Self-hosted, Self-Improving, Federated Everything App — dev management, agent orchestration, persistent mind, digital twin, second brain, and creative pipeline from a single dashboard over Tailscale to port the user into the OS
JavaScript ai-agentsdashboarddeveloper-toolsdigital-twinquantified-selfsecond-brainself-hostedtailscale
/**
* Initialize the per-task-type recent-outcomes ring on data/cos/learning.json
* and bump the store version to 2 (issue #2460).
*
* Background: the CoS task-learning counters (`byTaskType[type].completed/
* succeeded/failed/successRate`) are lifetime-cumulative and never decay, so a
* burst of since-resolved failures permanently depresses the success rate the
* Layered Intelligence reasoner reads via its `cosMetrics` source. `metrics.js`
* now appends every completion to a bounded `recentOutcomes` ring, and LI reads a
* recency-windowed rate derived from it (`computeWindowedStats`). Existing
* installs' learning.json predate the ring; this migration adds an empty ring to
* each task-type bucket so the store shape matches v2.
*
* Deliberately EMPTY, not backfilled: `failureSignatures.recent` holds only
* FAILURE samples, so seeding the ring from it would fabricate an all-failure
* recent window — the exact false signal this issue removes. The ring fills from
* real outcomes going forward; until it holds samples, LI's windowed rate is null
* and the reasoner falls back to the intact lifetime rate.
*
* Idempotent: a second run finds every bucket already has a `recentOutcomes`
* array and the version already at 2, so it writes nothing.
*/
import { readFile, writeFile } from 'fs/promises';
import { join } from 'path';
const readJson = async (path, fallback) => {
const raw = await readFile(path, 'utf-8').catch((err) => {
if (err.code === 'ENOENT') return null;
throw err;
});
if (raw == null) return fallback;
try { return JSON.parse(raw); } catch { return fallback; }
};
const writeJson = (path, value) =>
writeFile(path, JSON.stringify(value, null, 2) + '\n');
export async function up({ rootDir }) {
const learningPath = join(rootDir, 'data', 'cos', 'learning.json');
const data = await readJson(learningPath, null);
// No learning.json yet — a fresh install seeds v2 with empty rings on first
// completion (DEFAULT_LEARNING_DATA + bucket init in metrics.js). Nothing to do.
if (!data || typeof data !== 'object') {
console.log('📚 cos-learning ring: no learning.json — nothing to migrate (fresh install seeds v2)');
return { migrated: false, initialized: 0 };
}
const byTaskType = data.byTaskType && typeof data.byTaskType === 'object' ? data.byTaskType : {};
let initialized = 0;
for (const metrics of Object.values(byTaskType)) {
if (metrics && typeof metrics === 'object' && !Array.isArray(metrics.recentOutcomes)) {
metrics.recentOutcomes = [];
initialized++;
}
}
// Forward-only: only bump an absent/older version up to 2. Never downgrade a
// store that is already at 2+ (a future v3, or a re-run against edited data).
const bumpedVersion = !(typeof data.version === 'number' && data.version >= 2);
if (bumpedVersion) data.version = 2;
if (initialized > 0 || bumpedVersion) {
await writeJson(learningPath, data);
console.log(`📚 cos-learning ring: initialized recentOutcomes on ${initialized} task type(s), version → 2`);
return { migrated: true, initialized };
}
console.log('✅ cos-learning ring: already initialized — no changes');
return { migrated: false, initialized: 0 };
}
export default { up };