Merge main: wave-4 backend (activity abort endpoint, lastStep capture)

src/agent/prompts/20-jobs.md

@@ -27,3 +27,7 @@ Dispatch as many research jobs as you genuinely need — in one turn or across t
 parallel when the questions are independent — but don't fan out on one question; the
 researcher searches multiple angles on its own. If research fails, you'll get a failure
 message: say so plainly and continue with what you can establish, flagging the uncertainty.
+
+A job can also be **aborted by the learner** — you'll get a one-line "was aborted"
+message instead of the result. Acknowledge it in one sentence and continue teaching
+without it (offering to re-dispatch is fine if the learner wants the result).

src/agent/subagents/aborts.ts

@@ -0,0 +1,50 @@
+/**
+ * Job abort registry — WHO can actually kill a running sub-agent.
+ *
+ * The runner creates one AbortController per running job and registers it
+ * here; the abort path (src/server/jobs.ts abortJob, driven by the dock's
+ * END button via POST /api/jobs/[jobId]/abort) fires it through fireJobAbort.
+ * Two module trees meet: the runner (src/agent) and the endpoint
+ * (src/server) — the registry lives on globalThis for the same reason as
+ * the turn registry and the wake pipeline: Next dev gives each route entry
+ * its own module instances, so a module-level Map would be a different
+ * object in each. Placed under src/agent (server→agent imports are an
+ * established direction; agent→server is not — see chat-handler).
+ *
+ * 2026-09 abort semantics (user-decided): REAL KILL, not mark-and-abandon —
+ * the controller aborts the in-flight model call (agent.generate /
+ * generateText abortSignal); headless tool runs are best-effort (the next
+ * model call is where a hung one dies). The route owns the bookkeeping
+ * (status error "Aborted by learner" + the abort steer); the runner only unwinds.
+ */
+
+interface LhAbortGlobal {
+	__lhJobAborts?: Map<string, AbortController>;
+}
+
+function registry(): Map<string, AbortController> {
+	const g = globalThis as typeof globalThis & LhAbortGlobal;
+	return (g.__lhJobAborts ??= new Map());
+}
+
+/** The runner registers its controller at job start (idempotent overwrite). */
+export function registerJobAbort(jobId: string, controller: AbortController): void {
+	registry().set(jobId, controller);
+}
+
+/** The runner unregisters in its finally block (job settled or unwound). */
+export function unregisterJobAbort(jobId: string): void {
+	registry().delete(jobId);
+}
+
+/** Abort a running job's loop. Returns false when no controller is live —
+ * the job's loop is not executing (crashed, or in the delivering-result
+ * phase where the runner has already returned); the caller then owns the
+ * whole abort on the record side. */
+export function fireJobAbort(jobId: string): boolean {
+	const controller = registry().get(jobId);
+	if (!controller) return false;
+	registry().delete(jobId);
+	controller.abort();
+	return true;
+}

src/agent/subagents/demo-builder.ts

@@ -1,7 +1,7 @@
 import { ToolLoopAgent, stepCountIs, tool, type LanguageModel, type ToolSet } from "ai";
 import { z } from "zod";
 import { modelFor } from "../models.ts";
-import type { DemoResult } from "../../core/contracts.ts";
+import type { DemoResult, UsageStepLike } from "../../core/contracts.ts";
 import { getHeadless, type DemoRunResult, type HeadlessApi } from "./headless.ts";
 
 /**
@@ -32,6 +32,10 @@ export interface DemoBuilderOptions {
 	maxSteps?: number;
 	/** Live activity-line callback (updates the job record). */
 	onActivity?: (line: string) => void;
+	/** Kills the run mid-model-call (2026-09 real-kill abort semantics). */
+	abortSignal?: AbortSignal;
+	/** Money channel: one call per agent step, folded into JobRecord.usage. */
+	onUsage?: (usage: UsageStepLike) => void;
 }
 
 const INSTRUCTIONS = `You are a demo builder — a specialist sub-agent inside a learning application. You
@@ -189,6 +193,8 @@ export async function runDemoBuilder(input: DemoBuilderInput, opts: DemoBuilderO
 	try {
 		const result = await agent.generate({
 			prompt,
+			abortSignal: opts.abortSignal,
+			onStepFinish: ({ usage }) => opts.onUsage?.(usage),
 			onToolExecutionStart: ({ toolCall }) => {
 				opts.onActivity?.(toolCall.toolName);
 			},

src/agent/subagents/extraction.ts

@@ -1,7 +1,15 @@
 import { ToolLoopAgent, stepCountIs, tool, type LanguageModel, type UIMessage } from "ai";
 import { z } from "zod";
 import { modelFor } from "../models.ts";
-import type { AuditResult, ChatRecord, GraphNode, SessionFindings } from "../../core/contracts.ts";
+import type {
+	AuditResult,
+	CallUsage,
+	ChatRecord,
+	ExtractionRunUsage,
+	ExtractionStageUsage,
+	GraphNode,
+	SessionFindings,
+} from "../../core/contracts.ts";
 import { getGraph, mergeFindings, type EmissionStats, type GraphDeps } from "../../core/graph.ts";
 
 /**
@@ -211,12 +219,6 @@ function createExtractor(model: LanguageModel) {
 			}),
 		},
 		stopWhen: stepCountIs(4),
-		// One line per step: a background worker's progress must be legible
-		// from the server log (§5 observability baseline).
-		onStepFinish: (step) =>
-			console.info(
-				`[extraction] extractor step ${step.stepNumber}: ${step.finishReason}, tools: ${step.toolCalls.map((c) => c.toolName).join(", ") || "(none)"}`,
-			),
 	});
 }
 
@@ -300,10 +302,6 @@ function createAuditor(model: LanguageModel) {
 			}),
 		},
 		stopWhen: stepCountIs(4),
-		onStepFinish: (step) =>
-			console.info(
-				`[extraction] auditor step ${step.stepNumber}: ${step.finishReason}, tools: ${step.toolCalls.map((c) => c.toolName).join(", ") || "(none)"}`,
-			),
 	});
 }
 
@@ -312,11 +310,18 @@ function createAuditor(model: LanguageModel) {
 /** The pipeline phases, in order — what the home status endpoint shows. */
 export type ExtractionPhase = "extracting" | "auditing" | "merging";
 
