PatternForge AI — Complete User Guide
Everything PatternForge AI does, every formula it runs, every tab in the Command Center, and every keyboard shortcut on the chart overlay — written from and checked against the extension's own source code, not just the marketing copy. Free core forever; Pro $9/mo or Pro+ $79/yr unlock higher limits. Brought to you by ExtensionHub.app.
What Is PatternForge AI?
PatternForge AI is a Chrome extension (Manifest V3) that reads the symbol and timeframe off a live chart, matches the current price shape against thousands of bars of history, and reports what actually happened next — a BUY, SELL, or HOLD call with an entry, stop-loss, and take-profit level, backed by a measured historical win rate rather than a guess.
An optional AI layer (bring your own API key) adds a plain-language explanation on top of the quantitative call — it never overrides the direction, entry, stop, or target, all of which come from the replay engine.
Key Trader Pain Points & How PatternForge AI Solves Them
| Pain point | Why it hurts | How the extension solves it |
|---|---|---|
| "Is this actually a pattern, or am I imagining it?" | Manual pattern-reading is subjective under pressure. | Deterministic candlestick detector over the last 1–3 bars — same geometric rules every time. |
| "Where do I put my stop and target?" | Guessed distances cause premature stop-outs or oversized risk. | Stop distance is an ATR-style proxy from the signal bar's own range, floored at 0.4% of price so it never goes microscopic on a quiet bar. |
| "Has this setup actually worked before?" | Most indicators show a signal with zero evidence behind it. | Every matched analogue is walked forward bar-by-bar with the live signal's own stop/target and scored WIN/LOSS/TIMEOUT. |
| "How big should my position be?" | Oversized positions blow accounts; undersized ones waste good setups. | Risk Calculator tab turns balance + risk% + stop distance into an exact position size. |
| "Am I actually improving?" | Memory is a bad trading journal. | Journal tab auto-computes win rate, expectancy (in R), and a 12-week heat-map from logged trades. |
| "I can't watch five charts at once." | Missed setups on symbols you're not staring at. | Scanner tab runs the same engine across your whole watchlist and ranks by expectancy. |
| "My symbol/timeframe won't auto-detect." | Canvas-rendered terminals (MT5 web, some broker WebTraders) hide the DOM values the extension reads. | Manual symbol pinning — type the symbol once, it's remembered per website. |
| "I keep overtrading after a bad day." | Revenge-trading after a losing streak compounds losses. | Built-in daily risk gate (max daily loss in R, max open trades, max trades/day, news blackout) warns before you log another trade. |
PatternForge AI vs Other Chrome Trading Extensions: Where It Fits
Competitor map (what shows up in Store + SERP)
| Cluster | Examples | Core approach | Typical scope |
|---|---|---|---|
| Vision / “scan the chart image” AI | ChartRead, ChartSutra, TRAID, USstockRadar | Screenshot or VLM reads the chart → BULLISH/BEARISH + levels | Often “any site” or TradingView-heavy |
| LLM one-click on TradingView | Crypto SignAI, ChartMentor, ChartLense | Gemini/ChatGPT on the open TV chart → LONG/SHORT | Mostly TradingView only |
| Named-pattern / ICT overlays | SuperChartAI, various SMC tools | Order blocks, FVG, Elliott, BOS/CHOCH | TV or a few platforms |
| Execution / power tools | TVH Trading Power Tool | Risk size, multi-TP, order send | TV + some exchange charts |
| Backtest / optimizer | TradingView Assistant, TradeSearcher, Pineify-style optimizers | Parameter search on Pine strategies | TradingView only |
| Journal + AI on TV | ChartLense | AI read + visual journal | TradingView-focused |
PatternForge’s lane: quantitative historical shape matching + barrier replay on many broker chart URLs (TradingView, Forex.com WebTrader, MT5 web, Binance, OANDA, IG, etc.), with journal, risk calc, scanner, alerts, and optional BYOK AI—not “AI is the only engine.”
