pi-mega-compact 0.7.8 → 0.7.9

package/src/store/sqlite/raw-transcript.ts

@@ -0,0 +1,221 @@
+/**
+ * raw-transcript.ts — S27 durable raw-transcript mirror + checkpoint-epoch registry.
+ *
+ * ADDITIVE + flag-default-OFF (MEGACOMPACT_DB_MIRROR). No behavior change
+ * until the flag is flipped on; the tables are created IF NOT EXISTS on
+ * open so existing stores are untouched. These helpers are storage primitives
+ * only — the runtime hook (Task 5) wires them into the compaction flow. All
+ * queries use @-parameterized placeholders (PREVENT-002); no `any` types
+ * (PREVENT-011).
+ */
+import type { DatabaseSync } from "node:sqlite";
+import { withTx } from "./utils.js";
+
+/** One appended raw-message row in the durable mirror. */
+export interface RawTranscriptRow {
+  contentHash: string;
+  sessionId: string;
+  seq: number;
+  role: string;
+  contentBytes: string;
+  toolName: string | null;
+  /** ORIGINAL message timestamp captured at append time (NOT a served ts). */
+  messageTimestamp: number | null;
+  checkpointEpoch: string;
+}
+
+/** One checkpoint-epoch bookkeeping row (informational registry). */
+export interface CheckpointEpoch {
+  epochId: string;
+  sessionId: string;
+  startedSeq: number;
+  committedSeq: number;
+  summaryMessageText: string;
+  cutIndex: number;
+  checkpointId: string;
+  createdAt: number;
+}
+
+/** DB row shape for raw_transcript (snake_case column names). */
+interface RawTranscriptDBRow {
+  content_hash: string;
+  session_id: string;
+  seq: number;
+  role: string;
+  content_bytes: string;
+  tool_name: string | null;
+  message_timestamp: number | null;
+  checkpoint_epoch: string;
+}
+
+/** DB row shape for checkpoint_epochs (snake_case column names). */
+interface CheckpointEpochDBRow {
+  epoch_id: string;
+  session_id: string;
+  started_seq: number;
+  committed_seq: number;
+  summary_message_text: string;
+  cut_index: number;
+  checkpoint_id: string;
+  created_at: number;
+}
+
+function rowToRawTranscript(row: RawTranscriptDBRow): RawTranscriptRow {
+  return {
+    contentHash: row.content_hash,
+    sessionId: row.session_id,
+    seq: Number(row.seq),
+    role: row.role,
+    contentBytes: row.content_bytes,
+    toolName: row.tool_name ?? null,
+    messageTimestamp:
+      row.message_timestamp == null ? null : Number(row.message_timestamp),
+    checkpointEpoch: row.checkpoint_epoch,
+  };
+}
+
+function rowToCheckpointEpoch(row: CheckpointEpochDBRow): CheckpointEpoch {
+  return {
+    epochId: row.epoch_id,
+    sessionId: row.session_id,
+    startedSeq: Number(row.started_seq),
+    committedSeq: Number(row.committed_seq),
+    summaryMessageText: row.summary_message_text,
+    cutIndex: Number(row.cut_index),
+    checkpointId: row.checkpoint_id,
+    createdAt: Number(row.created_at),
+  };
+}
+
+/**
+ * Append one raw-message row to the durable mirror. Idempotent by
+ * (content_hash, session_id) via INSERT OR IGNORE — re-appending the same
+ * content for the same session is a no-op. seq is assigned server-side as
+ * COALESCE(MAX(seq),0)+1 within the session, so callers never need to compute
+ * it. Pass an open store handle (openStore) — matches the other DatabaseSync
+ * helpers. Parameterized (PREVENT-002).