-/** One in-flight pipeline run. Module-level: one process, one registry. */
-interface RunningExtraction {
+/** One in-flight pipeline run. Module-level: one process, one registry.
+ *  Exported for the status payload's consumer typing (the dock). */
+export interface RunningExtraction {
 	chatId: string;
 	phase: ExtractionPhase;
 	startedAt: number;
+	/** Live per-stage usage, accumulated as steps bill (the dock's cost /
+	 *  context-window blobs read this through /api/graph/status). */
+	usage: { extractor?: ExtractionStageUsage; auditor?: ExtractionStageUsage };
+	/** True-cancellation handle (the dock's abort button): one controller per
+	 *  run, shared across phases; abort() cancels the in-flight model call. */
+	controller: AbortController;
 }
 
 const runningExtractions = new Map<string, RunningExtraction>();
@@ -326,6 +331,18 @@ export function extractionStatuses(): RunningExtraction[] {
 	return [...runningExtractions.values()].sort((a, b) => a.startedAt - b.startedAt);
 }
 
+/** True cancellation of the in-flight run for a chat (the dock's abort
+ *  button): the pending model call receives the signal — the provider
+ *  cancels the request, spend stops — and the run ends like a failure:
+ *  nothing merged, nothing checkpointed, partial usage still persisted.
+ *  Returns false when no run is in flight (idempotent no-op). */
+export function abortExtraction(chatId: string): boolean {
+	const run = runningExtractions.get(chatId);
+	if (!run) return false;
+	run.controller.abort();
+	return true;
+}
+
 // ─── the pipeline ───────────────────────────────────────────────────────────
 
 export interface ExtractionModels {
@@ -350,21 +367,62 @@ export function createExtractionPipeline(deps: ExtractionPipelineDeps): Extracti
 	const inFlight = new Set<string>();
 
 	function setPhase(chatId: string, phase: ExtractionPhase): void {
+		const prev = runningExtractions.get(chatId);
 		runningExtractions.set(chatId, {
 			chatId,
 			phase,
-			startedAt: runningExtractions.get(chatId)?.startedAt ?? Date.now(),
+			startedAt: prev?.startedAt ?? Date.now(),
+			usage: prev?.usage ?? {}, // phase changes never lose usage
+			controller: prev?.controller ?? new AbortController(),
 		});
 	}
 
+	/** Accumulate one billed step into the run's registry entry (live). */
+	function recordStep(chatId: string, stage: "extractor" | "auditor", usage: CallUsage, model: string): void {
+		const run = runningExtractions.get(chatId);
+		if (!run) return; // not running (tests driving stages directly)
+		run.usage[stage] ??= { model, steps: [] };
+		run.usage[stage]!.steps.push(usage);
+	}
+
+	/** The SDK's per-step usage, flattened to the persisted CallUsage fields
+	 *  (the money channel's capture shape, contracts/chat.ts). */
+	function callUsage(usage: {
+		inputTokens: number | undefined;
+		outputTokens: number | undefined;
+		inputTokenDetails?: { cacheReadTokens?: number; cacheWriteTokens?: number };
+		outputTokenDetails?: { reasoningTokens?: number };
+	}): CallUsage {
+		// Optional-chained: the SDK types the detail objects as required, but
+		// providers (and mocks) may leave them unset.
+		return {
+			inputTokens: usage.inputTokens,
+			outputTokens: usage.outputTokens,
+			cacheReadTokens: usage.inputTokenDetails?.cacheReadTokens,
+			cacheWriteTokens: usage.inputTokenDetails?.cacheWriteTokens,
+			reasoningTokens: usage.outputTokenDetails?.reasoningTokens,
+		};
+	}
+
 	async function extract(chatId: string): Promise<SessionFindings> {
 		const record = chatStore.getChat(chatId);
 		if (!record) throw new Error(`Chat not found: ${chatId}`);
+		const signal = runningExtractions.get(chatId)?.controller.signal;
+		signal?.throwIfAborted(); // an abort during a sync gap ends the run here
 		const view = transcriptView(record);
 		console.info(`[extraction ${chatId}] extractor starting (${view.split("\n").length} transcript lines)`);
 		const t0 = performance.now();
 
-		const extraction = await extractorAgent.generate({ prompt: `Transcript:\n\n${view}` });
+		const extraction = await extractorAgent.generate({
+			prompt: `Transcript:\n\n${view}`,
+			abortSignal: signal,
+			onStepFinish: (step) => {
+				recordStep(chatId, "extractor", callUsage(step.usage), step.model.modelId);
+				console.info(
+					`[extraction] extractor step ${step.stepNumber}: ${step.finishReason}, tools: ${step.toolCalls.map((c) => c.toolName).join(", ") || "(none)"}`,
+				);
+			},
+		});
 		for (const part of extraction.content) {
 			if (part.type === "tool-result" && part.toolName === "submit_findings") {
 				console.info(`[extraction ${chatId}] extractor done in ${Math.round(performance.now() - t0)}ms`);
@@ -377,6 +435,8 @@ export function createExtractionPipeline(deps: ExtractionPipelineDeps): Extracti
 	async function audit(chatId: string, findings: SessionFindings): Promise<AuditResult> {
 		const record = chatStore.getChat(chatId);
 		if (!record) throw new Error(`Chat not found: ${chatId}`);
+		const signal = runningExtractions.get(chatId)?.controller.signal;
+		signal?.throwIfAborted(); // an abort during the extract→audit gap ends here
 		const view = transcriptView(record);
 		const graph = getGraph(graphStore);
 		console.info(
@@ -389,6 +449,13 @@ export function createExtractionPipeline(deps: ExtractionPipelineDeps): Extracti
 			prompt:
 				`Transcript:\n\n${view}\n\nProposed findings:\n\n${JSON.stringify(findings, null, 2)}\n\n` +
 				`Current knowledge graph (${graph.nodes.length} nodes):\n\n${graphView(graph.nodes)}`,
+			abortSignal: signal,
+			onStepFinish: (step) => {
+				recordStep(chatId, "auditor", callUsage(step.usage), step.model.modelId);
+				console.info(
+					`[extraction] auditor step ${step.stepNumber}: ${step.finishReason}, tools: ${step.toolCalls.map((c) => c.toolName).join(", ") || "(none)"}`,
+				);
+			},
 		});
 		let auditResult: AuditResult | null = null;
 		for (const part of review.content) {
@@ -412,6 +479,25 @@ export function createExtractionPipeline(deps: ExtractionPipelineDeps): Extracti
 		return auditResult;
 	}
 
