@oh-my-pi/omp-stats 15.0.0 → 15.0.2

package/src/client/types.ts

@@ -1,8 +1,28 @@
 /**
  * Client-side type definitions.
- * Duplicated from ../types.ts to avoid pulling in server dependencies.
+ *
+ * Shared shapes (aggregations, time-series, dashboard payloads) live in
+ * `../shared-types` and are re-exported here. The types declared inline below
+ * are deliberately client-only because:
+ *   - `Usage` is redeclared locally so the client bundle avoids importing
+ *     `@oh-my-pi/pi-ai` (the server-side AI types package).
+ *   - `MessageStats.stopReason` is widened from the server's `StopReason`
+ *     enum to `string`, again to keep the client free of pi-ai types.
+ *   - `TimeRange`, `OverviewStats`, `ModelDashboardStats`,
+ *     `CostDashboardStats` are UI-only view shapes the server never produces.
  */
 
+import type {
+	AggregatedStats,
+	CostTimeSeriesPoint,
+	ModelPerformancePoint,
+	ModelStats,
+	ModelTimeSeriesPoint,
+	TimeSeriesPoint,
+} from "../shared-types";
+
+export * from "../shared-types";
+
 export interface Usage {
 	input: number;
 	output: number;
@@ -40,80 +60,7 @@ export interface RequestDetails extends MessageStats {
 	output: unknown;
 }
 
-export interface AggregatedStats {
-	totalRequests: number;
-	successfulRequests: number;
-	failedRequests: number;
-	errorRate: number;
-	totalInputTokens: number;
-	totalOutputTokens: number;
-	totalCacheReadTokens: number;
-	totalCacheWriteTokens: number;
-	cacheRate: number;
-	totalCost: number;
-	totalPremiumRequests: number;
-	avgDuration: number | null;
-	avgTtft: number | null;
-	avgTokensPerSecond: number | null;
-	firstTimestamp: number;
-	lastTimestamp: number;
-}
-
 export type TimeRange = "1h" | "24h" | "7d" | "30d" | "90d" | "all";
-export interface ModelStats extends AggregatedStats {
-	model: string;
-	provider: string;
-}
-
-export interface FolderStats extends AggregatedStats {
-	folder: string;
-}
-
-export interface TimeSeriesPoint {
-	timestamp: number;
-	requests: number;
-	errors: number;
-	tokens: number;
-	cost: number;
-}
-
-export interface ModelTimeSeriesPoint {
-	timestamp: number;
-	model: string;
-	provider: string;
-	requests: number;
-}
-
-export interface ModelPerformancePoint {
-	timestamp: number;
-	model: string;
-	provider: string;
-	requests: number;
-	avgTtft: number | null;
-	avgTokensPerSecond: number | null;
-}
-
-export interface CostTimeSeriesPoint {
-	timestamp: number;
-	model: string;
-	provider: string;
-	cost: number;
-	costInput: number;
-	costOutput: number;
-	costCacheRead: number;
-	costCacheWrite: number;
-	requests: number;
-}
-
-export interface DashboardStats {
-	overall: AggregatedStats;
-	byModel: ModelStats[];
-	byFolder: FolderStats[];
-	timeSeries: TimeSeriesPoint[];
-	modelSeries: ModelTimeSeriesPoint[];
-	modelPerformanceSeries: ModelPerformancePoint[];
-	costSeries: CostTimeSeriesPoint[];
-}
 
