jeopi-stats 16.2.13

package/src/shared-types.ts

@@ -0,0 +1,271 @@
+/**
+ * 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. `jeopi-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[];
+	byAgentType: AgentTypeStats[];
+	timeSeries: TimeSeriesPoint[];
+	modelSeries: ModelTimeSeriesPoint[];
+	modelPerformanceSeries: ModelPerformancePoint[];
+	costSeries: CostTimeSeriesPoint[];
+}
+
+/**
+ * Which agent produced a message, derived from its transcript file location
+ * inside the session directory: the top-level `<project>/<file>.jsonl` is the
+ * `main` agent, an `__advisor.jsonl` is the passive `advisor`, and any other
+ * nested transcript is a task `subagent`.
+ */
+export type AgentType = "main" | "subagent" | "advisor";
+
+/**
+ * Token usage aggregated by {@link AgentType} over the active range. Token
+ * columns are explicit so the dashboard's share denominator matches the
+ * counts it renders (input + output + cache read + cache write).
+ */
+export interface AgentTypeStats {
+	agentType: AgentType;
+	/** Total number of requests */
+	totalRequests: number;
+	/** Total input tokens */
+	totalInputTokens: number;
+	/** Total output tokens */
+	totalOutputTokens: number;
+	/** Total cache read tokens */
+	totalCacheReadTokens: number;
+	/** Total cache write tokens */
+	totalCacheWriteTokens: number;
+	/** Total cost */
+	totalCost: number;
+}
+
+/**
+ * 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[];
+}
+
+/** Token savings from a single source type. */
+export interface GainSourceTotals {
+	savedTokens: number;
+	savedBytes: number;
+	hits: number;
+	/** originalBytes - savedBytes, when original is known */
+	outputBytes: number;
+	/** Total original bytes before compression, when known */
+	originalBytes: number;
+	/** savedBytes / originalBytes when both are known, else null */
+	reductionPercent: number | null;
+}
+
+/** Per-source breakdown. */
+export type GainSource = "snapcompact";
+
+/** Time-series point for gain (daily bucket). */
+export interface GainTimeSeriesPoint {
+	date: string;
+	snapcompact: number;
+	total: number;
+}
+
+/** Complete gain dashboard payload. */
+export interface GainDashboardStats {
+	/** Aggregate across all sources for the active range. */
+	overall: GainSourceTotals;
+	/** Per-source breakdown. */
+	bySource: Record<GainSource, GainSourceTotals>;
+	/** Daily time series. */
+	timeSeries: GainTimeSeriesPoint[];
+	/** Active project filter (cwd prefix), or null for all projects. */
+	project: string | null;
+	/** All distinct projects seen in the data, for the selector. */
+	projects: string[];
+}

package/src/sync-worker.ts

@@ -0,0 +1,40 @@
+/**
+ * Stateless parse worker for `syncAllSessions`. The main thread owns the
+ * SQLite handle; workers receive `{ sessionFile, fromOffset }`, run
+ * `parseSessionFile` (which is pure I/O + CPU, no DB), and post the
+ * structured-clone-safe result back. One in-flight request per worker so
+ * the main thread can fan jobs out 1:1 with the pool size.
+ *
+ * A `{ kind: "ping" }` request is also accepted and replies with
+ * `{ ok: true, kind: "pong" }` — used by `smokeTestSyncWorker` to prove the
+ * worker actually spawns and runs in compiled binaries (regression coverage
+ * for issue #1011 / PR #1027, where the worker silently failed to load).
+ */
+
+import { type ParseSessionResult, parseSessionFile } from "./parser";
+
+export type SyncWorkerRequest = { kind?: "parse"; sessionFile: string; fromOffset: number } | { kind: "ping" };
+
+export type SyncWorkerResponse =
+	| { ok: true; kind?: "parse"; result: ParseSessionResult }
+	| { ok: true; kind: "pong" }
+	| { ok: false; error: string };
+
+declare const self: Worker & {
+	onmessage: ((event: MessageEvent<SyncWorkerRequest>) => void) | null;
+};
+
+self.onmessage = async event => {
+	const request = event.data;
+	try {
+		if (request.kind === "ping") {
+			self.postMessage({ ok: true, kind: "pong" } satisfies SyncWorkerResponse);
+			return;
+		}
+		const result = await parseSessionFile(request.sessionFile, request.fromOffset);
+		self.postMessage({ ok: true, result } satisfies SyncWorkerResponse);
+	} catch (err) {
+		const error = err instanceof Error ? (err.stack ?? err.message) : String(err);
+		self.postMessage({ ok: false, error } satisfies SyncWorkerResponse);
+	}
+};

