fix: align compaction cuts and native usage accounting

src/core/contracts/chat.ts

@@ -87,8 +87,8 @@ export interface ChatSettings {
  *  billed (CallUsage per agent step). Steps, not a sum: cost and context
  *  folds happen at read time (core/prices.ts). */
 export interface ExtractionStageUsage {
-	/** The model that ran this stage, provider-qualified ("provider:id" —
-	 *  model slots wave, 2026-09-20; matches PRICES' uniform keys). */
+	/** The model that ran this stage, provider-qualified ("provider:id")
+	 * for key-independent Pi catalog lookup at read time. */
 	model: string;
 	steps: CallUsage[];
 }
@@ -118,7 +118,7 @@ export interface ExtractionRunUsage {
  * flattened to the fields we persist (undefined = provider didn't report;
  * renderers show absent, never zero). */
 export interface CallUsage {
-	/** Total input (prompt) tokens, cached portion included. */
+	/** Pi-normalized fresh input tokens; cache buckets are disjoint. */
 	inputTokens: number | undefined;
 	outputTokens: number | undefined;
 	/** Cached input tokens READ (billed at the cache-read rate, if any). */
@@ -126,6 +126,9 @@ export interface CallUsage {
 	/** Input tokens written to the cache this call (OpenRouter bills writes
 	 * for some providers; undefined = not reported/not billed). */
 	cacheWriteTokens: number | undefined;
+	/** Subset of cacheWriteTokens written with one-hour retention. Missing
+	 * on older logs; new native captures explicitly record zero if absent. */
+	cacheWrite1hTokens?: number;
 	/** The reasoning portion of the output tokens. */
 	reasoningTokens: number | undefined;
 }
@@ -144,8 +147,8 @@ export interface TurnUsage {
 	 *  entry to the transcript (the block this turn "belongs" to). */
 	messageId: string;
 	at: string;
-	/** The model that ran the turn, provider-qualified ("provider:id" —
-	 *  model slots wave, 2026-09-20; matches PRICES' uniform keys). */
+	/** The model that ran the turn, provider-qualified ("provider:id")
+	 * for key-independent Pi catalog lookup at read time. */
 	model: string;
 	/** One entry per agent step (model call) in this turn. */
 	steps: CallUsage[];

src/core/contracts/jobs.ts

@@ -43,7 +43,7 @@ export interface JobRecord {
  * money-channel doctrine; cost renders "—", never $0). */
 export interface JobUsage {
 	/** The model that ran the sub-agent, provider-qualified ("provider:id" —
-	 *  model slots wave, 2026-09-20; matches PRICES' uniform keys). */
+	 *  model slots wave, 2026-09-20; matches Pi catalog lookup keys). */
 	model: string;
 	/** Running totals summed over the run's model calls so far. */
 	inputTokens: number | undefined;
@@ -51,6 +51,10 @@ export interface JobUsage {
 	cacheReadTokens: number | undefined;
 	cacheWriteTokens: number | undefined;
 	reasoningTokens: number | undefined;
+	cacheWrite1hTokens?: number;
+	/** Individual calls for current-catalog per-call tier pricing. Absent on
+	 * older jobs; never backfill them from cumulative totals. */
+	calls?: CallUsage[];
 	/** 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
@@ -67,7 +71,7 @@ export interface JobUsage {
  * object with these fields folds, including the real thing and test fakes). */
 export interface UsageStepLike {
 	inputTokens?: number | undefined;
-	inputTokenDetails?: { cacheReadTokens?: number; cacheWriteTokens?: number } | undefined;
+	inputTokenDetails?: { cacheReadTokens?: number; cacheWriteTokens?: number; cacheWrite1hTokens?: number } | undefined;
 	outputTokens?: number | undefined;
 	outputTokenDetails?: { reasoningTokens?: number } | undefined;
 }
@@ -80,12 +84,13 @@ export function piUsageToStep(usage: {
 	output: number;
 	cacheRead: number;
 	cacheWrite: number;
+	cacheWrite1h?: number;
 	reasoning?: number;
 }): UsageStepLike {
 	return {
 		inputTokens: usage.input,
 		outputTokens: usage.output,
-		inputTokenDetails: { cacheReadTokens: usage.cacheRead, cacheWriteTokens: usage.cacheWrite },
+		inputTokenDetails: { cacheReadTokens: usage.cacheRead, cacheWriteTokens: usage.cacheWrite, cacheWrite1hTokens: usage.cacheWrite1h ?? 0 },
 		outputTokenDetails: usage.reasoning !== undefined ? { reasoningTokens: usage.reasoning } : undefined,
 	};
 }
@@ -101,6 +106,14 @@ export function accumulateJobUsage(
 ): JobUsage {
 	const add = (a: number | undefined, b: number | undefined): number | undefined =>
 		a === undefined && b === undefined ? undefined : (a ?? 0) + (b ?? 0);
+	const call: CallUsage = {
+		inputTokens: step.inputTokens,
+		outputTokens: step.outputTokens,
+		cacheReadTokens: step.inputTokenDetails?.cacheReadTokens,
+		cacheWriteTokens: step.inputTokenDetails?.cacheWriteTokens,
+		cacheWrite1hTokens: step.inputTokenDetails?.cacheWrite1hTokens,
+		reasoningTokens: step.outputTokenDetails?.reasoningTokens,
+	};
 	return {
 		model,
 		inputTokens: add(total?.inputTokens, step.inputTokens),
@@ -108,14 +121,12 @@ export function accumulateJobUsage(
 		cacheReadTokens: add(total?.cacheReadTokens, step.inputTokenDetails?.cacheReadTokens),
 		cacheWriteTokens: add(total?.cacheWriteTokens, step.inputTokenDetails?.cacheWriteTokens),
 		reasoningTokens: add(total?.reasoningTokens, step.outputTokenDetails?.reasoningTokens),
+		cacheWrite1hTokens: ((total?.cacheWriteTokens ?? 0) > 0 && total?.cacheWrite1hTokens === undefined) ||
+			((call.cacheWriteTokens ?? 0) > 0 && call.cacheWrite1hTokens === undefined)
+			? undefined : add(total?.cacheWrite1hTokens, call.cacheWrite1hTokens),
 		steps: (total?.steps ?? 0) + 1,
-		lastStep: {
-			inputTokens: step.inputTokens,
-			outputTokens: step.outputTokens,
-			cacheReadTokens: step.inputTokenDetails?.cacheReadTokens,
-			cacheWriteTokens: step.inputTokenDetails?.cacheWriteTokens,
-			reasoningTokens: step.outputTokenDetails?.reasoningTokens,
-		},
+		...(!total || total.calls ? { calls: [...(total?.calls ?? []), call] } : {}),
+		lastStep: call,
 	};
 }
 

src/core/contracts/models.ts

@@ -10,7 +10,7 @@ import type { BackgroundModelRole, ModelSlotName } from "../settings.ts";
 
 /** One picker row: everything the UI shows about a model, projected from
  *  pi-ai's catalog. Costs are USD per 1M tokens (the same catalog, the same
- *  units core/prices.ts's generated table carries). */
+ *  units as Pi's server-side pricing calculator). */
 export interface ModelCatalogEntry {
 	id: string;
 	name: string;