Trading System — Mean Reversion Lab

Jump to: Top bar & live prices · Can I trust this? · Chart & Levels · Backtest · Walk-Forward · Research · Portfolio · Paper Trading · Alerts · Self-learning · Data engine · What it is NOT · Invite friends

0 · Honest expectations first

The validated strategies here historically produce out-of-sample Sharpe ratios around 0.5–1.2 — that means realistic expectations of roughly 0.5–2% per month at moderate risk, with losing weeks and losing months along the way. Anything promising much more is either leverage (which multiplies losses identically) or overfitting. "Easily" in trading means: the process is easy to follow — the returns still require patience and discipline.

0½ · "How do I know this is actually tested correctly?" — check it yourself

Don't take it on faith. Press the button: it runs a set of deterministic proofs against the live engine right now and shows you each result. These verify the machinery is honest — it can't secretly peek at the future, it charges the costs it claims, its Sharpe/drawdown math is correct, and its out-of-sample test is genuinely out-of-sample.

Seven independent safeguards stand between a random-looking pattern and your advised book: (1) the engine decides at a bar's close and trades the NEXT bar — proven no look-ahead; (2) every strategy is scored only on walk-forward out-of-sample data it never trained on; (3) the qualification bar is deflated — the more strategies scanned, the higher the Sharpe required (multiple-testing correction); (4) the luck test (Backtest → Validate) checks whether the timing beats random with the same exposure; (5) the weekly sweep judges each strategy by its median across hundreds of markets/windows, never its best cell; (6) paper trading executes the advice for real and measures the live-vs-backtest gap; (7) anything that loses real money despite good backtests is quarantined. What none of this can promise: that a validated edge makes money tomorrow — no backtest can. It proves the process is honest, not that the future is known.

1 · Your 3-minute daily routine (the app works even when you don't)

☀️ 07:30 — the morning brief lands on your phone: account & every book's health, throttles/kill-switches, the 📏 reality gaps, what the overnight learning changed, and today's macro calendar (FOMC/CPI/NFP). Nothing orange? You're done for the morning.
During the day: pushes arrive only for NEW walk-forward-validated entries (entry/stop/target/leverage) or protective events (⚡ kill switch, 🛑 brake, quarantine). No news = healthy and trading.
Evening (optional): Paper tab — the “all portfolios” candles are your combined equity; 📏 on each card tells you if reality tracks the promise.
Weekly ritual (~10 min): Portfolio → 🎯 Build → read the 📋 decision card → still happy with the dial? If weights drifted, re-save the set to rebalance. Saturday nights the sweep auditions new lab strategies automatically.
The nightly report (~23:35 push) confirms the re-fit ran, what got quarantined or promoted, and that tomorrow's advice is ready.

1½ · ☁️ One brain in the cloud, many screens

The system runs 24/7 on its own server (a free Oracle cloud machine in Paris). Your computer, your phone, your friends' browsers are all just windows onto the same live state — open the app anywhere and you see the same books, positions and prices.
Nothing depends on any laptop being on. Books trade, exits fire, the nightly learning runs at 23:15, alerts push — all server-side.
Local changes do NOT mirror anywhere: the server is the single source of truth for your money-state; code improvements reach it only through a deliberate, tested deploy. By design there is exactly ONE writer — two would double-trade every book.
Security: the server exposes zero open ports; the public address rides an encrypted tunnel; login + invite code protect the app itself; real-money trading remains triple-locked OFF.

