (effort) => {
+    messages.push({
+      type: "configuration_update",
+      reasoning: { effort }
+    });
+  };
   const compat = model.compat;
   const instructionRole = model.reasoning && compat?.supportsDeveloperRole !== false ? "developer" : "system";
   let msgIndex = 0;
@@ -946,6 +948,9 @@ function convertResponsesMessages(model, context, allowedToolCallProviders, opti
       }
       if (output.length === 0)
         continue;
+      if (options?.configurationUpdateEffort !== undefined && isSameModel && msg.providerThinkingLevel !== undefined) {
+        pushConfigurationUpdate(msg.providerThinkingLevel);
+      }
       messages.push(...output);
     } else if (msg.role === "toolResult") {
       const [callId] = msg.toolCallId.split("|");
@@ -967,6 +972,8 @@ function convertResponsesMessages(model, context, allowedToolCallProviders, opti
     if (!isLeadingSystemMessage)
       msgIndex++;
   }
+  if (options?.configurationUpdateEffort !== undefined)
+    pushConfigurationUpdate(options.configurationUpdateEffort);
   return messages;
 }
 function convertResponsesTools(tools, options) {
@@ -1223,6 +1230,15 @@ async function collectProviderUsage(stream, signal) {
     throw new Error("Native compaction provider stream ended without usage");
   return usage;
 }
+
+class NativeCompactionValidationError extends Error {
+  usage;
+  constructor(message, usage) {
+    super(message);
+    this.usage = usage;
+    this.name = "NativeCompactionValidationError";
+  }
+}
 async function requestResponsesCompact(request) {
   let preparedPayload;
   let compacted;
@@ -1249,9 +1265,13 @@ async function requestResponsesCompact(request) {
       if (!response.ok)
         return response;
       const parsed = await readJsonBounded(response, request.signal);
-      if (!isObject(parsed))
-        throw new Error("Responses Compact returned a non-object response");
-      compacted = { response: parsed, output: validateCompactedOutput(parsed) };
+      try {
+        if (!isObject(parsed))
+          throw new Error("Responses Compact returned a non-object response");
+        compacted = { response: parsed, output: validateCompactedOutput(parsed) };
+      } catch (error) {
+        throw new NativeCompactionValidationError(error instanceof Error ? error.message : String(error));
+      }
       return syntheticCompletion(parsed, preparedPayload);
     } catch (error) {
       bridgeError = error;
@@ -1263,6 +1283,7 @@ async function requestResponsesCompact(request) {
     headers: request.headers,
     env: request.env,
     signal: request.signal,
+    ...request.serviceTier === undefined ? {} : { serviceTier: request.serviceTier },
     transport: "sse",
     cacheRetention: "none",
     maxRetries: 0,
@@ -1281,6 +1302,8 @@ async function requestResponsesCompact(request) {
       throw bridgeError;
     throw error;
   });
+  if (bridgeError instanceof NativeCompactionValidationError)
+    throw new NativeCompactionValidationError(bridgeError.message, usage);
   if (bridgeError)
     throw bridgeError;
   if (!compacted || dispatches !== 1)
@@ -1425,6 +1448,7 @@ async function requestRemoteV2(request) {
     headers: request.headers,
     env: request.env,
     signal: request.signal,
+    ...request.serviceTier === undefined ? {} : { serviceTier: request.serviceTier },
     transport: "sse",
     cacheRetention: "none",
     maxRetries: 0,
@@ -1452,7 +1476,11 @@ async function requestRemoteV2(request) {
 }
 async function requestProviderCompaction(request) {
   const result = await (request.protocol === "remote-v2" ? requestRemoteV2(request) : requestResponsesCompact(request));
-  return { ...result, output: validateCompactedOutput(result) };
+  try {
+    return { ...result, output: validateCompactedOutput(result) };
+  } catch (error) {
+    throw new NativeCompactionValidationError(error instanceof Error ? error.message : String(error), result.usage);
+  }
 }
 
