Methodik · vorbereitet
Methodik v2
v2 extends resolution coverage beyond crypto. The migration that flips the live ledger from v1 to v2 sits in supabase/migrations/009_methodology_v2_scaffold.sql.template and is intentionally not applied yet — it requires the resolve-cron to be smoke-tested against stocks, metals and forex first.
What changes vs v1
- Resolution coverage: stocks via Yahoo's
v7/finance/quotebatch endpoint, forex via the same endpoint with=Xtickers, metals via metals-api spot rates (USD base). v1 was crypto-only via CoinGecko. - BIST coverage: Turkish equities (THYAO, GARAN, …) are resolved via the suffix-mapped tickers (
THYAO.ISetc.). This closes the Turkish-retail buyer-persona's BIST gap. - Thresholds: unchanged. Bullish accurate at ≥+0.5%, bearish at ≤-0.5%, neutral at |Δ| < 2.0%. Wilson 95% CI canonical, point estimate is the midpoint.
- Aggregation: v1 rows continue to live in the table at
methodology_version = 1. The live stats view is filtered to v2 only — figures from the two versions are never mixed.
When v2 goes live
The cron path is already deployed; the only gate is the migration. The team will flip the switch once we have run a manual smoke test that the resolve loop returns sane prices for at least one row in each non-crypto class. The migration also requires the page at /methodology/v2 (this one) to exist before it's applied — bumping the badge on /track-record without a target is a deliberate footgun the runbook prevents.
Why we don't backfill
v1 rows were scored under a different resolution-source set. Re-scoring them with v2 prices would produce numbers that look more authoritative than they are. We leave v1 rows alone and publish v2 as a fresh dataset that grows over time.
Audit hash
The thresholds in §1 and the resolution-source list in §1 jointly produce v2's identity. A change to either requires
methodology_version = 3 and a new page. The migration that introduces v2 is 009_methodology_v2.sql (currently .template until applied). Git history of that file is the audit trail.Past performance does not predict future performance. See Disclaimer. For live numbers see track record.