2 · The top bar & live prices — your cockpit (always visible)

🔍 Search any market (left): type a name or ticker — apple, gold, nasdaq 24. Results blend Yahoo symbols with the full 15,000-instrument eToro catalog, including the always-open variants marked with a teal 24/7 badge (NSDQ100.24-7, TSLA.24-7, GOLD.24-7…). Click a result and it charts instantly.
Category · Symbol dropdowns: the curated universe (indices, stocks, ETFs, commodities, forex, crypto, eToro tradeable prices, local CSVs).
Interval · Range: 5m→monthly, 1 day→max. Whatever you pick here is what Backtest and Walk-Forward run on — day-trading research means choosing 15m here first.
Market-state chip: ● green = market open, red = closed, teal pulsing = a 24/7 market, plus pre-market / after-hours — same idea as eToro's "Market Open / Closed".
Live price: the selected asset's price ticks every ~5 s and flashes green/red; the chart's last candle moves with it. Crypto and .24-7 instruments tick all weekend.
Ticker strip (under the header): live prices for the whole current category — click any of them to chart it. It's your watchlist.

3 · Chart & Levels — see the setup

Everything drawn on the chart is explained by the Key strip above the candles (toggle it with the "Key" checkbox once you've learned it). The full colour/notation scheme:

Colour = which side: teal/green = support (price tends to bounce UP off it), rose/pink = resistance (price tends to be capped BELOW it). This holds for both the diagonal trendlines and the horizontal zones.
Trendlines (diagonal): a solid line is the fitted trendline through the pivots; a dashed line is its channel — the parallel boundary on the opposite side (same slope). Toggle Log scale and they re-fit in log geometry — essential for crypto and multi-year charts.
S/R zones (horizontal dashed bands, labelled on the right price axis): S×N = a Support zone, R×N = a Resistance zone, and ×N is the number of times price has tested that level. More touches = a stronger, more respected level (S×24 is a wall; S×2 is the minimum to qualify). Round-number zones rank higher — that's where real orders cluster.
Unfilled gaps (orange dotted): open gaps that never filled — price often returns to "fill" them, so they act as magnets/hurdles.
Signals overlay (Signals dropdown): pick any strategy to stamp its trades on the candles — ▲ L = go Long (buy), ▼ S = go Short (sell), ● x = eXit the position back to cash. The fastest way to see what a strategy actually does before trusting its numbers.
Pattern callouts (text above the chart): wedges/triangles with apex distance, and head-&-shoulders with neckline + measured target when present.
MA row: up to three SMA/EMA overlays, any period, computed on the visible timeframe.
eToro banner: when charting an ETORO: instrument you also get its leverage tiers and fee rules right above the candles.

Intraday history grows every night: the system archives fresh 15m/1h bars into its own vault (already past Yahoo's 60-day window), and crypto intraday pulls years straight from Binance. The longer the app runs, the deeper your intraday backtests reach.

4 · Backtest — the first filter, never the proof

The lab holds 2,180 strategies in 80 families, every family cited to published research or a canonical trading text (the 📚 line under the picker; hover any entry for its source).

Filter box: search by name, family or source — try fomc, vwap, connors, overnight, breakout. Entries like RSI(2) Pullback [deep5] are named parameter presets ("variants") of a family, each testable on its own.
Parameters: pre-filled with the published defaults; hover a field to see the grid walk-forward would search. Change anything and re-run instantly.
Execution: leverage 1×/2×/5×, fee and slippage in bps. Rule of thumb: eToro crypto ≈ 100 bps fee, index CFDs ≈ 5 bps spread, stocks CFD ≈ 15 bps.
Run backtest → equity vs buy-and-hold, metric tiles, trade list. It runs on the top bar's symbol / interval / range.
Validate (luck test) → bootstrap Sharpe CI (if it includes zero, the edge may not exist) + permutation p-value (would random timing do as well? p < 0.05 = real timing information).

Rule: a good-looking backtest that fails validation is a coincidence, not a strategy.

Still worth it once Walk-Forward exists? Yes — they do different jobs. Backtest is the microscope: instant, one strategy, the whole equity curve + trade list + entries drawn on the candles, plus the luck test — so you see how and why a strategy trades before spending a walk-forward run on it. Walk-Forward is the verdict (does the edge survive out-of-sample?), and it's what the advised books trust. Explore in Backtest, decide in Walk-Forward — a backtest that looks great but fails walk-forward is exactly the trap the two-step is built to catch.