 // replay.ts
@@ -1463,7 +1491,7 @@ var SNAPSHOT_TTL_MS = 5 * 60000;
 function equal(left, right) {
   return JSON.stringify(left) === JSON.stringify(right);
 }
-function serializeInput(model, context, declarations = context) {
+function serializeInput(model, context, declarations = context, configurationUpdateEffort) {
   const compat = model.compat;
   const flag = (name) => compat?.[name] === true;
   const codex = model.api === "openai-codex-responses";
@@ -1471,6 +1499,7 @@ function serializeInput(model, context, declarations = context) {
   const tools = resolveTranscript2(declarations, flag("supportsMidConvoSystemMessages"));
   return convertResponsesMessages(model, normalized, model.api === "azure-openai-responses" ? AZURE_TOOL_CALL_PROVIDERS : TOOL_CALL_PROVIDERS, {
     includeSystemPrompt: !codex,
+    configurationUpdateEffort,
     grammarToolInputProperties: createGrammarToolInputProperties(getDeclaredTools(tools.messages), flag("supportsOpenAIGrammarTools")),
     supportsMidConvoSystemMessages: flag("supportsMidConvoSystemMessages"),
     supportsAdditionalTools: flag("supportsAdditionalTools"),
@@ -1495,12 +1524,15 @@ function capturePreparedRequest(model, context, canonical, payload, reject = ()
     reject("snapshot exceeds size limit");
     return;
   }
-  const expected = serializeInput(model, context);
+  const last = payload.input.at(-1);
+  const compat = model.compat;
+  const configurationUpdateEffort = model.api === "openai-responses" && compat?.supportsMidConvoEffort === true && last?.type === "configuration_update" && isObject(last.reasoning) && typeof last.reasoning.effort === "string" ? last.reasoning.effort : undefined;
+  const expected = serializeInput(model, context, context, configurationUpdateEffort);
   if (!equal(expected, payload.input)) {
     reject("prepared input differs from serializer");
     return;
   }
-  return { context: structuredClone(context), canonical: structuredClone(canonical), payload: structuredClone(payload), portableInput: expected, createdAt: Date.now() };
+  return { context: structuredClone(context), canonical: structuredClone(canonical), payload: structuredClone(payload), portableInput: expected, configurationUpdateEffort, createdAt: Date.now() };
 }
 function promptCount(input) {
   let count = 0;
@@ -1508,6 +1540,10 @@ function promptCount(input) {
     count++;
   return count;
 }
+function serializePrefix(model, snapshot, prefix) {
+  const input = serializeInput(model, prefix, snapshot.context, snapshot.configurationUpdateEffort);
+  return snapshot.configurationUpdateEffort === undefined ? input : input.slice(0, -1);
+}
 function closedTools(context) {
   const pending = new Set;
   for (const message of context.messages) {
@@ -1551,7 +1587,7 @@ function selectCoveredPrefix(model, snapshot, prefix, reject = () => {}) {
     reject("projected prefix differs from request context");
     return;
   }
-  const expected = serializeInput(model, mapped, snapshot.context);
+  const expected = serializePrefix(model, snapshot, mapped);
   if (!equal(snapshot.portableInput.slice(0, expected.length), expected)) {
     reject("serialized prefix does not match prepared input");
     return;
@@ -1584,7 +1620,7 @@ function summarySlot(model, snapshot, summaryPrefix) {
   const mapped = mappedPrefix(snapshot, summaryPrefix);
   if (!mapped)
     return;
-  const prefix = serializeInput(model, mapped, snapshot.context);
+  const prefix = serializePrefix(model, snapshot, mapped);
   const index = promptCount(prefix);
   if (prefix.length !== index + 1 || !equal(snapshot.portableInput.slice(0, prefix.length), prefix))
     return;
@@ -1763,7 +1799,8 @@ Last attempt: ${compactionStatus}`, "info");
       return;
     }
     if (!options.onPayload) {
-      captureRejected("no payload callback");
+      if (captureStatus !== "eligible prepared request captured")
+        captureRejected("no payload callback");
       return;
     }
     if (generic(owner)) {
@@ -1782,7 +1819,7 @@ Last attempt: ${compactionStatus}`, "info");
       if (generation !== ticket || ctx !== owner || options.signal?.aborted || !supportedRouteOptions(model, options))
         return false;
       try {
-        return digest(owner.sessionManager.getBranch()) === branchHash;
+        return unchanged(owner.sessionManager.getBranch(), branch.length, branchHash);
       } catch {
         retire();
         return false;
@@ -1814,8 +1851,9 @@ Last attempt: ${compactionStatus}`, "info");
       const routeHash = route(model, options);
       const policyHash = digest(requestPolicy(model, prepared));
       const latest = latestCompaction(branch);
+      const payloadModel = isObject(payload) && typeof payload.model === "string" ? payload.model : undefined;
       let result = payload;
-      if (active && latest?.id === active.entryId && latest.summary === active.summary && digest(latest.details) === active.detailsHash && unchanged(branch, active.branchLength, active.branchHash) && active.routeHash === routeHash && active.policyHash === policyHash) {
+      if (active && latest?.id === active.entryId && latest.summary === active.summary && active.payloadModel === payloadModel && unchanged(branch, active.branchLength, active.branchHash) && active.routeHash === routeHash && active.policyHash === policyHash) {
         const details = readNativeCompactionDetails(latest.details);