Feature fit (honest comparison)
| Capability | Typical AI signal extension | Typical TV-only optimizer | PatternForge AI |
|---|---|---|---|
| Works outside TradingView | Sometimes (image drag) | Rarely | Yes — multi-broker host list |
| Direction from opaque LLM | Common | N/A | Optional only |
| Direction from history + same SL/TP rules | Rare | N/A (strategy tester) | Primary path |
| Explicit HOLD when no edge | Rare (tools push a call) | N/A | Built-in |
| Trade journal in R | Some (e.g. ChartLense) | No | Yes + heatmap / CSV |
| Position size calculator | Occasional | No | Yes |
| Watchlist pattern scan | Some scanners | No | Yes |
| Manual pin for canvas terminals | Rare | No | Yes (Forex.com / MT5) |
| Does not place trades | Usually | Usually | Yes (analysis only) |
Where PatternForge fits: between pure signal apps (fast opinion) and pure calculators/journals (discipline without setup detection). It is closest to a research + risk workflow that still lives on the live chart.
Where it is not competing: Pine parameter optimizers, order-routing “power tools,” or paid order-flow feeds. Those solve different jobs.
Why Traders Rate Calculators Higher Than Signal Apps (and How PatternForge Responds)
What Google / review culture rewards
SERP and Store narratives keep returning to the same themes:
- Signals without process fail — blogs stress R:R, measured targets, and journaling more than “another BUY alert.”
- Win rate without risk is misleading — expectancy and R-multiples show up in serious pattern guides.
- Trust = explainable rules — pure LLM “looks bullish” scores poorly next to tools that show why and what invalidates.
- Calculators feel controllable — position size and risk % are auditable; black-box signals feel like tip services.
That is why risk calculators, journals, and backtest helpers often earn steadier ratings than high-claim signal bots with thin samples.
How PatternForge responds (by design)
| Trader complaint about signal apps | PatternForge response |
|---|---|
| “Always gives a side” | HOLD when expectancy is not positive |
| “No idea how SL/TP were chosen” | Same risk geometry for live card and historical replay (stopFactor, min stop %, R multiple) |
| “Can’t measure if I am profitable” | Journal in R, heatmap, Win/Loss, CSV |
| “Size is guesswork” | Risk Calc (balance × risk % ÷ distance × pip/point value) |
| “AI changed its mind with no audit” | Quant path is deterministic on given bars; AI remarks are optional BYOK |
| “Only works on TradingView” | Multi-broker content scripts + manual pin |
So PatternForge is positioned less as “trust our signal” and more as calculator-grade discipline around pattern research.
Multi-Broker Chart Patterns vs TradingView-Only AI Overlays
TradingView-only AI overlays (dominant SERP cluster)
Strengths
- Deep UI integration on the chart most people already use
- Easy one-click “analyze this screen”
- Rich ecosystem (Pine, alerts, optimizers)
Weaknesses for many active traders
- Analysis lives on TV; execution is on FXCM, Forex.com, MT5, OANDA, Binance…
- Copying levels between tabs introduces error
- Canvas / white-label broker terminals never get the overlay
- Model quality varies; few expose historical barrier statistics under your SL/TP rules
Examples in this cluster: ChartMentor, Crypto SignAI, ChartLense (TV-first), many “AI Elliott / Signal Pro” listings.
Multi-broker / any-chart tools
Strengths
- ChartRead / ChartSutra-style image or any-page analysis follows the trader to the broker
- Less tab friction for people who never leave the broker WebTrader
Weaknesses
- Often vision/LLM first → hard to verify reproducibility
- Weaker journaling + risk math in the same product
- Scan limits / Pro tiers without a local quantitative core
PatternForge’s multi-broker angle
| Dimension | TradingView-only AI | Image/any-chart AI | PatternForge |
|---|---|---|---|
| Primary surface | tradingview.com | Screenshot / many sites | Broker + TV URL list |
| Engine | LLM / vision | Vision / LLM | OHLCV shape + replay (+ optional AI) |
| Canvas brokers (MT5, some WebTraders) | Weak | Capture may work | Manual symbol pin |
| Process tools | Usually thin | Thin | Journal, sizer, alerts, scanner |
| “No trade” outcome | Uncommon | Uncommon | HOLD / NO DATA explicit |
Practical takeaway for ExtensionHub positioning
- vs TV AI chat: “Works on the broker chart you actually trade; quant first, AI optional.”
- vs screenshot pattern AI: “Uses price history and the same SL/TP rules on past analogues—not only a picture label.”