 export interface OverviewStats {
 	overall: AggregatedStats;
@@ -129,50 +76,3 @@ export interface ModelDashboardStats {
 export interface CostDashboardStats {
 	costSeries: CostTimeSeriesPoint[];
 }
-
-export interface BehaviorTimeSeriesPoint {
-	timestamp: number;
-	model: string;
-	provider: string;
-	messages: number;
-	yelling: number;
-	profanity: number;
-	anguish: number;
-	negation: number;
-	repetition: number;
-	blame: number;
-	chars: number;
-}
-
-export interface BehaviorOverallStats {
-	totalMessages: number;
-	totalYelling: number;
-	totalProfanity: number;
-	totalAnguish: number;
-	totalNegation: number;
-	totalRepetition: number;
-	totalBlame: number;
-	totalChars: number;
-	firstTimestamp: number;
-	lastTimestamp: number;
-}
-
-export interface BehaviorModelStats {
-	model: string;
-	provider: string;
-	totalMessages: number;
-	totalYelling: number;
-	totalProfanity: number;
-	totalAnguish: number;
-	totalNegation: number;
-	totalRepetition: number;
-	totalBlame: number;
-	totalChars: number;
-	lastTimestamp: number;
-}
-
-export interface BehaviorDashboardStats {
-	overall: BehaviorOverallStats;
-	byModel: BehaviorModelStats[];
-	behaviorSeries: BehaviorTimeSeriesPoint[];
-}

package/src/db.ts

@@ -38,6 +38,7 @@ const BACKFILL_COMPLETE = "complete";
 const BACKFILL_PENDING = "pending";
 const USER_MESSAGES_BACKFILL_KEY = "user_messages_v5";
 const USER_MESSAGE_LINKS_REPAIR_KEY = "user_message_links_v1";
+const PRIORITY_PREMIUM_REQUESTS_BACKFILL_KEY = "premium_requests_priority_v1";
 function shouldResetBackfill(value: string | undefined): boolean {
 	return value !== BACKFILL_COMPLETE && value !== BACKFILL_PENDING;
 }
@@ -179,6 +180,7 @@ export async function initDb(): Promise<Database> {
 	}
 	backfillUserMessages(db);
 	repairUserMessageLinks(db);
+	backfillPriorityPremiumRequests(db);
 	backfillMissingCatalogCosts(db);
 	return db;
 }
@@ -300,13 +302,21 @@ export function setFileOffset(sessionFile: string, offset: number, lastModified:
 export function insertMessageStats(stats: MessageStats[]): number {
 	if (!db || stats.length === 0) return 0;
 
+	// Use UPSERT so a re-sync can fix up `premium_requests` for rows persisted
+	// before priority service-tier traffic was counted as premium. The guard
+	// `WHERE messages.premium_requests < excluded.premium_requests` keeps every
+	// other column immutable and never demotes an existing count (e.g. when a
+	// later parse drops back to 0 for the same row).
 	const stmt = db.prepare(`
-		INSERT OR IGNORE INTO messages (
+		INSERT INTO messages (
 			session_file, entry_id, folder, model, provider, api, timestamp,
 			duration, ttft, stop_reason, error_message,
 			input_tokens, output_tokens, cache_read_tokens, cache_write_tokens, total_tokens, premium_requests,
 			cost_input, cost_output, cost_cache_read, cost_cache_write, cost_total
 		) VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)
+		ON CONFLICT(session_file, entry_id) DO UPDATE SET
+			premium_requests = excluded.premium_requests
+		WHERE messages.premium_requests < excluded.premium_requests
 	`);
 
 	let inserted = 0;
@@ -780,6 +790,36 @@ function repairUserMessageLinks(database: Database): void {
 		.run(USER_MESSAGE_LINKS_REPAIR_KEY, BACKFILL_PENDING);
 }
 