         const projection = owner.sessionManager.buildSessionProjection();
         const contribution = projection.entries.find((entry) => entry.sourceEntry.id === latest.id);
@@ -1838,7 +1876,7 @@ Last attempt: ${compactionStatus}`, "info");
         prepared,
         model: structuredClone(model),
         provider,
-        options: { apiKey: options.apiKey, headers: structuredClone(options.headers), env: structuredClone(options.env), fetch: options.fetch },
+        options: { apiKey: options.apiKey, headers: structuredClone(options.headers), env: structuredClone(options.env), fetch: options.fetch, ...typeof options.serviceTier === "string" ? { serviceTier: options.serviceTier } : {} },
         branchLength: branch.length,
         branchHash,
         routeHash,
@@ -1892,6 +1930,7 @@ Last attempt: ${compactionStatus}`, "info");
       removeInvalidation();
   });
   pi.on("session_before_compact", async (event, context) => {
+    compactionStatus = "preparing compaction";
     const capture = snapshot;
     snapshot = undefined;
     pending = undefined;
@@ -1925,7 +1964,7 @@ Last attempt: ${compactionStatus}`, "info");
     const input = selectCoveredPrefix(capture.model, capture.prepared, prefix, fallback);
     if (!input)
       return;
-    const current = () => ticket === generation && !event.signal.aborted && supportedRouteOptions(capture.model, capture.options) && !generic(context) && context.modelRegistry.getProvider(capture.model.provider) === capture.provider && digest(context.sessionManager.getBranch()) === branchHash;
+    const current = () => ticket === generation && !event.signal.aborted && supportedRouteOptions(capture.model, capture.options) && !generic(context) && context.modelRegistry.getProvider(capture.model.provider) === capture.provider && unchanged(context.sessionManager.getBranch(), branch.length, branchHash);
     const auth = await context.modelRegistry.getApiKeyAndHeaders(context.model);
     if (!current())
       return { cancel: true };
@@ -1938,7 +1977,10 @@ Last attempt: ${compactionStatus}`, "info");
     try {
       portable = await compact(event.preparation, capture.model, auth.apiKey, auth.headers, event.customInstructions, event.signal, context.thinkingLevel, capture.provider.streamSimple.bind(capture.provider), auth.env, undefined, undefined, capture.sessionId);
     } catch {
-      return !current() ? { cancel: true } : undefined;
+      if (!current())
+        return { cancel: true };
+      fallback("portable summary generation failed");
+      return;
     }
     if (!current() || !portable.summary.trim() || portable.summary === NATIVE_COMPACTION_SUMMARY)
       return { cancel: true };
@@ -1962,7 +2004,8 @@ Last attempt: ${compactionStatus}`, "info");
         headers: capture.options.headers,
         env: capture.options.env,
         signal: event.signal,
-        fetch: capture.options.fetch
+        fetch: capture.options.fetch,
+        ...capture.options.serviceTier === undefined ? {} : { serviceTier: capture.options.serviceTier }
       });
       if (!current())
         return { cancel: true };
@@ -1989,11 +2032,13 @@ Last attempt: ${compactionStatus}`, "info");
       };
       pending = { details, capture, summary: portable.summary };
       return { compaction: { ...portable, usage: combineUsage(native.usage, portable.usage), details } };
-    } catch {
+    } catch (error) {
       if (!current())
         return { cancel: true };
       fallback("native acquisition or validation failed");
-      return { compaction: portable };
+      return {
+        compaction: error instanceof NativeCompactionValidationError && error.usage !== undefined ? { ...portable, usage: combineUsage(error.usage, portable.usage) } : portable
+      };
     }
   });
   pi.on("session_compact", (event, context) => {
@@ -2011,9 +2056,18 @@ Last attempt: ${compactionStatus}`, "info");
       return;
     const branch = context.sessionManager.getBranch();
     const coverage = candidate.details.coverage;
-    if (!coverage || event.compactionEntry.firstKeptEntryId !== coverage.firstKeptEntryId || event.compactionEntry.parentId !== coverage.sourceLeafId || branch.at(-1)?.id !== event.compactionEntry.id || digest(branch.slice(0, -1)) !== coverage.sourceHash)
+    const index = branch.findIndex((entry) => entry.id === event.compactionEntry.id);
+    if (!coverage || index < 0 || event.compactionEntry.firstKeptEntryId !== coverage.firstKeptEntryId || event.compactionEntry.parentId !== coverage.sourceLeafId || digest(branch.slice(0, index)) !== coverage.sourceHash)
       return;
-    active = { entryId: event.compactionEntry.id, detailsHash: digest(candidate.details), summary: candidate.summary, branchLength: branch.length, branchHash: digest(branch), routeHash: candidate.capture.routeHash, policyHash: candidate.capture.policyHash };
+    active = {
+      entryId: event.compactionEntry.id,
+      summary: candidate.summary,
+      branchLength: index + 1,
+      branchHash: digest(branch.slice(0, index + 1)),
+      routeHash: candidate.capture.routeHash,
+      policyHash: candidate.capture.policyHash,
+      ...typeof candidate.capture.prepared.payload.model === "string" ? { payloadModel: candidate.capture.prepared.payload.model } : {}
+    };
   });
   pi.on("session_compact_failed", () => {
     pending = undefined;