- vs calculators alone: “Finds candidate patterns and sizes/journals them.”
- vs optimizers: “Not a Pine optimizer; different job (live analogue research).”
One-line strategic summary
PatternForge AI sits in a sparse cell of the Chrome trading-extension market: multi-broker chart presence + historical pattern correlation with barrier-tested R geometry + journal/risk tooling—while most Store/SERP winners are either TradingView-only LLM overlays or pure backtest/execution utilities.
Step-by-Step Quick Start for Chart Pattern Signals
1. Install and grant site access
- Install PatternForge AI from the Chrome Web Store (linked from ExtensionHub.app).
- Open a chart on a supported platform — TradingView, MetaTrader 4/5 web, OANDA, FXCM, Forex.com, IG, Binance, Coinbase, Kraken, or any of the 60+ domains listed under Supported Platforms.
- Chrome may ask you to confirm site access the first time. If the overlay doesn't appear, click the toolbar icon and use the "Enable on this site" button in the popup — PatternForge never requests access silently.
2. Confirm the overlay is reading your chart
- Within a few seconds, a floating signal card should appear on the chart (default position: top right).
- If the card shows "No symbol detected" instead (common on canvas-based terminals like MT5 web or some broker WebTraders), follow the Manual Symbol Pinning steps below — type the symbol, pick the timeframe, click Start.
EURUSD, XAUUSD, or US30 — use your broker's own code, not prose like "gold" or "the S&P". Timeframe labels use M5, M15, H1, H4, D1.3. Read your first signal card
- Check the source badge first — LIVE means a confirmed live quote, SIM means a fallback estimate.
- Note the direction (BUY/SELL/HOLD), Entry/SL/TP, and the Confidence, Pattern Correlation, and Historical Win Rate percentages.
- If the card reads HOLD, that's a valid result — it means the engine could not find a historically profitable side, not that something is broken.
4. Open the Command Center for the full toolset
- Click Open Command Center in the popup (or Ctrl/Cmd+Shift+F → Command Center) to reach all 8 tabs: Signals, AI Analysis, Journal, Scanner, Alerts, Risk Calculator, API Keys, Settings.
- You don't need any API key to start — Yahoo Finance history is enabled keyless by default and covers major indices and gold out of the box.
5. Optional setup for a smoother workflow
- API Keys tab — add a market-data key (Twelve Data, EODHD, etc.) for wider symbol coverage, and/or an AI key (BYOK) for written analysis.
- Settings tab — set your Scanner watchlist, adjust Min Confidence to display, and review the default risk-gate limits (3R daily loss, 3 open trades, 6 trades/day).
- Risk Calculator tab — confirm your Pip/Point Value matches the instrument you trade before sizing your first position.
6. Build your track record
- Place trades manually on your broker's own platform — PatternForge never executes trades itself.
- Click ✓ Win or ✗ Loss on the signal card (or log it in the Journal tab) as each trade closes, so Win Rate and Expectancy start reflecting real results.
How a Signal Is Built
The pipeline runs entirely inside your browser — the content script reads the chart, the service worker fetches history and runs the match/replay, and the result is rendered back on the page and in the side panel.
pattern-engine.js (pfFetchHistory, pfMatchPatterns, pfReplayTrade, pfDetectCandlestick) and background.js, not from marketing copy.Every Formula, With the Real Constants
These are the exact constants shipped in the extension's engine. Defaults can be changed in Settings; the values below are what the code falls back to unless you move a slider.
1. Shape Feature Vector (Scale-Invariant Window)
Purpose: strip out absolute price so two moves at different price levels can be recognized as "the same shape."
2. Pattern Correlation — Shape Similarity %, 0–100
Matching ladder (verified in code): the engine first tries your Min Correlation setting, then relaxes step-by-step through 0.75 → 0.70 → 0.65 → 0.60 until it finds at least 5 matching windows. If it had to relax, the card shows the threshold actually used and the confidence score is scaled down proportionally (confidence × usedSim/minSim).