+/**
+ * One-shot wipe of `file_offsets` so the next sync re-parses every session
+ * and re-derives `premium_requests` from recorded `service_tier_change`
+ * entries. Earlier ingestions captured priority OpenAI traffic with
+ * `premium_requests = 0` because the AI layer only set the field for GitHub
+ * Copilot traffic. The parser now folds priority requests into the same
+ * counter; combined with the UPSERT in `insertMessageStats`, a single sync
+ * pass brings the messages table up to date without touching any other
+ * column. Idempotent: gated by a sentinel row in `meta`.
+ */
+function backfillPriorityPremiumRequests(database: Database): void {
+	const row = database.prepare("SELECT value FROM meta WHERE key = ?").get(PRIORITY_PREMIUM_REQUESTS_BACKFILL_KEY) as
+		| { value: string }
+		| undefined;
+	if (!shouldResetBackfill(row?.value)) return;
+
+	database.exec("DELETE FROM file_offsets");
+	database
+		.prepare("INSERT OR REPLACE INTO meta (key, value) VALUES (?, ?)")
+		.run(PRIORITY_PREMIUM_REQUESTS_BACKFILL_KEY, BACKFILL_PENDING);
+}
+
+export function markPriorityPremiumRequestsBackfillComplete(): void {
+	if (!db) return;
+	db.prepare("INSERT OR REPLACE INTO meta (key, value) VALUES (?, ?)").run(
+		PRIORITY_PREMIUM_REQUESTS_BACKFILL_KEY,
+		BACKFILL_COMPLETE,
+	);
+}
+
 export function markUserMessagesBackfillComplete(): void {
 	if (!db) return;
 	db.prepare("INSERT OR REPLACE INTO meta (key, value) VALUES (?, ?)").run(

package/src/parser.ts

@@ -1,8 +1,15 @@
 import * as fs from "node:fs/promises";
 import * as path from "node:path";
-import type { AssistantMessage } from "@oh-my-pi/pi-ai";
+import { type AssistantMessage, getPriorityPremiumRequests, type ServiceTier } from "@oh-my-pi/pi-ai";
 import { getSessionsDir, isEnoent } from "@oh-my-pi/pi-utils";
-import type { MessageStats, SessionEntry, SessionMessageEntry, UserMessageLink, UserMessageStats } from "./types";
+import type {
+	MessageStats,
+	SessionEntry,
+	SessionMessageEntry,
+	SessionServiceTierChangeEntry,
+	UserMessageLink,
+	UserMessageStats,
+} from "./types";
 import { computeUserMessageMetrics } from "./user-metrics";
 
 /**
@@ -24,6 +31,10 @@ function extractFolderFromPath(sessionPath: string): string {
 function isAssistantMessage(entry: SessionEntry): entry is SessionMessageEntry {
 	if (entry.type !== "message") return false;
 	const msgEntry = entry as SessionMessageEntry;
+	// Legacy sessions (pre-id tracking) recorded message entries without an `id`.
+	// They're not linkable and would violate the messages.entry_id NOT NULL
+	// constraint, so skip them at the parser boundary.
+	if (typeof msgEntry.id !== "string" || msgEntry.id.length === 0) return false;
 	return msgEntry.message?.role === "assistant";
 }
 
@@ -33,9 +44,17 @@ function isAssistantMessage(entry: SessionEntry): entry is SessionMessageEntry {
 function isUserMessage(entry: SessionEntry): entry is SessionMessageEntry {
 	if (entry.type !== "message") return false;
 	const msgEntry = entry as SessionMessageEntry;
+	if (typeof msgEntry.id !== "string" || msgEntry.id.length === 0) return false;
 	return msgEntry.message?.role === "user";
 }
 
+/**
+ * Check if an entry is a service-tier change.
+ */
+function isServiceTierChange(entry: SessionEntry): entry is SessionServiceTierChangeEntry {
+	return entry.type === "service_tier_change";
+}
+
 /**
  * Extract plain text from a user message content payload.
  */
@@ -83,10 +102,25 @@ function extractUserStats(sessionFile: string, folder: string, entry: SessionMes
 /**
  * Extract stats from an assistant message entry.
  */
-function extractStats(sessionFile: string, folder: string, entry: SessionMessageEntry): MessageStats | null {
+function extractStats(
+	sessionFile: string,
+	folder: string,
+	entry: SessionMessageEntry,
+	currentServiceTier: ServiceTier | undefined,
+): MessageStats | null {
 	const msg = entry.message as AssistantMessage;
 	if (!msg || msg.role !== "assistant") return null;
 