package/index.ts

@@ -5,7 +5,7 @@ import { compact, convertToLlm, type ExtensionAPI, type ExtensionContext, type S
 import { resolveCompactionModelPolicy } from "./policy";
 import { installObserver, type InstalledObserver, type ObserveRequest } from "./provider";
 import { capturePreparedRequest, equal, requestPolicy, selectCoveredPrefix, substituteSummary, summarySlot, type PreparedRequest } from "./replay";
-import { isObject, requestProviderCompaction, resolveNativeProtocol, supportsResponsesApi, validateCompactedOutput, type NativeProtocol, type ResponseItem } from "./protocol";
+import { isObject, NativeCompactionValidationError, requestProviderCompaction, resolveNativeProtocol, supportsResponsesApi, validateCompactedOutput, type NativeProtocol, type ResponseItem } from "./protocol";
 
 export const NATIVE_COMPACTION_TYPE = "pi-provider-compaction/openai-responses";
 export const NATIVE_COMPACTION_VERSION = 3;
@@ -124,7 +124,7 @@ interface Captured {
 	prepared: PreparedRequest;
 	model: Model<any>;
 	provider: Provider;
-	options: ProviderStreamOptions | SimpleStreamOptions;
+	options: (ProviderStreamOptions | SimpleStreamOptions) & { serviceTier?: string };
 	branchLength: number;
 	branchHash: string;
 	routeHash: string;
@@ -134,12 +134,13 @@ interface Captured {
 }
 interface Active {
 	entryId: string;
-	detailsHash: string;
 	summary: string;
 	branchLength: number;
 	branchHash: string;
 	routeHash: string;
 	policyHash: string;
+	/** Model field of the prepared payload; a request hook rewrite invalidates native state. */
+	payloadModel?: string;
 }
 