+ */
+export function appendRawTranscript(db: DatabaseSync, row: RawTranscriptRow): void {
+  withTx(db, () => {
+    db.prepare(
+      `INSERT OR IGNORE INTO raw_transcript
+        (content_hash, session_id, seq, role, content_bytes, tool_name, message_timestamp, checkpoint_epoch)
+       VALUES (
+         @content_hash, @session_id,
+         COALESCE((SELECT MAX(seq) FROM raw_transcript WHERE session_id = @session_id), 0) + 1,
+         @role, @content_bytes, @tool_name, @message_timestamp, @checkpoint_epoch
+       )`,
+    ).run({
+      "@content_hash": row.contentHash,
+      "@session_id": row.sessionId,
+      "@role": row.role,
+      "@content_bytes": row.contentBytes,
+      "@tool_name": row.toolName,
+      "@message_timestamp": row.messageTimestamp,
+      "@checkpoint_epoch": row.checkpointEpoch,
+    });
+  });
+}
+
+/**
+ * List raw-transcript rows for a session in [fromSeq, toSeq], ordered by seq
+ * ascending. Returns camel-cased RawTranscriptRow[]. Parameterized.
+ */
+export function listRawTranscriptRange(
+  db: DatabaseSync,
+  sessionId: string,
+  fromSeq: number,
+  toSeq: number,
+): RawTranscriptRow[] {
+  const rows = db
+    .prepare(
+      `SELECT content_hash, session_id, seq, role, content_bytes, tool_name, message_timestamp, checkpoint_epoch
+       FROM raw_transcript
+       WHERE session_id = @session_id AND seq >= @from_seq AND seq <= @to_seq
+       ORDER BY seq ASC`,
+    )
+    .all({
+      "@session_id": sessionId,
+      "@from_seq": fromSeq,
+      "@to_seq": toSeq,
+    }) as unknown as RawTranscriptDBRow[];
+  return rows.map(rowToRawTranscript);
+}
+
+/**
+ * Insert (or refresh) a checkpoint-epoch row. ON CONFLICT(epoch_id) DO UPDATE
+ * so re-running the same compaction epoch is idempotent / refresh-safe.
+ * Parameterized (PREVENT-002).
+ */
+export function writeCheckpointEpoch(db: DatabaseSync, epoch: CheckpointEpoch): void {
+  withTx(db, () => {
+    db.prepare(
+      `INSERT INTO checkpoint_epochs
+        (epoch_id, session_id, started_seq, committed_seq, summary_message_text, cut_index, checkpoint_id, created_at)
+       VALUES (@epoch_id, @session_id, @started_seq, @committed_seq, @summary_message_text, @cut_index, @checkpoint_id, @created_at)
+       ON CONFLICT(epoch_id) DO UPDATE SET
+         session_id = excluded.session_id,
+         started_seq = excluded.started_seq,
+         committed_seq = excluded.committed_seq,
+         summary_message_text = excluded.summary_message_text,
+         cut_index = excluded.cut_index,
+         checkpoint_id = excluded.checkpoint_id,
+         created_at = excluded.created_at`,
+    ).run({
+      "@epoch_id": epoch.epochId,
+      "@session_id": epoch.sessionId,
+      "@started_seq": epoch.startedSeq,
+      "@committed_seq": epoch.committedSeq,
+      "@summary_message_text": epoch.summaryMessageText,
+      "@cut_index": epoch.cutIndex,
+      "@checkpoint_id": epoch.checkpointId,
+      "@created_at": epoch.createdAt,
+    });
+  });
+}
+
+/** Read one checkpoint-epoch row by id (or null if absent). Parameterized. */
+export function readCheckpointEpoch(db: DatabaseSync, epochId: string): CheckpointEpoch | null {
+  const row = db
+    .prepare(
+      `SELECT epoch_id, session_id, started_seq, committed_seq, summary_message_text, cut_index, checkpoint_id, created_at
+       FROM checkpoint_epochs WHERE epoch_id = @epoch_id`,
+    )
+    .get({ "@epoch_id": epochId }) as unknown as CheckpointEpochDBRow | undefined;
+  return row ? rowToCheckpointEpoch(row) : null;
+}
+
+/**
+ * Latest checkpoint-epoch row for a session (highest created_at), or null if
+ * none. Parameterized (PREVENT-002).