+	// Backfill: when the session recorded `priority` as the active service tier
+	// at this point but the AI usage payload was captured before priority
+	// requests were folded into `premiumRequests`, derive the count here so the
+	// "Premium Reqs" stat aggregates priority traffic on re-sync. Trust any
+	// non-zero value already in `usage.premiumRequests` (Copilot multipliers or
+	// the new AI code path) and only synthesise when the field is missing/zero.
+	const recorded = msg.usage.premiumRequests ?? 0;
+	const derived = recorded > 0 ? recorded : getPriorityPremiumRequests(currentServiceTier, msg.provider);
+	const usage = derived === recorded ? msg.usage : { ...msg.usage, premiumRequests: derived };
+
 	return {
 		sessionFile,
 		entryId: entry.id,
@@ -99,7 +133,7 @@ function extractStats(sessionFile: string, folder: string, entry: SessionMessage
 		ttft: msg.ttft ?? null,
 		stopReason: msg.stopReason,
 		errorMessage: msg.errorMessage ?? null,
-		usage: msg.usage,
+		usage,
 	};
 }
 
@@ -158,7 +192,12 @@ export async function parseSessionFile(sessionPath: string, fromOffset = 0): Pro
 	const start = Math.max(0, Math.min(fromOffset, bytes.length));
 	const unprocessed = bytes.subarray(start);
 	const { entries, read } = parseSessionEntriesLenient(unprocessed);
+	let currentServiceTier: ServiceTier | undefined;
 	for (const entry of entries) {
+		if (isServiceTierChange(entry)) {
+			currentServiceTier = entry.serviceTier ?? undefined;
+			continue;
+		}
 		if (isUserMessage(entry)) {
 			const userMsg = extractUserStats(sessionPath, folder, entry);
 			if (userMsg) {
@@ -168,7 +207,7 @@ export async function parseSessionFile(sessionPath: string, fromOffset = 0): Pro
 			continue;
 		}
 		if (isAssistantMessage(entry)) {
-			const msgStats = extractStats(sessionPath, folder, entry);
+			const msgStats = extractStats(sessionPath, folder, entry, currentServiceTier);
 			if (msgStats) stats.push(msgStats);
 			// Link assistant's responding model back to the user message it answered.
 			const parentId = (entry as SessionMessageEntry).parentId;