 export default function (pi: ExtensionAPI): void {
@@ -190,7 +191,12 @@ export default function (pi: ExtensionAPI): void {
 		}
 		if (options.sessionId !== manager.getSessionId()) return;
 		if (!supportsNativeCompaction(model)) { captureRejected("unsupported API"); return; }
-		if (!options.onPayload) { captureRejected("no payload callback"); return; }
+		if (!options.onPayload) {
+			// The portable-summary request re-enters this observer without a payload
+			// callback. Do not overwrite the main request's capture diagnostic.
+			if (captureStatus !== "eligible prepared request captured") captureRejected("no payload callback");
+			return;
+		}
 		if (generic(owner)) { captureRejected("generic compaction model takes precedence"); return; }
 		// The Azure adapter also accepts API-specific overrides and process-env
 		// fallbacks. Those must not silently select a route absent from our evidence.
@@ -201,7 +207,7 @@ export default function (pi: ExtensionAPI): void {
 		const canonical = transcript(manager.buildSessionProjection().messages);
 		const current = () => {
 			if (generation !== ticket || ctx !== owner || options.signal?.aborted || !supportedRouteOptions(model, options)) return false;
-			try { return digest(owner.sessionManager.getBranch()) === branchHash; }
+			try { return unchanged(owner.sessionManager.getBranch(), branch.length, branchHash); }
 			catch { retire(); return false; }
 		};
 		return async (payload) => {
@@ -225,8 +231,9 @@ export default function (pi: ExtensionAPI): void {
 			const routeHash = route(model, options);
 			const policyHash = digest(requestPolicy(model, prepared));
 			const latest = latestCompaction(branch);
+			const payloadModel = isObject(payload) && typeof payload.model === "string" ? payload.model : undefined;
 			let result = payload;
-			if (active && latest?.id === active.entryId && latest.summary === active.summary && digest(latest.details) === active.detailsHash && unchanged(branch, active.branchLength, active.branchHash) && active.routeHash === routeHash && active.policyHash === policyHash) {
+			if (active && latest?.id === active.entryId && latest.summary === active.summary && active.payloadModel === payloadModel && unchanged(branch, active.branchLength, active.branchHash) && active.routeHash === routeHash && active.policyHash === policyHash) {
 				const details = readNativeCompactionDetails(latest.details);
 				const projection = owner.sessionManager.buildSessionProjection();
 				const contribution = projection.entries.find((entry) => entry.sourceEntry.id === latest.id);
@@ -244,7 +251,7 @@ export default function (pi: ExtensionAPI): void {
 			captureStatus = "eligible prepared request captured";
 			snapshot = {
 				prepared, model: structuredClone(model), provider,
-				options: { apiKey: options.apiKey, headers: structuredClone(options.headers), env: structuredClone(options.env), fetch: options.fetch },
+				options: { apiKey: options.apiKey, headers: structuredClone(options.headers), env: structuredClone(options.env), fetch: options.fetch, ...(typeof (options as { serviceTier?: unknown }).serviceTier === "string" ? { serviceTier: (options as { serviceTier: string }).serviceTier } : {}) },
 				branchLength: branch.length, branchHash, routeHash, authHash: routeDigest(auth), policyHash,
 				sessionId: owner.sessionManager.getSessionId(),
 			};
@@ -281,6 +288,7 @@ export default function (pi: ExtensionAPI): void {
 	pi.on("session_shutdown", () => { if (typeof removeInvalidation === "function") removeInvalidation(); });
 