+	/** Append the run's usage record to the chat's extractions log — success
+	 *  or failure: a failed run still spent money, and its partial usage is
+	 *  exactly what the record should show (stats absent = never merged).
+	 *  Idempotent per startedAt (the store dedupes). */
+	function persistRunUsage(chatId: string, durationMs: number, stats: EmissionStats | undefined): void {
+		const run = runningExtractions.get(chatId);
+		if (!run) return;
+		chatStore.appendExtractionUsage(chatId, {
+			startedAt: run.startedAt,
+			at: new Date().toISOString(),
+			durationMs,
+			stats,
+			stages: {
+				extractor: run.usage.extractor ?? { model: "unknown", steps: [] },
+				auditor: run.usage.auditor ?? { model: "unknown", steps: [] },
+			},
+		});
+	}
+
 	async function runExtraction(chatId: string): Promise<EmissionStats> {
 		if (inFlight.has(chatId)) return { nodes: 0, edges: 0, events: 0 };
 		inFlight.add(chatId);
@@ -426,12 +512,22 @@ export function createExtractionPipeline(deps: ExtractionPipelineDeps): Extracti
 			// The merger re-reads the chat record itself — events are rebuilt
 			// from the transcript, never from stale copies.
 			const stats = mergeFindings(chatId, auditResult, { chatStore, graphStore });
+			persistRunUsage(chatId, Math.round(performance.now() - t0), stats);
 			console.info(
 				`[extraction ${chatId}] merged: ${stats.nodes} nodes, ${stats.edges} edges, ${stats.events} events ` +
 					`(${Math.round(performance.now() - t0)}ms total)`,
 			);
 			return stats;
 		} catch (err) {
+			persistRunUsage(chatId, Math.round(performance.now() - t0), undefined);
+			// Aborted (the dock's abort button): a deliberate stop, not a
+			// failure — same ledger discipline as a failure (nothing merged,
+			// nothing checkpointed, partial usage persisted), but the run
+			// reports zero emissions instead of throwing.
+			if (err instanceof Error && err.name === "AbortError") {
+				console.info(`[extraction ${chatId}] aborted (${Math.round(performance.now() - t0)}ms) — nothing merged, nothing checkpointed`);
+				return { nodes: 0, edges: 0, events: 0 };
+			}
 			// The throw travels to the caller; this line is what makes the
 			// run's log legible end-to-end (no hanging "pipeline start").
 			console.error(`[extraction ${chatId}] pipeline failed (${Math.round(performance.now() - t0)}ms):`, err);

src/agent/subagents/runner.ts

@@ -1,16 +1,21 @@
 import { isToolUIPart, type LanguageModel, type UIMessage } from "ai";
 import {
+	accumulateJobUsage,
 	buildSubagentSteer,
 	type DemoResult,
 	type DemoToolInput,
 	type JobKind,
 	type JobRecord,
+	type JobUsage,
 	type ResearchResult,
 	type ResearchToolInput,
 	type SubagentSteerMessage,
+	type UsageStepLike,
 	type VisualResult,
 	type VisualToolInput,
 } from "../../core/contracts.ts";
+import { registerJobAbort, unregisterJobAbort } from "./aborts.ts";
+import { MODELS } from "../../../config.ts";
 import type { ChatStore, JobStore } from "../../core/ports.ts";
 import { runDemoBuilder } from "./demo-builder.ts";
 import { createResearcherAgent } from "./researcher.ts";
@@ -67,16 +72,70 @@ export interface JobRunner {
 	runJob(jobId: string, input?: ResearchToolInput | DemoToolInput | VisualToolInput): Promise<void>;
 }
 
+/** One job loop's injected machinery: the kill switch (real abort, 2026-09
+ * semantics) and the usage tap (money channel — every model call folds into
+ * JobRecord.usage as it completes). */
+interface LoopContext {
+	/** Fires when abortJob kills the run; abortSignal is passed into every
+	 * model call so the spend stops mid-call, not at the next loop check. */
+	abortSignal: AbortSignal;
+	/** Fold one model call's usage into the job record (progress write). */
+	onUsage: (step: UsageStepLike) => void;
+}
+
+/** The OpenRouter model id for a job kind — JobUsage.model. The kind→role
+ * table mirrors MODELS in config.ts (research→researcher, demo→demoBuilder,
+ * visual→visualMaker); the id is taken from config so a price-table sync
+ * re-prices jobs for free. Model overrides (tests) still bill under the
+ * configured model — usage is what is being captured, not which instance. */
+const KIND_MODEL: Record<JobKind, string> = {
+	research: MODELS.researcher.model,
+	demo: MODELS.demoBuilder.model,
+	visual: MODELS.visualMaker.model,
+};
+
 export function createJobRunner(deps: JobRunnerDeps): JobRunner {
 	const { chatStore, jobStore, deliver } = deps;
 
+	/** Running usage totals per live job (the loop's onUsage folds here,
+	 * then writes through the job store so the poll sees progress). */
+	const usageTotals = new Map<string, JobUsage>();
+
+	function makeLoopContext(jobId: string, kind: JobKind): LoopContext {
+		const model = KIND_MODEL[kind];
+		// Real-kill wiring (2026-09 abort semantics): the controller is what
+		// the abort endpoint fires (agent/subagents/aborts.ts registry); the
+		// signal rides into every model call of the loop.
+		const controller = new AbortController();
+		registerJobAbort(jobId, controller);
+		return {
+			abortSignal: controller.signal,
+			onUsage: (step) => {
+				const next = accumulateJobUsage(usageTotals.get(jobId), model, step);
+				usageTotals.set(jobId, next);
+				jobStore.updateJob(jobId, { usage: next }); // merge patch — status untouched
+			},
+		};
+	}
+
+	/** True while the job may still write terminal state (result, error,
+	 * steer). False once abortJob has flipped it to error ("Aborted by learner") —
+	 * the runner then unwinds silently; the abort path owns the record and
+	 * the abort steer. */
+	function stillRunning(jobId: string): boolean {
+		return jobStore.getJob(jobId)?.status === "running";
+	}
+
 	/** The kind→runner map: one sub-agent loop per job kind. */
 	// Each kind's runner narrows the union input itself (dispatch-time inputs
 	// are kind-correct by construction; transcript fallbacks are cast after).
-	const KIND_RUNNERS: Record<JobKind, (jobId: string, input?: JobToolInput) => Promise<void>> = {
-		research: (jobId, input) => runResearchJob(jobId, researchModelOverride.current, input),
-		demo: (jobId, input) => runDemoJob(jobId, demoModelOverride.current, input),
-		visual: (jobId, input) => runVisualJob(jobId, visualModelOverride.current, input),
+	const KIND_RUNNERS: Record<
+		JobKind,
+		(jobId: string, input: JobToolInput | undefined, loop: LoopContext) => Promise<void>
+	> = {
+		research: (jobId, input, loop) => runResearchJob(jobId, researchModelOverride.current, input, loop),
+		demo: (jobId, input, loop) => runDemoJob(jobId, demoModelOverride.current, input, loop),
+		visual: (jobId, input, loop) => runVisualJob(jobId, visualModelOverride.current, input, loop),
 	};
 