package/src/shared-types.ts

@@ -0,0 +1,204 @@
+/**
+ * Shared type definitions consumed by both the server-side stats code and the
+ * standalone client bundle. Keep this file free of any imports from server-only
+ * packages (e.g. `@oh-my-pi/pi-ai`, `bun:sqlite`) so the client can import it
+ * without dragging server dependencies into its bundle.
+ */
+
+/**
+ * Aggregated stats for a model or folder.
+ */
+export interface AggregatedStats {
+	/** Total number of requests */
+	totalRequests: number;
+	/** Number of successful requests */
+	successfulRequests: number;
+	/** Number of failed requests */
+	failedRequests: number;
+	/** Error rate (0-1) */
+	errorRate: number;
+	/** Total input tokens */
+	totalInputTokens: number;
+	/** Total output tokens */
+	totalOutputTokens: number;
+	/** Total cache read tokens */
+	totalCacheReadTokens: number;
+	/** Total cache write tokens */
+	totalCacheWriteTokens: number;
+	/** Cache hit rate (0-1) */
+	cacheRate: number;
+	/** Total cost */
+	totalCost: number;
+	/** Total premium requests */
+	totalPremiumRequests: number;
+	/** Average duration in ms */
+	avgDuration: number | null;
+	/** Average TTFT in ms */
+	avgTtft: number | null;
+	/** Average tokens per second (output tokens / duration) */
+	avgTokensPerSecond: number | null;
+	/** Time range */
+	firstTimestamp: number;
+	lastTimestamp: number;
+}
+
+/**
+ * Stats grouped by model.
+ */
+export interface ModelStats extends AggregatedStats {
+	model: string;
+	provider: string;
+}
+
+/**
+ * Stats grouped by folder.
+ */
+export interface FolderStats extends AggregatedStats {
+	folder: string;
+}
+
+/**
+ * Time series data point.
+ */
+export interface TimeSeriesPoint {
+	/** Bucket timestamp (start of hour/day) */
+	timestamp: number;
+	/** Request count */
+	requests: number;
+	/** Error count */
+	errors: number;
+	/** Total tokens */
+	tokens: number;
+	/** Total cost */
+	cost: number;
+}
+
+/**
+ * Model usage time series data point (daily buckets).
+ */
+export interface ModelTimeSeriesPoint {
+	/** Bucket timestamp (start of day) */
+	timestamp: number;
+	/** Model name */
+	model: string;
+	/** Provider name */
+	provider: string;
+	/** Request count */
+	requests: number;
+}
+
+/**
+ * Model performance time series data point (daily buckets).
+ */
+export interface ModelPerformancePoint {
+	/** Bucket timestamp (start of day) */
+	timestamp: number;
+	/** Model name */
+	model: string;
+	/** Provider name */
+	provider: string;
+	/** Request count */
+	requests: number;
+	/** Average TTFT in ms */
+	avgTtft: number | null;
+	/** Average tokens per second */
+	avgTokensPerSecond: number | null;
+}
+
+/**
+ * Cost time series data point (daily buckets).
+ */
+export interface CostTimeSeriesPoint {
+	/** Bucket timestamp (start of day) */
+	timestamp: number;
+	/** Model name */
+	model: string;
+	/** Provider name */
+	provider: string;
+	/** Total cost for this bucket */
+	cost: number;
+	/** Cost breakdown */
+	costInput: number;
+	costOutput: number;
+	costCacheRead: number;
+	costCacheWrite: number;
+	/** Request count */
+	requests: number;
+}
+
+/**
+ * Overall dashboard stats.
+ */
+export interface DashboardStats {
+	overall: AggregatedStats;
+	byModel: ModelStats[];
+	byFolder: FolderStats[];
+	timeSeries: TimeSeriesPoint[];
+	modelSeries: ModelTimeSeriesPoint[];
+	modelPerformanceSeries: ModelPerformancePoint[];
+	costSeries: CostTimeSeriesPoint[];
+}
+
+/**
+ * Behavior time-series point (daily bucket, per responding model).
+ */
+export interface BehaviorTimeSeriesPoint {
+	/** Bucket timestamp (start of day) */
+	timestamp: number;
+	/** Responding model ("unknown" if user msg never got a reply) */
+	model: string;
+	/** Responding provider */
+	provider: string;
+	/** Number of user messages in bucket */
+	messages: number;
+	/** Total yelling sentences in bucket */
+	yelling: number;
+	/** Total profanity hits in bucket */
+	profanity: number;
+	/** Total anguish signal in bucket */
+	anguish: number;
+	/** Total corrective-negation hits in bucket */
+	negation: number;
+	/** Total user-repeating-themselves hits in bucket */
+	repetition: number;
+	/** Total second-person blame hits in bucket */
+	blame: number;
+	/** Total characters in bucket */
+	chars: number;
+}
+
+export interface BehaviorOverallStats {
+	totalMessages: number;
+	totalYelling: number;
+	totalProfanity: number;
+	totalAnguish: number;
+	totalNegation: number;
+	totalRepetition: number;
+	totalBlame: number;
+	totalChars: number;
+	firstTimestamp: number;
+	lastTimestamp: number;
+}
+
+/**
+ * Per-model behavioral aggregate over the active range.
+ */
+export interface BehaviorModelStats {
+	model: string;
+	provider: string;
+	totalMessages: number;
+	totalYelling: number;
+	totalProfanity: number;
+	totalAnguish: number;
+	totalNegation: number;
+	totalRepetition: number;
+	totalBlame: number;
+	totalChars: number;
+	lastTimestamp: number;
+}
+
+export interface BehaviorDashboardStats {
+	overall: BehaviorOverallStats;
+	byModel: BehaviorModelStats[];
+	behaviorSeries: BehaviorTimeSeriesPoint[];
+}