 	pi.on("session_before_compact", async (event, context) => {
+		compactionStatus = "preparing compaction";
 		const capture = snapshot;
 		snapshot = undefined; pending = undefined;
 		if (!capture) { fallback(`no eligible snapshot (${captureStatus})`); return; }
@@ -297,7 +305,7 @@ export default function (pi: ExtensionAPI): void {
 		const prefix = transcript(projection.entries.slice(0, boundary).flatMap((entry) => entry.messages));
 		const input = selectCoveredPrefix(capture.model, capture.prepared, prefix, fallback);
 		if (!input) return;
-		const current = () => ticket === generation && !event.signal.aborted && supportedRouteOptions(capture.model, capture.options) && !generic(context) && context.modelRegistry.getProvider(capture.model.provider) === capture.provider && digest(context.sessionManager.getBranch()) === branchHash;
+		const current = () => ticket === generation && !event.signal.aborted && supportedRouteOptions(capture.model, capture.options) && !generic(context) && context.modelRegistry.getProvider(capture.model.provider) === capture.provider && unchanged(context.sessionManager.getBranch(), branch.length, branchHash);
 		const auth = await context.modelRegistry.getApiKeyAndHeaders(context.model);
 		if (!current()) return { cancel: true };
 		if (!auth.ok || routeDigest(auth) !== capture.authHash) { fallback("auth changed since capture"); return; }
@@ -307,7 +315,9 @@ export default function (pi: ExtensionAPI): void {
 			portable = await compact(event.preparation, capture.model, auth.apiKey, auth.headers as Record<string, string> | undefined, event.customInstructions, event.signal, context.thinkingLevel, capture.provider.streamSimple.bind(capture.provider), auth.env, undefined, undefined, capture.sessionId);
 		} catch {
 			// No result exists to return; let Pi own its ordinary failure/fallback behavior.
-			return !current() ? { cancel: true } : undefined;
+			if (!current()) return { cancel: true };
+			fallback("portable summary generation failed");
+			return undefined;
 		}
 		if (!current() || !portable.summary.trim() || portable.summary === NATIVE_COMPACTION_SUMMARY) return { cancel: true };
 		try {
@@ -320,6 +330,7 @@ export default function (pi: ExtensionAPI): void {
 				input, preparedPayload: capture.prepared.payload, protocol,
 				apiKey: capture.options.apiKey, headers: capture.options.headers, env: capture.options.env,
 				signal: event.signal, fetch: capture.options.fetch,
+				...(capture.options.serviceTier === undefined ? {} : { serviceTier: capture.options.serviceTier }),
 			});
 			if (!current()) return { cancel: true };
 			const finalAuth = await context.modelRegistry.getApiKeyAndHeaders(context.model);
@@ -334,12 +345,18 @@ export default function (pi: ExtensionAPI): void {
 			};
 			pending = { details, capture, summary: portable.summary };
 			return { compaction: { ...portable, usage: combineUsage(native.usage, portable.usage), details } };
-		} catch {
+		} catch (error) {
 			if (!current()) return { cancel: true };
 			// Do not expose exception text: provider errors may contain request data.
 			fallback("native acquisition or validation failed");
-			// The portable request already succeeded. Never pay for a duplicate summary.
-			return { compaction: portable };
+			// The portable request already succeeded. Never pay for a duplicate summary,
+			// and keep the native call's usage when the endpoint answered but produced
+			// an unusable window.
+			return {
+				compaction: error instanceof NativeCompactionValidationError && error.usage !== undefined
+					? { ...portable, usage: combineUsage(error.usage, portable.usage) }
+					: portable,
+			};
 		}
 	});
 	pi.on("session_compact", (event, context) => {
@@ -352,8 +369,17 @@ export default function (pi: ExtensionAPI): void {
 		if (!candidate || !equal(event.compactionEntry.details, candidate.details) || event.compactionEntry.summary !== candidate.summary) return;
 		const branch = context.sessionManager.getBranch();
 		const coverage = candidate.details.coverage;
-		if (!coverage || event.compactionEntry.firstKeptEntryId !== coverage.firstKeptEntryId || event.compactionEntry.parentId !== coverage.sourceLeafId || branch.at(-1)?.id !== event.compactionEntry.id || digest(branch.slice(0, -1)) !== coverage.sourceHash) return;
-		active = { entryId: event.compactionEntry.id, detailsHash: digest(candidate.details), summary: candidate.summary, branchLength: branch.length, branchHash: digest(branch), routeHash: candidate.capture.routeHash, policyHash: candidate.capture.policyHash };
+		const index = branch.findIndex((entry) => entry.id === event.compactionEntry.id);
+		if (!coverage || index < 0 || event.compactionEntry.firstKeptEntryId !== coverage.firstKeptEntryId || event.compactionEntry.parentId !== coverage.sourceLeafId || digest(branch.slice(0, index)) !== coverage.sourceHash) return;
+		active = {
+			entryId: event.compactionEntry.id,
+			summary: candidate.summary,
+			branchLength: index + 1,
+			branchHash: digest(branch.slice(0, index + 1)),
+			routeHash: candidate.capture.routeHash,
+			policyHash: candidate.capture.policyHash,
+			...(typeof candidate.capture.prepared.payload.model === "string" ? { payloadModel: candidate.capture.prepared.payload.model } : {}),
+		};
 	});
 	pi.on("session_compact_failed", () => { pending = undefined; snapshot = undefined; compactionStatus = "compaction failed or cancelled"; });
 }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@nijaru/pi-provider-compaction",
-  "version": "0.0.6",
+  "version": "0.0.8",
   "type": "module",
   "description": "Provider-native context compaction adapters for Pi",
   "keywords": [
@@ -44,9 +44,9 @@
     "LICENSE"
   ],
   "devDependencies": {
-    "@earendil-works/pi-ai": "1.2.0",
-    "@earendil-works/pi-agent-core": "1.2.0",
-    "@earendil-works/pi-coding-agent": "1.2.0",
+    "@earendil-works/pi-ai": "1.3.0",
+    "@earendil-works/pi-agent-core": "1.3.0",
+    "@earendil-works/pi-coding-agent": "1.3.0",
     "@types/bun": "latest",
     "oxlint": "^1.81.0",
     "typescript": "^6.0.3"

package/protocol.ts

@@ -225,10 +225,20 @@ interface TransportRequest {
 	apiKey?: string;
 	headers?: ProviderHeaders;
 	env?: Record<string, string>;
+	/** Cost-relevant request option captured with the original request. */
+	serviceTier?: string;
 	signal: AbortSignal;
 	fetch?: typeof globalThis.fetch;
 }
 