 	async function runJob(jobId: string, input?: JobToolInput): Promise<void> {
@@ -92,7 +151,17 @@ export function createJobRunner(deps: JobRunnerDeps): JobRunner {
 			jobStore.updateJob(jobId, { status: "error", error: `unknown kind ${job.kind}`, activity: "failed" });
 			return;
 		}
-		await run(jobId, input);
+		try {
+			await run(jobId, input, makeLoopContext(jobId, job.kind));
+		} finally {
+			unregisterJobAbort(jobId);
+			// The last usage write may have been mid-flight when the loop died;
+			// persist what was folded (the abort path patches only status, so
+			// this never resurrects an aborted job — merge patch).
+			const usage = usageTotals.get(jobId);
+			if (usage) jobStore.updateJob(jobId, { usage });
+			usageTotals.delete(jobId);
+		}
 	}
 
 	/** Resolve the job's tool input: dispatch passes it directly; only a
@@ -132,6 +201,10 @@ export function createJobRunner(deps: JobRunnerDeps): JobRunner {
 	async function finishJob(jobId: string, result: ResearchResult | DemoResult | VisualResult): Promise<void> {
 		const job = jobStore.getJob(jobId);
 		if (!job) return;
+		// Aborted mid-run or mid-delivery (abortJob flips the record before
+		// the loop unwinds): no result write, no steer — the abort path owns
+		// the record and queues the abort steer itself.
+		if (job.status !== "running") return;
 		jobStore.updateJob(jobId, { result, activity: "delivering result" });
 		const steer = buildSubagentSteer(job, result);
 		try {
@@ -157,7 +230,12 @@ export function createJobRunner(deps: JobRunnerDeps): JobRunner {
 
 	// ─── research ───────────────────────────────────────────────────────────
 
-	async function runResearchJob(jobId: string, model?: LanguageModel, passedInput?: JobToolInput): Promise<void> {
+	async function runResearchJob(
+		jobId: string,
+		model: LanguageModel | undefined,
+		passedInput: JobToolInput | undefined,
+		loop: LoopContext,
+	): Promise<void> {
 		const ctx = jobContext(jobId);
 		if (!ctx) return;
 		const { job } = ctx;
@@ -181,6 +259,10 @@ export function createJobRunner(deps: JobRunnerDeps): JobRunner {
 			jobStore.updateJob(jobId, { activity: "researching" });
 			const result = await agent.generate({
 				prompt,
+				abortSignal: loop.abortSignal,
+				// Money channel: each agent step is one model call — fold its
+				// usage into the record as it completes (progress write).
+				onStepFinish: ({ usage }) => loop.onUsage(usage),
 				onToolExecutionStart: ({ toolCall }) => {
 					const input = toolCall.input as { query?: string; url?: string } | undefined;
 					if (toolCall.toolName === "web_fetch" && typeof input?.url === "string") {
@@ -195,15 +277,24 @@ export function createJobRunner(deps: JobRunnerDeps): JobRunner {
 			const research: ResearchResult = { brief, sources: parseSources(brief, fetchedUrls) };
 			await finishJob(jobId, research);
 		} catch (err) {
+			// Real-kill abort (2026-09): the abort path flipped the record and
+			// queued the abort steer — unwind silently, keep the usage
+			// folded so far (flushed by runJob's finally).
+			if (loop.abortSignal.aborted) return;
 			const reason = err instanceof Error ? err.message : String(err);
-			jobStore.updateJob(jobId, { error: reason });
+			if (stillRunning(jobId)) jobStore.updateJob(jobId, { error: reason });
 			await finishJob(jobId, { brief: `Research failed: ${reason}`, sources: [] });
 		}
 	}
 
 	// ─── demo + visual ──────────────────────────────────────────────────────
 
-	async function runDemoJob(jobId: string, model?: LanguageModel, passedInput?: JobToolInput): Promise<void> {
+	async function runDemoJob(
+		jobId: string,
+		model: LanguageModel | undefined,
+		passedInput: JobToolInput | undefined,
+		loop: LoopContext,
+	): Promise<void> {
 		const ctx = jobContext(jobId);
 		if (!ctx) return;
 		const { job } = ctx;
@@ -218,19 +309,28 @@ export function createJobRunner(deps: JobRunnerDeps): JobRunner {
 				{ title, brief },
 				{
 					model,
+					abortSignal: loop.abortSignal,
+					onUsage: loop.onUsage,
 					onActivity: (line) => jobStore.updateJob(jobId, { activity: line }),
 				},
 			);
-			if (isDemoFailure(result)) jobStore.updateJob(jobId, { error: result.notes ?? "demo build failed" });
+			if (isDemoFailure(result) && stillRunning(jobId))
+				jobStore.updateJob(jobId, { error: result.notes ?? "demo build failed" });
 			await finishJob(jobId, result);
 		} catch (err) {
+			if (loop.abortSignal.aborted) return;
 			const reason = err instanceof Error ? err.message : String(err);
-			jobStore.updateJob(jobId, { error: reason });
+			if (stillRunning(jobId)) jobStore.updateJob(jobId, { error: reason });
 			await finishJob(jobId, { title, html: "", tested: true, notes: `FAILED: ${reason}` });
 		}
 	}
 