5 · Walk-Forward — the honest test

Optimizes on a training window, trades the NEXT window blind, slides, repeats — simulating how you would actually have traded. This is the number the whole app trusts.

Strategies panel: the 80 families are listed by group with per-family all/none links. Tick include lab variants to see all 2,180; the filter box searches names, families and sources. Your selection survives filtering.
No size limit: select 700 strategies if you want — the run splits into batches of 100 and the leaderboard fills in live between batches, ranked by OOS Sharpe.
Symbols panel: tick any set; filter by name; "all shown" respects the filter. Type any ticker in the box (full catalog, like the top search — NVDA, BTC-USD, any Yahoo/eToro symbol) and pick it, or press Enter, to add and analyse it.
Click a result row for the stitched OOS equity curve plotted against buy & hold — if the strategy line doesn't beat the grey buy-and-hold line out-of-sample, the edge isn't there.
Scheme: train/test window sizes in bars, with presets. If history is short the window auto-fits (the ⤵ marker tells you it did).
Reading a row: OOS Sharpe ≥ 0.5 is respectable · WF eff near 1 = robust, under 0.5 = overfit · low param stability = the grid is fitting noise. Click the row for the stitched OOS equity curve and per-window parameter choices.

Every Saturday the sweep (millions of backtests) stress-tests the core families plus a rotating batch of 100 lab variants across symbols × 9 window schemes, judging each by its median — never its best cell. Robust variants get promoted into the advised pools automatically; fading ones get demoted. That's how 2,180 lab strategies stay honest.

6 · Research — know what you're buying

Type any ticker in the Research search (or just use the top-bar search — they stay in sync). For the selected asset you get, on one page:

Sentiment gauge: Bearish ↔ Bullish needle blending technicals and news tone.
Key stats: market cap, P/E, EPS, dividend, beta, margins, 52-week and day ranges, volume — the eToro "Overview" numbers.
Analyst view: buy/hold/sell consensus bar and price targets (low/mean/high vs current). Third-party opinions, not gospel.
Live news: latest headlines via Google News — immune to Yahoo throttling.
Trading on eToro: the instrument's asset class, leverage tiers (ESMA retail caps), spread/fee rules and overnight-fee rules — read this BEFORE sizing a real trade.

7 · Portfolio — the advised book: scopes, dials, costs, save

Start simple — two dials and one button: set 🛡 Risk (the max drawdown you accept; it is a hard limit, never exceeded by anything else) and ⚖️ smooth ⟷ returns (left = straightest possible equity line, right = chase maximum return, middle = best balance), then click 🎯 Build my portfolio. Everything else lives behind ⚙ Advanced and is optional.

Read the result top-down: the 📋 decision card states your plan in the three numbers that matter (return /mo · worst dip · straightness), then the honest walk-forward line (the same blend re-decided using only past data — trust it over the headline), and a Confidence grade earned from four independent checks: walk-forward ✓, statistical range ✓ (bootstrap CI), luck-discount ✓ (survives multiple-testing), and real eToro costs ✓ including overnight financing charged for every bar held (forex/indices/commodities — added 2026-07-06, which trimmed forex's previously-flattered numbers by ~25%). The trade-off menu shows the same books at five dial settings — click use on any row. 💾 Create N tracked books in the 🎯 panel turns the whole diversified set into auto-trading paper books in one click (that is the way to track it — the single 💾 Save & track button saves only one scope's book).

Protections that only ever SHRINK exposure: the live regime gate 🛡 (blocks or halves entries whose family faces a hostile current regime), 📅 earnings blackouts (single stocks reporting within 3 days are skipped), CPI/NFP release-day halving for day books, the FOMC leverage cap, and an optional correlation-cluster cap. If your quality filters are too strict for today's market, 🔶 best-effort mode shows the closest achievable book, clearly labeled — never a silent empty answer, and never below the automatic anti-datamining bar.