+/** The provider answered and was billed, but its window cannot be committed. */
+export class NativeCompactionValidationError extends Error {
+	constructor(message: string, readonly usage?: Usage) {
+		super(message);
+		this.name = "NativeCompactionValidationError";
+	}
+}
+
 export interface TransportResponse {
 	output: ResponseItem[];
 	usage: Usage;
@@ -258,8 +268,12 @@ async function requestResponsesCompact(request: TransportRequest): Promise<Trans
 			});
 			if (!response.ok) return response;
 			const parsed = await readJsonBounded(response, request.signal);
-			if (!isObject(parsed)) throw new Error("Responses Compact returned a non-object response");
-			compacted = { response: parsed, output: validateCompactedOutput(parsed) };
+			try {
+				if (!isObject(parsed)) throw new Error("Responses Compact returned a non-object response");
+				compacted = { response: parsed, output: validateCompactedOutput(parsed) };
+			} catch (error) {
+				throw new NativeCompactionValidationError(error instanceof Error ? error.message : String(error));
+			}
 			return syntheticCompletion(parsed, preparedPayload);
 		} catch (error) {
 			bridgeError = error;
@@ -272,6 +286,7 @@ async function requestResponsesCompact(request: TransportRequest): Promise<Trans
 		headers: request.headers,
 		env: request.env,
 		signal: request.signal,
+		...(request.serviceTier === undefined ? {} : { serviceTier: request.serviceTier }),
 		transport: "sse",
 		cacheRetention: "none",
 		maxRetries: 0,
@@ -288,6 +303,7 @@ async function requestResponsesCompact(request: TransportRequest): Promise<Trans
 		if (bridgeError) throw bridgeError;
 		throw error;
 	});
+	if (bridgeError instanceof NativeCompactionValidationError) throw new NativeCompactionValidationError(bridgeError.message, usage);
 	if (bridgeError) throw bridgeError;
 	if (!compacted || dispatches !== 1) throw new Error("Responses Compact did not complete exactly one request");
 	return { output: compacted.output, usage };
@@ -410,6 +426,7 @@ async function requestRemoteV2(request: TransportRequest): Promise<TransportResp
 		headers: request.headers,
 		env: request.env,
 		signal: request.signal,
+		...(request.serviceTier === undefined ? {} : { serviceTier: request.serviceTier }),
 		transport: "sse",
 		cacheRetention: "none",
 		maxRetries: 0,
@@ -436,5 +453,9 @@ export async function requestProviderCompaction(
 	request: TransportRequest & { protocol: NativeProtocol },
 ): Promise<TransportResponse> {
 	const result = await (request.protocol === "remote-v2" ? requestRemoteV2(request) : requestResponsesCompact(request));
-	return { ...result, output: validateCompactedOutput(result) };
+	try {
+		return { ...result, output: validateCompactedOutput(result) };
+	} catch (error) {
+		throw new NativeCompactionValidationError(error instanceof Error ? error.message : String(error), result.usage);
+	}
 }

package/provider.ts

@@ -59,6 +59,11 @@ export function installObserver(pi: ExtensionAPI, ctx: ExtensionContext, observe
 	const id = model.provider;
 	const native = registry.getRegisteredNativeProvider(id);
 	if (native) {
+		// Pi composes a native base only when its catalog serves the active API.
+		// Otherwise the request goes to the global API provider and bypasses this
+		// wrapper; claiming an observer there would report native compaction as
+		// installed while never capturing anything.
+		if (typeof native.getModels !== "function" || !native.getModels().some((entry) => entry.api === model.api)) return;
 		const isInstalled = () => registry.getRegisteredNativeProvider(id) === wrapper;
 		const guarded: ObserveRequest = (...args) => {
 			if (!isInstalled()) return;

package/replay.ts

@@ -14,7 +14,7 @@ export function equal(left: unknown, right: unknown): boolean {
 }
 