-	async function runVisualJob(jobId: string, model?: LanguageModel, passedInput?: JobToolInput): Promise<void> {
+	async function runVisualJob(
+		jobId: string,
+		model: LanguageModel | undefined,
+		passedInput: JobToolInput | undefined,
+		loop: LoopContext,
+	): Promise<void> {
 		const ctx = jobContext(jobId);
 		if (!ctx) return;
 		const { job } = ctx;
@@ -244,14 +344,18 @@ export function createJobRunner(deps: JobRunnerDeps): JobRunner {
 				{ brief },
 				{
 					model,
+					abortSignal: loop.abortSignal,
+					onUsage: loop.onUsage,
 					onActivity: (line) => jobStore.updateJob(jobId, { activity: line }),
 				},
 			);
-			if (isVisualFailure(result)) jobStore.updateJob(jobId, { error: result.notes ?? "visual build failed" });
+			if (isVisualFailure(result) && stillRunning(jobId))
+				jobStore.updateJob(jobId, { error: result.notes ?? "visual build failed" });
 			await finishJob(jobId, result);
 		} catch (err) {
+			if (loop.abortSignal.aborted) return;
 			const reason = err instanceof Error ? err.message : String(err);
-			jobStore.updateJob(jobId, { error: reason });
+			if (stillRunning(jobId)) jobStore.updateJob(jobId, { error: reason });
 			await finishJob(jobId, { svg: "", notes: `FAILED: ${reason}` });
 		}
 	}

src/agent/subagents/visual-maker.ts

@@ -1,7 +1,7 @@
 import { generateText, tool, type LanguageModel, type ModelMessage, type ToolResultPart } from "ai";
 import { z } from "zod";
 import { modelFor } from "../models.ts";
-import type { VisualResult } from "../../core/contracts.ts";
+import type { UsageStepLike, VisualResult } from "../../core/contracts.ts";
 import { getHeadless, type HeadlessApi } from "./headless.ts";
 
 /**
@@ -37,6 +37,11 @@ export interface VisualMakerOptions {
 	maxIterations?: number;
 	/** Live activity-line callback (updates the job record). */
 	onActivity?: (line: string) => void;
+	/** Kills the run mid-model-call (2026-09 real-kill abort semantics). */
+	abortSignal?: AbortSignal;
+	/** Money channel: one call per loop iteration (generateText), folded
+	 * into JobRecord.usage. */
+	onUsage?: (usage: UsageStepLike) => void;
 }
 
 const INSTRUCTIONS = `You are a visual maker — a diagram author + renderer for precise, spatial figures. You
@@ -135,7 +140,15 @@ export async function runVisualMaker(input: VisualMakerInput, opts: VisualMakerO
 
 	try {
 		for (let iteration = 0; iteration < maxIterations; iteration++) {
-			const result = await generateText({ model: opts.model ?? modelFor("visualMaker"), system: INSTRUCTIONS, messages, tools });
+			const result = await generateText({
+				model: opts.model ?? modelFor("visualMaker"),
+				system: INSTRUCTIONS,
+				messages,
+				tools,
+				abortSignal: opts.abortSignal,
+			});
+			// Money channel: each iteration is one model call.
+			opts.onUsage?.(result.usage);
 			messages = [...messages, ...result.response.messages];
 
 			const toolCalls = result.toolCalls;

src/core/contracts/chat.ts

@@ -9,6 +9,7 @@
 import type { UIMessage } from "ai";
 import type { Phase3 } from "../phases.ts";
 import type { PlanNodeStatus, PlanTree } from "./plan.ts";
+import type { EmissionStats } from "./graph.ts";
 
 // ─── Chat records (the session store's on-disk format) ───────────────────────
 
@@ -32,6 +33,41 @@ export interface ChatRecord {
 	 * each entry onto its message; aggregates are folds over this log
 	 * (core/prices.ts). */
 	turns?: TurnUsage[];
+	/** Append-only background-extraction usage log (2026-09 observability):
+	 *  one entry per finished pipeline run on THIS chat — extraction is real
+	 *  spend too. Same parallel-log discipline as `turns` (nothing rides
+	 *  inside messages). Keyed by startedAt; idempotent appends. */
+	extractions?: ExtractionRunUsage[];
+}
+
+// ─── Background-extraction usage (the money channel, extraction side) ────────
+
+/** One pipeline stage's model usage — the stage's model + every step it
+ *  billed (CallUsage per agent step). Steps, not a sum: cost and context
+ *  folds happen at read time (core/prices.ts). */
+export interface ExtractionStageUsage {
+	/** The OpenRouter model id that ran this stage. */
+	model: string;
+	steps: CallUsage[];
+}
+
+/** One finished (or failed mid-flight) extraction run's usage record.
+ *  Appended to the chat's `extractions` log when the run ends — success or
+ *  failure, because money was spent either way; `stats` is present only
+ *  when the merger actually ran. */
+export interface ExtractionRunUsage {
+	/** The run's registry startedAt (ms) — the idempotency key. */
+	startedAt: number;
+	at: string;
+	durationMs: number;
+	/** What the run merged into the graph; absent when it failed first. */
+	stats?: EmissionStats;
+	/** Model usage per stage, in pipeline order (merger is deterministic —
+	 *  no model call, no entry). */
+	stages: {
+		extractor: ExtractionStageUsage;
+		auditor: ExtractionStageUsage;
+	};
 }
 
 // ─── Usage accounting (the money channel) ─────────────────────────────────────

src/core/contracts/graph.ts

@@ -22,6 +22,17 @@
  *  strength — strength is derived at read time from topology. */
 export type EdgeType = "follows" | "supports";
 
+/** What one extraction run merged into the graph ledgers — the merger's
+ *  output, also persisted in extraction-run usage records (contracts/chat). */
+export interface EmissionStats {
+	/** New graph nodes appended. */
+	nodes: number;
+	/** New edge assertions appended. */
+	edges: number;
+	/** Ledger events rebuilt (quiz outcomes + observations). */
+	events: number;
+}
+
 /** One provenance point — message-granular (DECIDED 2026-09-02): nodes and
  *  edges record every (chat, message) where they came up, deep-linkable to
  *  the exact transcript point. */

src/core/contracts/jobs.ts

@@ -7,6 +7,7 @@
  * of "ai" (safe everywhere: server, client, tests). See docs/vision.md.
  */
 import type { UIMessage } from "ai";
+import type { CallUsage } from "./chat.ts";
 