3. Stop-Loss Distance (Risk) price units
Every other level — stop, target, win-rate replay, position size — is built from this one number. It behaves like a lightweight ATR proxy: it widens on volatile bars and never shrinks below a floor.
stopFactor: 0.75, minStopPct: 0.004) are hard-coded defaults in pfReplayTrade() and pfMatchPatterns() and are what the Settings tab's "ATR Multiplier" and "Stop Loss Type" fields feed into.4. Entry, Stop-Loss & Take-Profit price units, symbol's own decimals
5. Historical Win Rate — Bar-by-Bar Trade Replay %, resolved trades only
Both BUY and SELL are replayed on every matched analogue; whichever side shows positive expectancy (mean R per trade, including timeouts) becomes the card's direction. If neither side clears at least 3 resolved trades, the card shows HOLD rather than a number built on too little evidence.
6. Confidence Score %, 0–100
7. Multi-Timeframe Confluence %, 0–100
8. Position Size lots — Risk Calculator tab
9. Journal Win Rate & Expectancy %, and R-multiple
Worked example: 20 closed trades, 9 wins averaging +1.8R, 11 losses averaging −1.0R → Win Rate = 45%, Expectancy = (0.45 × 1.8) − (0.55 × 1.0) = 0.81 − 0.55 = +0.26R per trade — profitable on average even below a 50% win rate, because winners are bigger than losers.
Named Candlestick Formations (Deterministic, Rule-Based)
Computed from the last 1–3 bars with fixed geometric rules — no model, no screenshot. Strength is 1–3 and ranks how strict the geometric match was; it is not a win probability.
| Formation | Bars | Bias | Strength | What it means |
|---|---|---|---|---|
| Morning Star | 3 | BUY | 3 | Selling stalled on a small middle bar, then buyers took back most of the drop. |
| Evening Star | 3 | SELL | 3 | Buying stalled on a small middle bar, then sellers gave back most of the rise. |
| Three White Soldiers | 3 | BUY | 3 | Three strong closes in a row, each above the last. |
| Three Black Crows | 3 | SELL | 3 | Three strong down closes in a row, each below the last. |
| Bullish / Bearish Engulfing | 2 | BUY / SELL | 3 | This bar's body swallows the previous bar's body — demand/supply overwhelmed the other side. |
| Bullish / Bearish Outside Bar | 2 | BUY / SELL | 2 | Took out both sides of the previous bar and closed decisively — a liquidity sweep. |
| Inside Bar (Harami) | 2 | context-dependent | 1 | Range contracted inside the previous bar — momentum paused, often before a break. |
| Dragonfly / Gravestone Doji | 1 | BUY / SELL | 2 | Hard rejection from below/above, closed back near the open. |
| Doji | 1 | HOLD | 1 | Open and close effectively equal — genuine indecision. |
| Hammer / Hanging Man | 1 | BUY / SELL | 2 | Long lower wick: sellers pushed down and were rejected before the close. |
| Shooting Star / Inverted Hammer | 1 | SELL / BUY | 2 | Long upper wick: buyers pushed up and were rejected before the close. |
| Bullish / Bearish Marubozu | 1 | BUY / SELL | 2 | Almost pure body, no wicks — one side controlled the entire bar. |
HOLD and NO SIGNAL
There is no fixed number of “HOLD / NO SIGNAL” refreshes after which you should trade. PatternForge is not on a timer that “owes” you a BUY or SELL.
What HOLD and NO SIGNAL mean
| Message | Meaning | What usually changes it |
|---|---|---|
| HOLD | History was matched, but barrier replay did not show positive expectancy for BUY or SELL at the current levels | A new bar (or a real shape change), different timeframe, or different risk settings |
| NO SIGNAL / NO DATA | Symbol, history, or quotes are missing or failed | Pin symbol, fix API/provider, wait for quotes, open a real chart |
The overlay may re-check about every 5 seconds (default poll), but that often recomputes the same HOLD if the last N candles have not changed in a meaningful way.
When a new directional signal can appear
Think in bars, not in “times the card blinked”:
New candle close on that timeframe
- M15 → often after the next 15‑minute close
- H1 → after the next hourly close
- H4 / D1 → much less often
Shape actually changes
The engine uses a window of recent bars (often ~20–30). Until those closes shift, correlation and replay stay similar → HOLD can persist for many polls.You change inputs
Timeframe, symbol pin, lookback / min correlation / R:R in Settings, or a fresh history fetch after cache age.
So: seeing HOLD 10 or 100 times on the same H1 bar is normal. Waiting for “enough HOLDs” is not a valid entry rule.