+ */
+export function getActiveEpochForSession(db: DatabaseSync, sessionId: string): CheckpointEpoch | null {
+  const row = db
+    .prepare(
+      `SELECT epoch_id, session_id, started_seq, committed_seq, summary_message_text, cut_index, checkpoint_id, created_at
+       FROM checkpoint_epochs
+       WHERE session_id = @session_id
+       ORDER BY created_at DESC
+       LIMIT 1`,
+    )
+    .get({ "@session_id": sessionId }) as unknown as CheckpointEpochDBRow | undefined;
+  return row ? rowToCheckpointEpoch(row) : null;
+}
+
+/** List all checkpoint epochs (diagnostic / test helper). */
+export function listCheckpointEpochs(db: DatabaseSync): CheckpointEpoch[] {
+  const rows = db
+    .prepare(
+      `SELECT epoch_id, session_id, started_seq, committed_seq, summary_message_text, cut_index, checkpoint_id, created_at
+       FROM checkpoint_epochs
+       ORDER BY created_at DESC`,
+    )
+    .all() as unknown as CheckpointEpochDBRow[];
+  return rows.map(rowToCheckpointEpoch);
+}
+
+/** Count raw transcript rows (diagnostic / test helper). */
+export function countRawTranscript(db: DatabaseSync): number {
+  const row = db.prepare(`SELECT COUNT(*) AS cnt FROM raw_transcript`).get() as { cnt: number };
+  return row.cnt;
+}

package/src/store/sqlite/schema.ts

@@ -0,0 +1,258 @@
+/**
+ * schema.ts — table creation, migrations, `ensureColumn`, PRAGMA setup.
+ */
+import { DatabaseSync } from "node:sqlite";
+
+const SCHEMA_VERSION = 2;
+
+/**
+ * Add `column` (with `decl`, e.g. "INTEGER") to `table` if it does not already
+ * exist. Idempotent: checks PRAGMA table_info first, so it is safe to run on
+ * every open. Table/column/decl are code-controlled constants (never user
+ * input), so the unavoidable identifier interpolation here does not violate
+ * PREVENT-002 (no external data reaches this SQL).
+ */
+export function ensureColumn(db: DatabaseSync, table: string, column: string, decl: string): void {
+  const cols = db.prepare(`PRAGMA table_info(${table})`).all() as Array<{ name: string }>;
+  if (cols.some((c) => c.name === column)) return;
+  db.exec(`ALTER TABLE ${table} ADD COLUMN ${column} ${decl}`);
+}
+
+export function initSchema(db: DatabaseSync): void {
+  db.exec(`
+    CREATE TABLE IF NOT EXISTS context_chunks (
+      id                  TEXT NOT NULL,
+      session_id         TEXT NOT NULL,
+      region_hash        TEXT,
+      content_hash       TEXT,
+      content_hash2      TEXT,
+      content_hash_version INTEGER,
+      normalized_text    TEXT,
+      summary            TEXT,
+      topic_summary      TEXT,
+      summary_hash       TEXT,
+      key_decisions      TEXT,           -- JSON array
+      next_steps         TEXT,           -- JSON array
+      files_modified     TEXT,           -- JSON array
+      embedding_blob     BLOB,           -- float32 vector
+      token_estimate     INTEGER,
+      original_token_estimate INTEGER,    -- dropped region size (tokens saved = orig − stored)
+      timestamp          INTEGER,
+      dedup_status       TEXT DEFAULT 'active',
+      compressed_original BLOB           -- optional DR copy
+    );
+    -- Primary key is (session_id, id): checkpoint ids are unique per session
+    -- (chkpt_001 per session), not globally, so a bare id PK would collide
+    -- across sessions on the nextCheckpointId sequence.