 // ─── Async sub-agent jobs (vision §2.5; plan.md Wave 0 protocol) ────────────
 
@@ -36,18 +37,64 @@ export interface JobRecord {
 
 /** A sub-agent run's accumulated usage — the same accounting shape as a
  * chat turn (contracts/chat.ts CallUsage), flattened to one running total
- * plus a step count. */
+ * plus a step count. Token fields are optional exactly like CallUsage: a
+ * field NO model call reported stays undefined — absent is not zero (the
+ * money-channel doctrine; cost renders "—", never $0). */
 export interface JobUsage {
 	/** The OpenRouter model id that ran the sub-agent. */
 	model: string;
 	/** Running totals summed over the run's model calls so far. */
-	inputTokens: number;
-	outputTokens: number;
-	cacheReadTokens: number;
-	cacheWriteTokens: number;
-	reasoningTokens: number;
+	inputTokens: number | undefined;
+	outputTokens: number | undefined;
+	cacheReadTokens: number | undefined;
+	cacheWriteTokens: number | undefined;
+	reasoningTokens: number | undefined;
 	/** How many model calls the run has made (each bills separately). */
 	steps: number;
+	/** The MOST RECENT model call's accounting (2026-09 dock contract): the
+	 * per-blob context line reads last-step input+output against the model's
+	 * CONTEXT_LIMITS — never the cumulative totals. Absent until the first
+	 * call completes; absent provider fields stay absent (never zero). */
+	lastStep?: CallUsage;
+}
+
+/** The per-model-call usage shape we fold into JobUsage — the SDK's
+ * LanguageModelUsage narrowed structurally (no runtime "ai" import: any
+ * object with these fields folds, including the real thing and test fakes). */
+export interface UsageStepLike {
+	inputTokens?: number | undefined;
+	inputTokenDetails?: { cacheReadTokens?: number; cacheWriteTokens?: number } | undefined;
+	outputTokens?: number | undefined;
+	outputTokenDetails?: { reasoningTokens?: number } | undefined;
+}
+
+/** Fold one model call's usage into the run's running total (pure). A
+ * field neither the total nor the step reported stays undefined; once
+ * reported, later unreported calls contribute 0. The `steps` counter is
+ * the billing count: one per model call, reported or not. */
+export function accumulateJobUsage(
+	total: JobUsage | undefined,
+	model: string,
+	step: UsageStepLike,
+): JobUsage {
+	const add = (a: number | undefined, b: number | undefined): number | undefined =>
+		a === undefined && b === undefined ? undefined : (a ?? 0) + (b ?? 0);
+	return {
+		model,
+		inputTokens: add(total?.inputTokens, step.inputTokens),
+		outputTokens: add(total?.outputTokens, step.outputTokens),
+		cacheReadTokens: add(total?.cacheReadTokens, step.inputTokenDetails?.cacheReadTokens),
+		cacheWriteTokens: add(total?.cacheWriteTokens, step.inputTokenDetails?.cacheWriteTokens),
+		reasoningTokens: add(total?.reasoningTokens, step.outputTokenDetails?.reasoningTokens),
+		steps: (total?.steps ?? 0) + 1,
+		lastStep: {
+			inputTokens: step.inputTokens,
+			outputTokens: step.outputTokens,
+			cacheReadTokens: step.inputTokenDetails?.cacheReadTokens,
+			cacheWriteTokens: step.inputTokenDetails?.cacheWriteTokens,
+			reasoningTokens: step.outputTokenDetails?.reasoningTokens,
+		},
+	};
 }
 
 /**
@@ -155,6 +202,11 @@ export interface SubagentResultMeta {
 	jobId: string;
 	kind: JobKind;
 	toolCallId: string;
+	/** Present only on an abort steer (2026-09 abort semantics): the
+	 * learner aborted the job (it lands as error "Aborted by learner");
+	 * the teacher is told so its promise of a result doesn't hang
+	 * silently in the lesson. */
+	aborted?: true;
 }
 
 /** The data-subagent-artifact part payload: what the UI renders. */
@@ -178,6 +230,20 @@ export interface SubagentSteerMessage extends UIMessage {
 	];
 }
 
+/** The abort steer: user-role like any steer, but NO artifact part —
+ * there is no result to render, only the teacher-facing text. Shares the
+ * result steer's deterministic id space (`steer_${jobId}`): steerDelivered
+ * matches on metadata jobId, so an abort and a (racing) result steer
+ * can never BOTH deliver — exactly one wakes the teacher. */
+export interface AbortSteerMessage extends UIMessage {
+	role: "user";
+	metadata: { subagentResult: SubagentResultMeta & { aborted: true } };
+	parts: [{ type: "text"; text: string }];
+}
+
+/** Anything the wake pipeline may queue and deliver. */
+export type SteerMessage = SubagentSteerMessage | AbortSteerMessage;
+
 /** Whether a result is the failure shape (html/svg empty, failure brief). */
 export function isSubagentFailure(kind: JobKind, result: ResearchResult | DemoResult | VisualResult): boolean {
 	if (kind === "research") {
@@ -234,6 +300,32 @@ export function buildSubagentSteer(
 	};
 }
 
+/** Build the abort steer for an aborted job (pure — shared by the
+ * abort path in src/server/jobs.ts and tests). Same id space as the result
+ * steer; see AbortSteerMessage. */
+export function buildAbortSteer(
+	job: Pick<JobRecord, "id" | "kind" | "toolCallId">,
+): AbortSteerMessage {
+	return {
+		id: `steer_${job.id}`,
+		role: "user",
+		metadata: {
+			subagentResult: { jobId: job.id, kind: job.kind, toolCallId: job.toolCallId, aborted: true },
+		},
+		parts: [{ type: "text", text: abortText(job.kind) }],
+	};
+}
+
+/** What the teacher reads: one line telling it the result is not coming,
+ * how to acknowledge, and that re-dispatching is allowed. */
+function abortText(kind: JobKind): string {
+	if (kind === "research")
+		return "Research aborted — the learner aborted the research job, so no brief is coming. Acknowledge it in one line and continue teaching without it (or offer to re-dispatch it).";
+	if (kind === "demo")
+		return "The demo job was aborted by the learner — no demo is coming. Acknowledge it in one line and teach the topic in prose instead (a re-dispatch is fine if the learner wants it).";
+	return "The figure job was aborted by the learner — no figure is coming. Acknowledge it in one line and teach without it (a re-dispatch is fine if the learner wants it).";
+}
+
 /** True when the transcript already carries this job's steer (idempotent delivery). */
 export function steerDelivered(messages: UIMessage[], jobId: string): boolean {
 	return messages.some(

src/core/graph.ts

@@ -42,6 +42,7 @@ import { randomBytes } from "node:crypto";
 import { isToolUIPart, type UIMessage } from "ai";
 import type {
 	AuditResult,
+	EmissionStats,
 	GraphEdge,
 	GraphNode,
 	ProvenancePoint,
@@ -108,15 +109,6 @@ const samePoint = (a: ProvenancePoint, b: ProvenancePoint): boolean =>
 
