Changelog

Track record correction: the headline now measures calls from the market price, after estimated costs — our track record measured each call from the middle of its suggested entry zone and counted no trading costs, and our method note wrongly said entries were the price when the call was logged. We now lead with a corrected figure: each call measured from the engine’s market price at the moment of the call, after estimated Bybit fees and slippage (funding excluded). For calls before today that price was not stored and is recovered from the stored entry zone — an estimate of the engine’s signal price, not a verified fill, within ±0.5% on some low-priced coins. On the record as exported today (377 rows), the corrected sum of per-call moves is −198.72% (−180.68% with fees only, −231.03% with high slippage), against +122.38% under the original method. This is a sum of equal-size per-call percentages, not an account or portfolio return. No win or loss label changed, and the 44% win rate is the same under both methods. 39 calls had been scored as wins at an entry the market never reached; they are now marked. The original rows are published unaltered with a SHA-256 verification fingerprint (how to verify). New calls record their market price at signal time and are scored both ways. The strategy that produced these calls is retired as a trading candidate and continues as a public benchmark.

Track record: every call is scored on its own window, and a call we could not price is no longer closed on the spot — a call is judged only on prices inside its 21-day horizon (plus the two-hour check that first sees it), however late we price it, so a move after the horizon can never turn it into a win or a loss. When no exchange could price a call at its horizon, it is now held and retried for up to three days instead of being closed at once as an unpriced expiry. Only calls resolved late are affected — ones we could not price on time. A call resolved on time is scored exactly as before, and calls closed before this date keep their recorded result. Track record →

Track record: the profit factor now counts every scored closed call — gross gain ÷ gross loss summed in the track worker from each call’s stored outcome, with wins, losses and expired calls each on the side its own sign puts it (a 0.00% outcome on neither side; open and unscoreable calls left out). Before this change the page computed it from rounded averages over wins and losses only, which left expired calls out, so a profit factor quoted from before this date is not directly comparable. Snapshots that carry gross_gain_pct use the new definition. Track record →

Regime classifier inputs moved to FRED-primary (VIXCLS, broad dollar DTWEXBGS, SP500, credit spreads, yield curve) with exchange/Yahoo fallback. End-of-day official series replace intraday scraped quotes for classification; regime output verified unchanged on switch day. Every payload now carries a provenance field.

Confidence-tier inversion disclosed — our HIGH-confidence signal tier was underperforming the MEDIUM tier in live resolution; an on-page warning now says so and advises against weighting by tier while calibration is under review.honesty

Market Conditions Monitor replaces public trade setups — same structural inputs (funding, open interest, order flow, phase), published as conditions without entries, stops, targets or direction instructions. Analytics, not advice.

Rotation and positioning layers went live on the intelligence surface — sector rotation from exchange category flows; positioning from Hyperliquid's public API (venue-scoped and labeled as such). Unported layers are listed in a roadmap disclosure instead of rendering as empty panels.

Stated confidence calibrated against realized outcomes — published setup confidence began showing a calibrated figure derived from the realized win rate of its confidence band (Bayesian shrinkage), alongside the raw number.

Regime archive introduced — ~2 years of weekly regime classification (trend vs volatility rules), recomputed live from history on each load rather than stored claims. Central-bank stances and token-unlock amounts moved from curated values to live official/on-chain sources.

Fed & rates panel switched to live FRED after our own review caught stale hardcoded values (policy rate ~3 cuts out of date). The failure and the fix are both part of this record.honesty

Regime classifier v1 on edge infrastructure — six-indicator rule-based classification (volatility, credit spreads, yield curve, dollar, equity trend) with explicit reason strings; simplified port of the earlier research prototype.

Google sign-in scaffold (provider-gated) · live status page with real browser probes · FRED-provenance regime inputs · intelligence page de-scheduled (roadmap disclosure) · landing rebuilt around the product (globe removed) · signal research moved to a collapsed research register

Server-side plan gating on premium data endpoints (flip-ready) · Market Intelligence Brief — print-ready, sourced, timestamped IC report · sidebar line icons · skeleton loading · page-title cleanup across 11 languages

Daily Delta strip (what changed since your last visit) · ⌘K command palette · workflow navigation regroup · CSV export on conditions and the full track record · per-user plan enforcement behind the beta flag — including a fix where a paying customer could have read as free foreverfix

Live Stripe payment links + webhook plan sync · trust page (company, security, data provenance, honest SLA) · Copy Desk made internal-only — the private engine's mirror is not a customer product · Early Access reframe

Honesty pass on our own marketing: removed a fabricated social-proof toast and placeholder testimonials, corrected a false privacy claim in the FAQ (11 languages), fixed trial-length copyhonesty

Public engine status page — plain-language safety readout with full trade history, wins and losses, dated so old losses can't be misread as today's

Security audit: closed an account-data exposure on an internal endpoint, removed a founder URL backdoor, added per-IP rate limiting across all public data servicessecurity