+    CREATE UNIQUE INDEX IF NOT EXISTS idx_chunks_pk
+      ON context_chunks(session_id, id);
+    CREATE INDEX IF NOT EXISTS idx_chunks_session ON context_chunks(session_id);
+    CREATE INDEX IF NOT EXISTS idx_chunks_region ON context_chunks(region_hash);
+    CREATE INDEX IF NOT EXISTS idx_chunks_content ON context_chunks(content_hash);
+    -- Partial UNIQUE (QA #1): null content_hash rows never violate the constraint;
+    -- ON CONFLICT DO NOTHING makes backfill + L0 inserts safe.
+    CREATE UNIQUE INDEX IF NOT EXISTS idx_content_hash
+      ON context_chunks(session_id, content_hash) WHERE content_hash IS NOT NULL;
+
+    -- Sprint 11: MinHash signature + LSH bucket tables for L1 near-dup dedup.
+    CREATE TABLE IF NOT EXISTS minhash_signatures (
+      chunk_id          TEXT NOT NULL,
+      session_id        TEXT NOT NULL,
+      signature_version INTEGER NOT NULL,
+      signatures        TEXT NOT NULL,   -- JSON array of 256 uint32
+      PRIMARY KEY (chunk_id, signature_version)
+    );
+    CREATE INDEX IF NOT EXISTS idx_minhash_session ON minhash_signatures(session_id);
+
+    CREATE TABLE IF NOT EXISTS dedup_lsh_buckets (
+      bucket_key        TEXT NOT NULL,
+      chunk_id          TEXT NOT NULL,
+      session_id        TEXT NOT NULL,
+      signature_version INTEGER NOT NULL,
+      PRIMARY KEY (bucket_key, chunk_id)
+    );
+    CREATE INDEX IF NOT EXISTS idx_lsh_bucket ON dedup_lsh_buckets(bucket_key, session_id);
+
+    CREATE TABLE IF NOT EXISTS session_state (
+      session_id               TEXT PRIMARY KEY,
+      injected_checkpoint_ids TEXT,      -- JSON array
+      stored_region_hashes    TEXT       -- JSON array
+    );
+
+    CREATE TABLE IF NOT EXISTS meta (
+      key   TEXT PRIMARY KEY,
+      value TEXT
+    );
+
+    -- Sprint 13 (RAPTOR): hierarchical summary tree nodes. children are a JSON
+    -- array of child node ids (or raw leaf ids at the bottom); embedding_blob
+    -- is the node centroid. Additive; retrieval ignores this table until
+    -- Sprint 14 promotes RAPTOR out of shadow mode.
+    CREATE TABLE IF NOT EXISTS raptor_nodes (
+      id            TEXT NOT NULL,
+      session_id   TEXT NOT NULL,
+      level        INTEGER NOT NULL,
+      parent_id    TEXT,
+      children     TEXT,           -- JSON array of child ids
+      summary      TEXT,
+      embedding_blob BLOB,         -- float32 centroid
+      quality_marker TEXT DEFAULT 'low',
+      token_estimate INTEGER,
+      built_at     INTEGER,        -- S25: epoch ms when the tree was built (freshness guard)
+      PRIMARY KEY (session_id, id)
+    );
+    CREATE INDEX IF NOT EXISTS idx_raptor_session ON raptor_nodes(session_id);
+    CREATE INDEX IF NOT EXISTS idx_raptor_parent ON raptor_nodes(parent_id);
+
+    -- Foundation for future features (resume sessions, daily log, lessons
+    -- learned). Scaffolded now so all store data lives in SQLite from day one;
+    -- population is minimal (touchSession / logDaily on compact) and the full
+    -- UI/recall for these lands in later sprints.
+
+    -- Per-session registry (resume + per-repo session history).