 // ─── the merger (extraction pipeline stage 3 — deterministic) ────────────────
 
-export interface EmissionStats {
-	/** New graph nodes appended. */
-	nodes: number;
-	/** New edge assertions appended. */
-	edges: number;
-	/** New ledger events appended. */
-	events: number;
-}
-
 /**
  * Merge one chat's audited findings into the graph ledgers — the pipeline's
  * apply stage. Deterministic + idempotent: re-merging the same verdicts (or
@@ -470,3 +462,6 @@ export function graphSnapshot(deps: GraphDeps): GraphSnapshot {
 
 	return { nodes, edges, sessions, nodeSessions, nodeObservations };
 }
+
+// Re-export: the contract owns the shape; legacy importers still reach it.
+export type { EmissionStats };

src/core/ports.ts

@@ -18,6 +18,7 @@ import type {
 	JobKind,
 	JobRecord,
 	LedgerEvent,
+	ExtractionRunUsage,
 	NodeMapping,
 	PlanTree,
 	SessionCheckpoint,
@@ -41,6 +42,8 @@ export interface ChatStore {
 	 * 2026-09): idempotent per messageId — a turn persisted twice (abort
 	 * replays, wake-turn retries) appends exactly once. */
 	appendTurnUsage(id: string, entry: TurnUsage): void;
+	/** Append one extraction-run usage entry — idempotent per startedAt. */
+	appendExtractionUsage(id: string, entry: ExtractionRunUsage): void;
 }
 
 // ─── GraphStore — the append-only evidence ledgers (data/graph) ──────────────

src/server/jobs.ts

@@ -33,13 +33,15 @@
 
 import type { UIMessage } from "ai";
 import {
+	buildAbortSteer,
 	buildSubagentSteer,
 	isSubagentFailure,
 	steerDelivered,
 	type JobRecord,
-	type SubagentSteerMessage,
+	type SteerMessage,
 } from "../core/contracts.ts";
 import { findPendingToolCall } from "../core/pending.ts";
+import { fireJobAbort } from "../agent/subagents/aborts.ts";
 import { stores } from "./stores.ts";
 import { isTurnRunning } from "./turn-liveness.ts";
 
@@ -50,7 +52,7 @@ export type TurnRunner = (chatId: string, messages: UIMessage[]) => Promise<numb
 export type JobRespawn = (jobId: string) => void;
 
 interface QueuedSteer {
-	steer: SubagentSteerMessage;
+	steer: SteerMessage;
 	resolve: () => void;
 	reject: (reason: Error) => void;
 }
@@ -108,7 +110,7 @@ function delay(ms: number): Promise<void> {
 	return new Promise((resolve) => setTimeout(resolve, ms));
 }
 
-function jobIdOf(steer: SubagentSteerMessage): string {
+function jobIdOf(steer: SteerMessage): string {
 	return (steer.metadata as { subagentResult: { jobId: string } }).subagentResult.jobId;
 }
 
@@ -117,7 +119,7 @@ function jobIdOf(steer: SubagentSteerMessage): string {
  * has completed (and the job status flipped); rejects when delivery fails
  * unrecoverably. Never rejects for queueing itself.
  */
-export function deliverJobResult(steer: SubagentSteerMessage): Promise<void> {
+export function deliverJobResult(steer: SteerMessage): Promise<void> {
 	return new Promise<void>((resolve, reject) => {
 		const s = state();
 		const chatId = chatIdOf(steer);
@@ -145,13 +147,13 @@ export function deliverJobResult(steer: SubagentSteerMessage): Promise<void> {
  * before firing, so a claim racing the pump converges to exactly one
  * delivery either way.
  */
-export function claimSteers(chatId: string): SubagentSteerMessage[] {
+export function claimSteers(chatId: string): SteerMessage[] {
 	const s = state();
 	const queue = s.queues.get(chatId);
 	if (!queue || queue.length === 0) return [];
 	const record = stores.chatStore.getChat(chatId);
 	if (!record) return []; // the handler validated the chat; defensive only
-	const claimed: SubagentSteerMessage[] = [];
+	const claimed: SteerMessage[] = [];
 	for (const entry of queue) {
 		const jobId = jobIdOf(entry.steer);
 		if (steerDelivered(record.messages, jobId)) {
@@ -167,7 +169,66 @@ export function claimSteers(chatId: string): SubagentSteerMessage[] {
 	return claimed;
 }
 
