Incremental sync keeps stale exchanges and skips new content after a transcript is truncated or rewritten #193

"""Generate only synthetic transcript records for the two incremental bugs."""
import argparse,json,os,time
from pathlib import Path
parser=argparse.ArgumentParser(description=__doc__)
parser.add_argument("case",choices=["continuation","rewrite"])
parser.add_argument("phase",choices=["initial","changed"])
parser.add_argument("--root",type=Path,default=Path("demo"))
args=parser.parse_args()
project=args.root/"claude/projects/example-project"
project.mkdir(parents=True,exist_ok=True)
file=project/"00000000-0000-4000-8000-000000000001.jsonl"
old_mtime=file.stat().st_mtime if file.exists() else 0
def line(role,text,seq):
    return json.dumps({"type":role,"uuid":role+"-PLACEHOLDER-"+str(seq),
        "sessionId":"00000000-0000-4000-8000-000000000001",
        "timestamp":"2026-01-01T00:00:00.000Z","isSidechain":False,
        "message":{"role":role,"content":text if role=="user" else [{"type":"text","text":text}]}},separators=(",",":"))+"\n"
if args.case=="continuation":
    if args.phase=="initial":
        file.write_text(line("user","USER_PLACEHOLDER",1)+line("assistant","FIRST_ANSWER_PLACEHOLDER",1))
    else:
        assert file.is_file(),"Generate initial transcript and index it first"
        with file.open("a") as stream: stream.write(line("assistant","CONTINUATION_PLACEHOLDER",2))
else:
    content=(line("user","OLD_USER_PLACEHOLDER",1)+line("assistant","OLD_ANSWER_PLACEHOLDER",1)+
             line("user","OLD_TAIL_USER_PLACEHOLDER",2)+line("assistant","OLD_TAIL_ANSWER_PLACEHOLDER",2))
    if args.phase=="changed": content=line("user","NEW_USER_PLACEHOLDER",1)+line("assistant","REPLACEMENT_PLACEHOLDER",1)
    file.write_text(content)
stamp=max(time.time(),old_mtime)+5
os.utime(file,(stamp,stamp))
print(json.dumps({"case":args.case,"phase":args.phase,"file":file.relative_to(args.root).as_posix()}))

// Run only against examples/make_logs.py output; no model, account or real session.
import fs from 'node:fs';
import path from 'node:path';
import { pathToFileURL } from 'node:url';
import { createRequire } from 'node:module';
const [caseName, phase, checkout = 'upstream', state = 'demo'] = process.argv.slice(2);
if (!['continuation', 'rewrite'].includes(caseName) || !['initial', 'changed'].includes(phase)) {
  throw new Error('Usage: reproduce.mjs continuation|rewrite initial|changed [checkout] [state-directory]');
}
const bundle = process.cwd();
const repo = path.resolve(checkout);
const root = path.resolve(state);
for (const dir of [repo, root]) {
  if (!dir.startsWith(bundle + path.sep)) throw new Error('Use a directory within this local bundle');
}
for (const key of ['home', 'archive', 'config']) fs.mkdirSync(path.join(root, key), { recursive: true });
Object.assign(process.env, {
  HOME: path.join(root, 'home'),
  TEST_PROJECTS_DIR: path.join(root, 'claude', 'projects'),
  TEST_ARCHIVE_DIR: path.join(root, 'archive'),
  TEST_DB_PATH: path.join(root, 'index.db'),
  EPISODIC_MEMORY_CONFIG_DIR: path.join(root, 'config'),
});
const transcript = path.join(process.env.TEST_PROJECTS_DIR, 'example-project', '00000000-0000-4000-8000-000000000001.jsonl');
if (!fs.readFileSync(transcript, 'utf8').includes('PLACEHOLDER')) throw new Error('Expected synthetic fixture');
const { syncConversations } = await import(pathToFileURL(path.join(repo, 'dist', 'sync.js')).href);
const result = await syncConversations(process.env.TEST_PROJECTS_DIR, process.env.TEST_ARCHIVE_DIR, { skipSummaries: true });
if (result.errors.length) throw new Error('Synthetic sync failed: ' + result.errors.map(item => item.error).join('; '));
const require = createRequire(path.join(repo, 'package.json'));
const Database = require('better-sqlite3');
const db = new Database(process.env.TEST_DB_PATH);
require('sqlite-vec').load(db);
try {
  const rows = db.prepare('SELECT user_message, assistant_message FROM exchanges ORDER BY line_start').all();
  const marker = caseName === 'continuation' ? 'CONTINUATION_PLACEHOLDER' : 'REPLACEMENT_PLACEHOLDER';
  const count = (value) => rows.filter(row => row.assistant_message.includes(value)).length;
  console.log(JSON.stringify({
    case: caseName, phase, deterministic_embedding_stub: true, indexed_rows: rows.length,
    target_marker_hits: count(marker), old_tail_hits: count('OLD_TAIL_ANSWER_PLACEHOLDER'),
    vector_rows: db.prepare('SELECT COUNT(*) AS n FROM vec_exchanges').get().n,
    rows,
  }));
} finally { db.close(); }

// Test-only encoder stub. Parser, sync, SQLite and vector persistence stay real.
const fake = `
export const initEmbeddingsFailed = false;
export async function initEmbeddings() {}
export async function generateExchangeEmbedding() { return new Array(384).fill(0); }
export async function generateQueryEmbedding() { return new Array(384).fill(0); }
export async function generateEmbedding() { return new Array(384).fill(0); }
`;
export async function load(url, context, nextLoad) {
  if (url.endsWith('/dist/embeddings.js')) return { format: 'module', shortCircuit: true, source: fake };
  return nextLoad(url, context);
}