+    CREATE TABLE IF NOT EXISTS sessions (
+      session_id    TEXT PRIMARY KEY,
+      repo          TEXT,
+      started_at    INTEGER,
+      ended_at      INTEGER,
+      last_compacted_at INTEGER,
+      status        TEXT DEFAULT 'active'
+    );
+
+    -- Append-only daily activity log (the "daily log" feature seed).
+    CREATE TABLE IF NOT EXISTS daily_log (
+      id            INTEGER PRIMARY KEY AUTOINCREMENT,
+      day           TEXT NOT NULL,     -- YYYY-MM-DD
+      session_id    TEXT,
+      event         TEXT,             -- e.g. 'compact'
+      detail        TEXT,
+      tokens_saved  INTEGER DEFAULT 0,
+      ts            INTEGER
+    );
+    CREATE INDEX IF NOT EXISTS idx_daily_log_day ON daily_log(day);
+
+    -- Active model/provider for cost estimation + the future multi-repo
+    -- dashboard (Phase 5b). One row per (repo, model change); latest wins.
+    CREATE TABLE IF NOT EXISTS model_snapshots (
+      id            INTEGER PRIMARY KEY AUTOINCREMENT,
+      repo_root     TEXT NOT NULL,
+      provider      TEXT NOT NULL,
+      provider_name TEXT,
+      model_id      TEXT NOT NULL,
+      model_name    TEXT,
+      input_rate    REAL,          -- USD per input token (Model.cost.input)
+      output_rate   REAL,          -- USD per output token (Model.cost.output)
+      context_window INTEGER,
+      max_tokens    INTEGER,
+      reasoning     INTEGER DEFAULT 0,
+      captured_at   INTEGER
+    );
+    CREATE INDEX IF NOT EXISTS idx_model_repo ON model_snapshots(repo_root);
+
+    -- Lessons learned (future recall/browse feature seed).
+    CREATE TABLE IF NOT EXISTS lessons (
+      id            INTEGER PRIMARY KEY AUTOINCREMENT,
+      session_id    TEXT,
+      repo          TEXT,
+      lesson        TEXT,
+      ts            INTEGER
+    );
+
+    -- Durable "save to memory" store (taken over from memory extensions).
+    -- One row per saved memory; scoped by repo so memory travels with the
+    -- clone. All params are parameterized (PREVENT-002).
+    CREATE TABLE IF NOT EXISTS memories (
+      id                INTEGER PRIMARY KEY AUTOINCREMENT,
+      repo              TEXT,
+      kind              TEXT DEFAULT 'note',   -- note | fact | decision | preference
+      content           TEXT NOT NULL,
+      tags              TEXT,                   -- JSON array of strings
+      created_at        INTEGER,
+      last_recalled_at  INTEGER,
+      -- S20 memory-RAG extension (auto-review add/replace/remove ops).
+      category          TEXT,                   -- typed bucket, e.g. decision | fact | preference
+      target            TEXT,                   -- optional subject/scope this memory targets
+      last_referenced   INTEGER,               -- last time memory was referenced by recall (epoch s)
+      source_turn       INTEGER                -- conversation turn that produced this memory
+    );
+    CREATE INDEX IF NOT EXISTS idx_memories_repo ON memories(repo);
+
+    -- FTS5 trigram virtual table (Sprint 9+ pg_trgm-equivalent verification).
+    CREATE VIRTUAL TABLE IF NOT EXISTS context_chunks_trgm USING fts5(
+      id UNINDEXED,
+      normalized_text,
+      tokenize='trigram'
+    );
+
+    -- S27: durable raw-transcript mirror (MEGACOMPACT_DB_MIRROR). Appended
+    -- RAW message bytes per session so a compacted window can be rehydrated
+    -- from the local store instead of the pi runtime transcript (which is
+    -- trimmed). PK is (content_hash, session_id) — NOT content_hash alone —
+    -- so identical content in different sessions never collides. Additive:
+    -- CREATE TABLE IF NOT EXISTS leaves existing DBs untouched on open until
+    -- the S27 mirror flag is flipped on. All queries parameterized (PREVENT-002).