Scope chips (top): Multi-asset, or a dedicated book per asset class — Stocks, ETFs, Indices, Commodities, Crypto, Forex. Each scope scans its own universe with only the strategy families the evidence supports there (daily-bar mean reversion never scans FX or commodities; intraday variants do).
Horizon slider: Day = true 15-minute bars, flat by the close · Swing = daily bars, 1–15 day holds · Long = decades of daily bars, trend & seasonal.
Max drawdown: your pain threshold. The book is scaled DOWN until its estimated drawdown fits — this cap always wins over the return target.
Target return /mo: a goal-check, not a promise. If validated edges can't reach it at your risk cap, the verdict banner says so instead of levering up.
Reading the table: every row passed walk-forward with a deflated bar (shown in the tiles — the more pairs scanned, the higher the required OOS Sharpe). You get direction, weight, suggested leverage (1× near FOMC, ≤2× crypto), entry/stop/target, OOS Sharpe, WF eff, and Est. cost — the real eToro round-trip: opening spread both ways + overnight CFD fees over the expected hold (weekend ×3). Hover it for the breakdown; a red ⚠ means fees eat ≥50% of the expected move — skip or shorten those.
💾 Save & track: freezes THIS combination (scope × horizon × sliders) as a named portfolio that trades automatically in Paper Trading with its own equity curve. This is how you turn advice into a track record.
The unused remainder stays in cash — cash is a position. Daily snapshots in reports/ keep the advice auditable.

8 · Paper Trading — your portfolios, live, day by day

New safety & honesty layer: each card shows a 📏 reality gap (the book's live %/mo vs what its backtest promised at save time — the number that keeps everyone honest), a ⛔ drawdown throttle (entries auto-halve past the book's dd target, pause at 1.5×), and an ⚡ daily max-loss kill switch you can arm per book (lose X% in one day → everything closes at market and entries pause until tomorrow, with a phone push). Next to Sync now sits the 🛑 account brake — a whole-account drawdown limit that slows ALL books together. The “all portfolios” equity renders as candles with 1m–1d timeframes; ⬇ CSV exports your full trade log; drag any column header to reorder tables.

This tab is a grid of portfolio cards — every book you saved, each with its own virtual capital ($10k), its own P&L and its own equity sparkline, updating live (the view refreshes every 30 s).

How they trade: the watcher executes each book on its horizon's rhythm — day books every minute on 15-minute marks (forced flat within the session), swing every ~15 minutes, long hourly. Fills at the latest price; slippage vs the advised entry is logged per trade.
Click a card to inspect that portfolio: its full day-by-day equity curve, its open positions and its closed trades. "Show all portfolios" returns to the combined view.
✕ deletes a book: its open positions close at market, its trade history stays in the account. Nothing is ever silently rewritten.
Exits are automatic: stop hit, target hit, or time (day: same session; swing: 20 days; long: 130 days).
The two numbers that matter: live win rate and average |slippage| — that's the live-vs-backtest reality gap. Give a book 4–8 weeks; 10 trades mean nothing, 100 start to.
Sync now forces an immediate pass; Reset account wipes everything (books included) back to a clean slate.

Each login (you, each friend) has its own private books, positions and history.

9 · Phone alerts — signals wherever you are

Install ntfy (Google Play, free) → subscribe to the personal topic shown in the Portfolio tab. Samsung: enable Instant delivery in ntfy and set Battery to Unrestricted, or pushes arrive late. You'll receive: new validated entries (entry / stop / target / leverage), FOMC warnings, and the nightly learning report. Day-trade signals are scanned every minute, swing at the cadence you set in the panel, long-term hourly — duplicates are never re-pushed.