-function chatIdOf(steer: SubagentSteerMessage): string {
+/** Result of an abort attempt — the endpoint maps these to HTTP. */
+export interface AbortOutcome {
+	ok: boolean;
+	reason?: "not_found" | "not_running";
+}
+
+/**
+ * ABORT (2026-09 dock semantics, user-decided): real kill, error with
+ * reason, the teacher still learns. The dock's END button reaches this
+ * through POST /api/activity/abort (body {kind: "job", id: jobId}); the
+ * route is master's thin composition — this is the job-side mechanism.
+ * Four things happen, in order:
+ *
+ *   1. KILL the spend: fire the job's registered AbortController (the
+ *      runner registered it at loop start) — the in-flight model call
+ *      aborts, not just the next loop check. The runner unwinds silently
+ *      (its catch checks abortSignal.aborted; finishJob checks status).
+ *   2. DROP any queued result steer for this job (the delivering-result
+ *      phase): resolve its entry without firing the wake. If the wake
+ *      turn is ALREADY in flight (the pump shifted the entry), it's too
+ *      late — the result steer lands, and the cancellation steer dedupes
+ *      against it via steerDelivered (same jobId metadata).
+ *   3. FLIP the record to error "Aborted by learner" immediately — the
+ *      poll reflects the user's action at once (the user chose
+ *      error-with-reason over a distinct cancelled status). finalizeJob
+ *      never overwrites a non-running status, so the flip is terminal no
+ *      matter what the wake does.
+ *   4. STEER the teacher: a one-line abort note (buildAbortSteer) through
+ *      the same wake pipeline, so the teacher's promise of a result
+ *      doesn't hang silently in the lesson.
+ *
+ * The runner writes the usage folded so far in its finally (merge patch),
+ * so an aborted job's money record survives the kill.
+ */
+export function abortJob(jobId: string): AbortOutcome {
+	const job = stores.jobStore.getJob(jobId);
+	if (!job) return { ok: false, reason: "not_found" };
+	if (job.status !== "running") return { ok: false, reason: "not_running" };
+
+	fireJobAbort(jobId); // no-op when the loop already returned (delivering)
+
+	// Drop any queued steers for this job (their deliver() resolves unfired).
+	const queue = state().queues.get(job.chatId);
+	if (queue) {
+		const remaining = queue.filter((e) => jobIdOf(e.steer) !== jobId);
+		if (remaining.length === 0) state().queues.delete(job.chatId);
+		else state().queues.set(job.chatId, remaining);
+		for (const entry of queue) {
+			if (jobIdOf(entry.steer) === jobId) entry.resolve();
+		}
+	}
+
+	stores.jobStore.updateJob(jobId, { status: "error", error: "Aborted by learner", activity: "aborted" });
+	void deliverJobResult(buildAbortSteer(job)).catch(() => {
+			/* deliverJobResult marks the job error on unrecoverable failure */
+	});
+	return { ok: true };
+}
+
+function chatIdOf(steer: SteerMessage): string {
 	// The chat is resolvable from the job record; the steer's metadata
 	// carries jobId, which indexes the store.
 	const jobId = jobIdOf(steer);
@@ -285,6 +346,11 @@ function finalizeJob(jobId: string): void {
 }
 
 function failJob(jobId: string, err: Error): void {
+	// An aborted job stays aborted even when its (abort) steer
+	// delivery fails — the user's action is the terminal fact, not the
+	// delivery error. Non-running records are never overwritten here.
+	const job = stores.jobStore.getJob(jobId);
+	if (!job || job.status !== "running") return;
 	stores.jobStore.updateJob(jobId, { status: "error", error: err.message, activity: "delivery failed" });
 }
 

src/server/observability.ts

@@ -13,6 +13,10 @@
  * (chat-transport); the dock is a different consumer with a different
  * shape, served here. Single sources of truth are unchanged: the job
  * store and the pipeline registry — this file only composes them.
+ *
+ * The extraction side of the dock's controls also surfaces here:
+ * abortExtraction (true mid-run cancellation, user-decided) is exported
+ * from the pipeline's module for the abort route to call.
  */
 
 import { extractionStatuses, type ExtractionPhase } from "../agent/subagents/extraction.ts";
@@ -20,8 +24,8 @@ import { chatStore } from "../store/chats.ts";
 import { listJobs } from "./jobs.ts";
 import type { JobRecord } from "../core/contracts.ts";
 
-export { extractionStatuses, type ExtractionPhase };
-export { chatStore };
+export { abortExtraction, extractionStatuses, type ExtractionPhase, type RunningExtraction } from "../agent/subagents/extraction.ts";
+export { chatStore } from "../store/chats.ts";
 
 /** A running sub-agent job, joined with its chat's title for the dock. */
 export interface ActivityJob extends JobRecord {

src/store/chats.ts

@@ -11,7 +11,7 @@ import * as fs from "node:fs";
 import * as path from "node:path";
 import { PATHS } from "../../config.ts";
 import type { ChatStore } from "../core/ports.ts";
-import type { ChatRecord, ChatSummary, PlanTree, TurnUsage } from "../core/contracts.ts";
+import type { ChatRecord, ChatSummary, ExtractionRunUsage, PlanTree, TurnUsage } from "../core/contracts.ts";
 import type { UIMessage } from "ai";
 import { findPendingToolCall } from "../core/pending.ts";
 import { phaseOfTree } from "../core/phases.ts";
@@ -51,9 +51,10 @@ function listChats(): ChatSummary[] {
 			.filter((f) => f.endsWith(".json"))
 			.map((f) => getChat(f.slice(0, -5)))
 			.filter((r): r is ChatRecord => r !== null)
-			.map(({ messages, tree, turns, ...summary }) => ({
-				// turns stay out of the list payload (list = summaries; usage
-				// aggregates ride the full record, the picker never needs them).
+			.map(({ messages, tree, turns, extractions, ...summary }) => ({
+				// turns and extractions stay out of the list payload (list =
+				// summaries; usage aggregates ride the full record, the picker
+				// never needs them).
 				...summary,
 				messageCount: messages.length,
 				// Record-derived status for the session picker: phase from the
@@ -90,6 +91,18 @@ function appendTurnUsage(id: string, entry: TurnUsage): void {
 	fs.writeFileSync(chatFile(id), JSON.stringify(record, null, "\t"), "utf-8");
 }
 
+/** Append one extraction-run usage entry to the record's extractions
+ * log — idempotent per startedAt: a retry that reaches the append twice
+ * (abort replays) does not double-count spend. */
+function appendExtractionUsage(id: string, entry: ExtractionRunUsage): void {
+	const record = getChat(id);
+	if (!record) throw new Error(`Chat not found: ${id}`);
+	record.extractions = record.extractions?.some((r) => r.startedAt === entry.startedAt)
+		? record.extractions
+		: [...(record.extractions ?? []), entry];
+	fs.writeFileSync(chatFile(id), JSON.stringify(record, null, "\t"), "utf-8");
+}
+
 /** Persist the session plan tree (plan tools call this in execute). */
 function saveTree(id: string, tree: PlanTree): void {
 	const record = getChat(id);
@@ -107,4 +120,5 @@ export const chatStore: ChatStore = {
 	saveMessages,
 	saveTree,
 	appendTurnUsage,
+	appendExtractionUsage,
 };