+    CREATE TABLE IF NOT EXISTS raw_transcript (
+      content_hash      TEXT NOT NULL,
+      session_id        TEXT NOT NULL,
+      seq               INTEGER NOT NULL,
+      role              TEXT NOT NULL,
+      content_bytes     TEXT NOT NULL,
+      tool_name         TEXT,
+      message_timestamp INTEGER,          -- ORIGINAL msg ts at append, NOT served
+      checkpoint_epoch  TEXT NOT NULL,
+      PRIMARY KEY (content_hash, session_id)
+    );
+    CREATE INDEX IF NOT EXISTS idx_rt_session_seq ON raw_transcript(session_id, seq);
+    CREATE INDEX IF NOT EXISTS idx_rt_epoch ON raw_transcript(checkpoint_epoch);
+
+    -- S27: checkpoint-epoch registry. One row per compaction epoch; the
+    -- summary_message_text is the verbatim system message that replaced the
+    -- trimmed prefix. Informational bookkeeping (the raw_transcript rows are
+    -- authoritative); refresh-safe via ON CONFLICT(epoch_id) DO UPDATE.
+    CREATE TABLE IF NOT EXISTS checkpoint_epochs (
+      epoch_id             TEXT PRIMARY KEY,
+      session_id           TEXT NOT NULL,
+      started_seq          INTEGER NOT NULL,
+      committed_seq        INTEGER NOT NULL,
+      summary_message_text TEXT NOT NULL,
+      cut_index            INTEGER NOT NULL,
+      checkpoint_id        TEXT NOT NULL,
+      created_at           INTEGER NOT NULL
+    );
+    CREATE INDEX IF NOT EXISTS idx_epoch_session ON checkpoint_epochs(session_id, created_at DESC);
+
+    -- S27 Task 6: dedup_mirror for space-efficient deduplicated storage.
+    -- Each unique content_hash stores its bytes ONCE; raw_transcript rows
+    -- reference this table via content_ref instead of storing duplicate content_bytes inline.
+    CREATE TABLE IF NOT EXISTS dedup_mirror (
+      content_hash    TEXT PRIMARY KEY,
+      content_bytes   TEXT NOT NULL,
+      ref_count       INTEGER NOT NULL DEFAULT 1,
+      first_seen_seq  INTEGER NOT NULL,
+      created_at      INTEGER NOT NULL
+    );
+  `);
+  // Idempotent column migrations. `CREATE TABLE IF NOT EXISTS` is a no-op on a
+  // pre-existing table, so new columns added to context_chunks after a store was
+  // first created (e.g. original_token_estimate in v0.4.2) must be ALTERed in for
+  // databases created by an older version — otherwise repoStats()/upsert crash
+  // with "no such column" and the extension fails to load. Additive only.
+  ensureColumn(db, "context_chunks", "original_token_estimate", "INTEGER");
+  // S27 Task 6: content_ref column in raw_transcript for dedup_mirror references.
+  ensureColumn(db, "raw_transcript", "content_ref", "TEXT");
+  // S20 memory-RAG extension: additive columns for auto-review ops. Idempotent —
+  // only alters DBs created by an older version that lack these columns.
+  ensureColumn(db, "memories", "category", "TEXT");
+  ensureColumn(db, "memories", "target", "TEXT");
+  ensureColumn(db, "memories", "last_referenced", "INTEGER");
+  ensureColumn(db, "memories", "source_turn", "INTEGER");
+  // S25: RAPTOR freshness-guard timestamp. Additive; old DBs have NULL → 0 →
+  // treated as stale → flat fallback (safe).
+  ensureColumn(db, "raptor_nodes", "built_at", "INTEGER");
+  const v = db.prepare("SELECT value FROM meta WHERE key='schema_version'").get() as
+    | { value: string }
+    | undefined;
+  if (!v) {
+    db.prepare("INSERT INTO meta(key, value) VALUES(?, ?)").run("schema_version", String(SCHEMA_VERSION));
+  }
+}

package/src/store/sqlite/session-state.ts

@@ -0,0 +1,38 @@
+/**
+ * session-state.ts — `session_state` table (injection tracking).