10 · How the system improves itself (nobody has to remember to)

23:15 nightly: archives fresh intraday bars into the vault → re-fits every (symbol, strategy) pair's parameters IN PARALLEL on data including today → rebuilds all 21 scope×horizon books → recalibrates cost assumptions from measured paper slippage (floored — it learns costs UP toward reality, never down into fantasy) → quarantines any strategy whose live expectancy goes negative on 10+ real trades. Live evidence outranks backtests.
Saturday weekly: the robustness sweep re-judges the families plus a rotating batch of lab variants across all markets and window schemes by MEDIAN performance; winners are promoted into the advice, faders demoted.
What it deliberately does NOT do: chase the best cell of a grid, or raise leverage to hit your return target. Self-improvement here means self-correction from evidence.

11 · Where the data comes from (and why you can trust the charts)

Yahoo Finance for stocks/ETFs/indices/FX/commodities (decades of daily bars, ~60 days of 15-minute bars); Binance for crypto intraday — years of 5m/15m/1h bars, ticking 24/7; eToro's own candle feed for anything ETORO: (the prices you'd actually trade, incl. the .24-7 variants); your local CSVs. The archive vault stores every intraday bar the app ever fetches, so 15m history keeps growing past Yahoo's window every single night. Live quotes poll every ~5 s (Binance ticker for crypto = weekends too). Commodity =F symbols are continuous futures — for long backtests prefer the ETF proxies (GLD, USO).

12 · What this app is NOT (read once, remember forever)

It does not place real orders (yet) — you execute manually on eToro. It does not model margin calls at 2×/5×, and backtests don't include CFD overnight fees (the Portfolio's Est. cost column estimates them per position — read it). It cannot remove market risk: validated edges lose money on many individual days. Nothing here is financial advice — it is a research instrument that enforces discipline you would otherwise need willpower for.

13 · Glossary (30 seconds)

Sharpe: return per unit of risk; >1 is very good after costs. Drawdown: worst peak-to-valley loss — the number that makes people quit. OOS (out-of-sample): performance on data the optimizer never saw — the only kind that counts. WF efficiency: how much of the in-sample edge survives OOS. Expectancy: average %-gain per trade. CFD: a derivative position with financing costs — what leverage/shorts are on eToro. Book: a saved portfolio tracked in paper trading. Quarantine: strategy suspended for losing real (paper) money despite good backtests. Promotion: a lab variant that earned its way into the advice via the weekly sweep.

14 · On your phone & from anywhere

One address, everywhere: open https://tradingsys.tail95e19f.ts.net in any browser and log in. Android: install the app from https://tradingsys.tail95e19f.ts.net/TradingSys.apk (v1.4 — points at the cloud server), or Chrome ⋮ → Add to Home screen. Desktop: a “TradingSystem” launcher can live on your Mac desktop/Dock (opens the app in its own clean window). Push alerts arrive anywhere regardless (ntfy is cloud-delivered). No computer needs to be on — the server never sleeps.

15 · Inviting friends (and if you ARE the friend — start here)

Owner's part: send the friend two things — the app address https://tradingsys.tail95e19f.ts.net and the current invite code (shown in Portfolio tab → Account panel, owner only). Nothing else to install or configure.

The friend's list:

Open the address in any browser (phone or computer — no VPN, no Tailscale needed).
Tap Create account → choose a username and password (8+ characters) → enter the invite code.
Android: Chrome ⋮ → Add to Home screen, or download the app at /TradingSys.apk.
Real-time signals: install the free ntfy app → subscribe to YOUR personal topic (shown in the Portfolio tab after login). Samsung: Instant delivery + Battery Unrestricted.
Build a portfolio you like (Portfolio tab) → 💾 Save & track → watch its equity curve grow in Paper Trading.
Read section 0 ("Honest expectations") and section 12 ("What this app is NOT").

Every account is private: own saved portfolios, own paper account, own alert topic, own settings. Heavy compute (the weekly sweep) stays owner-only.

Account

Simulated account — live track record

My portfolios — each saved book trades on its own; make any of them a live bot (auto or confirm-each-order)

How to use this app — the investor's manual