 /** Validator only. No serialized item from this function is transmitted. */
-export function serializeInput(model: Model<any>, context: TranscriptContext, declarations = context): ResponseItem[] {
+export function serializeInput(model: Model<any>, context: TranscriptContext, declarations = context, configurationUpdateEffort?: string): ResponseItem[] {
 	const compat = model.compat as Record<string, unknown> | undefined;
 	const flag = (name: string) => compat?.[name] === true;
 	const codex = model.api === "openai-codex-responses";
@@ -22,6 +22,7 @@ export function serializeInput(model: Model<any>, context: TranscriptContext, de
 	const tools = resolveTranscript(declarations, flag("supportsMidConvoSystemMessages"));
 	return convertResponsesMessages(model, normalized, model.api === "azure-openai-responses" ? AZURE_TOOL_CALL_PROVIDERS : TOOL_CALL_PROVIDERS, {
 		includeSystemPrompt: !codex,
+		configurationUpdateEffort,
 		grammarToolInputProperties: createGrammarToolInputProperties(getDeclaredTools(tools.messages), flag("supportsOpenAIGrammarTools")),
 		supportsMidConvoSystemMessages: flag("supportsMidConvoSystemMessages"),
 		supportsAdditionalTools: flag("supportsAdditionalTools"),
@@ -40,6 +41,8 @@ export interface PreparedRequest {
 	payload: JsonObject & { input: ResponseItem[] };
 	/** The portable wire input before our sole summary substitution. */
 	portableInput: ResponseItem[];
+	/** The final request-only effort update; historical efforts come from the transcript. */
+	configurationUpdateEffort?: string;
 	replacement?: { index: number; count: number };
 	createdAt: number;
 }
@@ -52,9 +55,16 @@ export function capturePreparedRequest(
 	// Reject stateful provider requests: coverage must be the complete visible input.
 	if (payload.previous_response_id || payload.conversation) { reject("stateful request"); return; }
 	if (Buffer.byteLength(JSON.stringify({ context, payload })) > MAX_SNAPSHOT_BYTES) { reject("snapshot exceeds size limit"); return; }
-	const expected = serializeInput(model, context);
+	// Read the actual prepared request's mode, not provider/auth guesses. Exact conversion
+	// below still validates every historical update, its position and its complete shape.
+	const last = payload.input.at(-1);
+	const compat = model.compat as Record<string, unknown> | undefined;
+	const configurationUpdateEffort = model.api === "openai-responses" && compat?.supportsMidConvoEffort === true &&
+		last?.type === "configuration_update" && isObject(last.reasoning) && typeof last.reasoning.effort === "string"
+		? last.reasoning.effort : undefined;
+	const expected = serializeInput(model, context, context, configurationUpdateEffort);
 	if (!equal(expected, payload.input)) { reject("prepared input differs from serializer"); return; }
-	return { context: structuredClone(context), canonical: structuredClone(canonical), payload: structuredClone(payload) as PreparedRequest["payload"], portableInput: expected, createdAt: Date.now() };
+	return { context: structuredClone(context), canonical: structuredClone(canonical), payload: structuredClone(payload) as PreparedRequest["payload"], portableInput: expected, configurationUpdateEffort, createdAt: Date.now() };
 }
 
 function promptCount(input: readonly ResponseItem[]): number {
@@ -63,6 +73,13 @@ function promptCount(input: readonly ResponseItem[]): number {
 	return count;
 }
 
+function serializePrefix(model: Model<any>, snapshot: PreparedRequest, prefix: TranscriptContext): ResponseItem[] {
+	const input = serializeInput(model, prefix, snapshot.context, snapshot.configurationUpdateEffort);
+	// The converter appends the next reply's effort even to a partial transcript.
+	// It is not historical coverage and must remain in the live suffix on replay.
+	return snapshot.configurationUpdateEffort === undefined ? input : input.slice(0, -1);
+}
+
 /** No open/synthetic tool exchange may cross the coverage boundary. */
 function closedTools(context: TranscriptContext): boolean {
 	const pending = new Set<string>();
@@ -94,7 +111,7 @@ export function selectCoveredPrefix(model: Model<any>, snapshot: PreparedRequest
 	if (!closedTools(prefix)) { reject("prefix has an incomplete tool exchange or failed assistant"); return; }
 	const mapped = mappedPrefix(snapshot, prefix);
 	if (!mapped) { reject("projected prefix differs from request context"); return; }
-	const expected = serializeInput(model, mapped, snapshot.context);
+	const expected = serializePrefix(model, snapshot, mapped);
 	if (!equal(snapshot.portableInput.slice(0, expected.length), expected)) { reject("serialized prefix does not match prepared input"); return; }
 	const start = promptCount(expected);
 	let end = expected.length;
@@ -121,7 +138,7 @@ export function substituteSummary(
 export function summarySlot(model: Model<any>, snapshot: PreparedRequest, summaryPrefix: TranscriptContext): number | undefined {
 	const mapped = mappedPrefix(snapshot, summaryPrefix);
 	if (!mapped) return;
-	const prefix = serializeInput(model, mapped, snapshot.context);
+	const prefix = serializePrefix(model, snapshot, mapped);
 	const index = promptCount(prefix);
 	if (prefix.length !== index + 1 || !equal(snapshot.portableInput.slice(0, prefix.length), prefix)) return;
 	return index;