+ */
+import type { DatabaseSync } from "node:sqlite";
+import type { SessionState } from "../../store.js";
+import { getStateDir, normalizeSessionId } from "../../store.js";
+import { openStore, jsonText } from "./utils.js";
+
+function loadSessionStateRow(sid: string, db: DatabaseSync): SessionState {
+  const row = db.prepare("SELECT * FROM session_state WHERE session_id = ?").get(sid) as any;
+  if (!row) {
+    return { injectedCheckpointIds: [], storedRegionHashes: [] };
+  }
+  return {
+    injectedCheckpointIds: row.injected_checkpoint_ids ? JSON.parse(row.injected_checkpoint_ids) : [],
+    storedRegionHashes: row.stored_region_hashes ? JSON.parse(row.stored_region_hashes) : [],
+  };
+}
+
+export function loadSessionState(sessionId: string, stateDir: string = getStateDir()): SessionState {
+  return loadSessionStateRow(normalizeSessionId(sessionId), openStore(stateDir));
+}
+
+export function saveSessionState(sessionId: string, state: SessionState, stateDir: string = getStateDir()): void {
+  const db = openStore(stateDir);
+  const sid = normalizeSessionId(sessionId);
+  db.prepare(
+    `INSERT INTO session_state(session_id, injected_checkpoint_ids, stored_region_hashes)
+     VALUES(@sid, @inj, @reg)
+     ON CONFLICT(session_id) DO UPDATE SET
+       injected_checkpoint_ids=excluded.injected_checkpoint_ids,
+       stored_region_hashes=excluded.stored_region_hashes`,
+  ).run({
+    sid,
+    inj: jsonText(state.injectedCheckpointIds),
+    reg: jsonText(state.storedRegionHashes),
+  });
+}

package/src/store/sqlite/stats.ts

@@ -0,0 +1,127 @@
+/**
+ * stats.ts — per-session, repo-wide, and data-invariant stats.
+ */
+import { getStateDir, normalizeSessionId } from "../../store.js";
+import { openStore } from "./utils.js";
+import { getMetaNumber, getDedupStats } from "./meta.js";
+
+export interface StoreStats {
+  checkpointCount: number;
+  totalTokenEstimate: number;
+  lastCheckpointId: string | undefined;
+  lastSummary: string | undefined;
+}
+
+export function storeStats(sessionId: string, stateDir: string = getStateDir()): StoreStats {
+  const db = openStore(stateDir);
+  const sid = normalizeSessionId(sessionId);
+  const row = db
+    .prepare(
+      `SELECT COUNT(*) AS c, COALESCE(SUM(token_estimate),0) AS tok,
+              MAX(id) AS lastId
+       FROM context_chunks WHERE session_id = ?`,
+    )
+    .get(sid) as { c: number; tok: number; lastId: string | null };
+  let lastSummary: string | undefined;
+  if (row.lastId) {
+    const s = db.prepare("SELECT summary FROM context_chunks WHERE id = ?").get(row.lastId) as
+      | { summary: string }
+      | undefined;
+    lastSummary = s?.summary;
+  }
+  return {
+    checkpointCount: row.c,
+    totalTokenEstimate: row.tok,
+    lastCheckpointId: row.lastId ?? undefined,
+    lastSummary,
+  };
+}
+
+/** Repo-wide stats — aggregates every session in this store (one per repo).