package/src/types.ts

@@ -0,0 +1,128 @@
+import type { AssistantMessage, ServiceTier, ServiceTierByFamily, StopReason, Usage } from "jeopi-ai";
+import type { AgentType } from "./shared-types";
+
+export * from "./shared-types";
+
+/**
+ * Extracted stats from an assistant message.
+ */
+export interface MessageStats {
+	/** Database ID */
+	id?: number;
+	/** Session file path */
+	sessionFile: string;
+	/** Entry ID within the session */
+	entryId: string;
+	/** Folder/project path (extracted from session filename) */
+	folder: string;
+	/** Model ID */
+	model: string;
+	/** Provider name */
+	provider: string;
+	/** API type */
+	api: string;
+	/** Unix timestamp in milliseconds */
+	timestamp: number;
+	/** Request duration in milliseconds */
+	duration: number | null;
+	/** Time to first token in milliseconds */
+	ttft: number | null;
+	/** Stop reason */
+	stopReason: StopReason;
+	/** Error message if stopReason is error */
+	errorMessage: string | null;
+	/** Token usage */
+	usage: Usage;
+	/** Which agent produced this message (main agent, task subagent, advisor) */
+	agentType: AgentType;
+}
+
+/**
+ * Full details of a request, including content.
+ */
+export interface RequestDetails extends MessageStats {
+	/** The full conversation history or just the last turn. */
+	messages: unknown[];
+	/** The model's response. */
+	output: unknown;
+}
+
+/**
+ * Session log entry types.
+ */
+export interface SessionHeader {
+	type: "session";
+	version: number;
+	id: string;
+	timestamp: string;
+	cwd: string;
+	title?: string;
+}
+
+export interface SessionMessageEntry {
+	type: "message";
+	id: string;
+	parentId: string | null;
+	timestamp: string;
+	message: AssistantMessage | { role: "user" | "toolResult" };
+}
+
+export interface SessionServiceTierChangeEntry {
+	type: "service_tier_change";
+	id: string;
+	parentId?: string | null;
+	timestamp: string;
+	serviceTier: ServiceTierByFamily | ServiceTier | null;
+}
+
+export type SessionEntry = SessionHeader | SessionMessageEntry | SessionServiceTierChangeEntry | { type: string };
+
+/**
+ * Behavioral stats extracted from a single user message.
+ */
+export interface UserMessageStats {
+	/** Database ID */
+	id?: number;
+	/** Session file path */
+	sessionFile: string;
+	/** Entry ID within the session */
+	entryId: string;
+	/** Folder/project path */
+	folder: string;
+	/** Unix timestamp in ms */
+	timestamp: number;
+	/** Model that responded to this user message, if linked */
+	model: string | null;
+	/** Provider that responded to this user message, if linked */
+	provider: string | null;
+	/** Total characters of message text */
+	chars: number;
+	/** Whitespace-delimited word count */
+	words: number;
+	/** Yelling sentences (> 50% uppercase letters) */
+	yelling: number;
+	/** Profanity hits */
+	profanity: number;
+	/** Catch-all upset signal: drama runs + `noooo`/`ughh`/... + `dude` + `..` */
+	anguish: number;
+	/** Corrective negation ("no", "nope", "thats not what i meant") */
+	negation: number;
+	/** User repeating themselves ("i meant", "still doesnt work", "like i said") */
+	repetition: number;
+	/** Second-person reproach ("you didnt", "you broke", "stop X-ing") */
+	blame: number;
+}
+
+/**
+ * Pair emitted by the parser when it sees an assistant message whose
+ * `parentId` points to a user message that wasn't parsed in the same pass
+ * (e.g. user prompt landed in an earlier incremental sync). The aggregator
+ * applies the link to the persisted `user_messages` row so it stops showing
+ * up in the "unknown" model bucket.
+ */
+export interface UserMessageLink {
+	sessionFile: string;
+	entryId: string;
+	model: string;
+	provider: string;
+}