+ *  Backed by the SQLite `meta` cumulative counters (`tokens_saved`,
+ *  `dedup_attempts`, `deduped`) plus a SUM over all `context_chunks`. This is the
+ *  cumulative, resumable, cross-device view the dashboard surfaces as "Repo …". */
+export interface RepoStats {
+  /** Total checkpoints across all sessions (excludes SemDeDup-removed rows). */
+  checkpointCount: number;
+  /** Sum of all stored checkpoint token estimates (repo-wide). */
+  totalTokenEstimate: number;
+  /** Total active sessions with at least one checkpoint. */
+  sessionCount: number;
+  /** Cumulative stored-summary tokens saved (Σ stored summaries). */
+  tokensSaved: number;
+  /** Sum of original dropped-region token estimates (repo-wide). */
+  originalTokens: number;
+  /** Cumulative dedup add() attempts (store-wide). */
+  dedupAttempts: number;
+  /** Cumulative deduped collapses (store-wide). */
+  dedupCollapsed: number;
+  /** Storage dedup rate (deduped / attempts), 0..1. */
+  storageDedupRate: number;
+}
+
+/**
+ * Data-safety invariant metrics (Phase 0 — trust foundation). Proves that every
+ * compacted region is still recoverable: we retain a compressed_original blob for
+ * each checkpoint and permanently delete nothing. "removed" rows are SemDeDup
+ * duplicates whose ORIGINAL is still retained on the surviving checkpoint — they
+ * are not data loss, so they are reported separately, not as deletions.
+ */
+export interface DataInvariantStats {
+  /** Checkpoints with a recoverable compressed_original blob. */
+  regionsRetained: number;
+  /** Total bytes of compressed_original retained (recoverable verbatim). */
+  compressedOriginalBytes: number;
+  /** Checkpoints missing a compressed_original blob (pre-blob or direct add). */
+  regionsWithoutBlob: number;
+  /** Bytes permanently deleted by the extension. ALWAYS 0 — the invariant. */
+  bytesPermanentlyDeleted: number;
+  /** Duplicate rows collapsed by dedup (original retained on the survivor). */
+  duplicatesCollapsed: number;
+}
+
+export function dataInvariantStats(stateDir: string = getStateDir()): DataInvariantStats {
+  const db = openStore(stateDir);
+  const row = db
+    .prepare(
+      `SELECT
+         COUNT(compressed_original) AS withBlob,
+         COALESCE(SUM(LENGTH(compressed_original)),0) AS blobBytes,
+         SUM(CASE WHEN compressed_original IS NULL THEN 1 ELSE 0 END) AS noBlob
+       FROM context_chunks WHERE dedup_status != 'removed'`,
+    )
+    .get() as { withBlob: number; blobBytes: number; noBlob: number };
+  const removed = db
+    .prepare(`SELECT COUNT(*) AS c FROM context_chunks WHERE dedup_status = 'removed'`)
+    .get() as { c: number };
+  return {
+    regionsRetained: row.withBlob,
+    compressedOriginalBytes: row.blobBytes,
+    regionsWithoutBlob: row.noBlob ?? 0,
+    bytesPermanentlyDeleted: 0,
+    duplicatesCollapsed: removed.c,
+  };
+}
+
+export function repoStats(stateDir: string = getStateDir()): RepoStats {
+  const db = openStore(stateDir);
+  const row = db
+    .prepare(
+      `SELECT COUNT(*) AS c, COALESCE(SUM(token_estimate),0) AS tok,
+              COALESCE(SUM(original_token_estimate),0) AS orig,
+              COUNT(DISTINCT session_id) AS sessions
+       FROM context_chunks WHERE dedup_status != 'removed'`,
+    )
+    .get() as { c: number; tok: number; orig: number; sessions: number };
+  const ds = getDedupStats(stateDir);
+  return {
+    checkpointCount: row.c,
+    totalTokenEstimate: row.tok,
+    originalTokens: row.orig,
+    sessionCount: row.sessions,
+    tokensSaved: getMetaNumber("tokens_saved", stateDir),
+    dedupAttempts: ds.attempts,
+    dedupCollapsed: ds.deduped,
+    storageDedupRate: ds.attempts === 0 ? 0 : ds.deduped / ds.attempts,
+  };
+}