Practical rule for trading
- Treat HOLD as “stand aside”, not “almost ready.”
- Prefer acting when the card flips to BUY or SELL after a bar close, with confidence above your minimum and risk sized in the Risk Calc.
- If it stays HOLD for many bars in a row, the market shape simply has no edge under the engine’s stop/target rules—forcing a trade is the opposite of what the math is for.
- For NO DATA, fix detection (manual pin on Forex.com / MT5, enable quotes) instead of waiting.
Bottom line: Do not count HOLD refreshes. Wait for a new bar and a real BUY/SELL (or change timeframe/symbol). HOLD can last for the whole life of the current candle window—and that is intentional.
The 8 Command Center Tabs
Opened from the popup's "Open Command Center" button, or by right-clicking the toolbar icon → side panel.
1. Signals Free
Live signal cards for the chart you have open: direction, source badge (LIVE vs SIM), symbol/timeframe/platform, Entry/SL/TP, R:R, Pattern Correlation %, Historical Win Rate %, plus a Multi-Timeframe Confluence meter across M15/H1/H4. Filter by timeframe, direction (BUY/SELL), or account tag. Click ✓ Win or ✗ Loss on a card to push it straight into the Journal.
2. AI Analysis Bring your own key
Connect your own OpenAI, Anthropic Claude, Google Gemini, DeepSeek, Moonshot Kimi, Alibaba Qwen, Z.AI GLM, or Morph key (or run fully offline with a local model). The AI writes a plain-language read of the setup — recommendation, reasoning, risks, and an invalidation condition — composed entirely from the signal's own numbers. Your key is stored only in chrome.storage.local and used directly by the service worker; chart pages never see it. Free tier: 5 AI analyses/day; Pro and Pro+ remove the cap.
3. Journal Free core CSV import is Pro
Logs every trade you mark Win/Loss with entry, SL, TP, and the R-multiple achieved. Shows Total, Wins, Losses, Win Rate, and Expectancy, plus a GitHub-style 12-week heat-map (green = more wins that day, red = more losses, darker = more trades). Export to CSV any time; CSV import and account filtering for multi-account tracking are available on Free, import specifically is Pro+.
4. Scanner Free
Runs the same historical pattern engine across your entire watchlist (set in Settings, comma-separated symbols) at a chosen timeframe (M15/H1/H4/D1) and minimum confidence, then ranks results by R per trade (expectancy) rather than win rate alone.
5. Alerts Free: 3 active Pro: unlimited
Set an Above/Below price trigger per symbol, optionally tagged to an account. Checked roughly once a minute against live quotes (Yahoo or your configured market-data key), with a browser notification on hit.
6. Risk Calculator Free
Account Balance, Risk % (slider, 0.1–5%), Entry, Stop-Loss, optional Take-Profit, and Pip/Point Value → Risk Amount, Position Size, R:R Ratio, Potential Profit, Potential Loss. Uses the exact position-size formula above.
7. API Keys Free tier works keyless via Yahoo
Connect market-data, AI, and notification providers. See the full provider tables in the Supported Platforms & Providers section below. Includes a cost estimator for AI usage and per-provider rate-limit status.
8. Settings Trading Profile presets are Pro
Signal Configuration (Pattern Lookback 5–100 candles, default 20; Min Pattern Correlation 50–95%, default 75%; Signal Sensitivity 1–10, default 5; Min Confidence to display, default 70%), Risk Management (Stop Loss Type: ATR-Based / Fixed Pips / Pattern-Based / Percentage; ATR Multiplier 1–3, default 1.5; Take-Profit R:R 1–5, default 1:2; Max Spread, default 5 pips), Notifications (browser/sound/voice, volume), Chart Overlay (Compact 160px / Normal 200px / Wide 280px; position; refresh interval 2–30s, default 5s; hide HOLD; show-only-LIVE), Appearance (colorblind mode, high-contrast mode, UI scale 75–150%), Workflow (copy format: plain/Telegram-Discord markdown/CSV; auto-AI on new signals — Pro; scanner watchlist), Accounts/Tags (add/manage multiple accounts), and Data & Privacy (offline mode toggle, export backup JSON, export signals CSV, import backup, delete all data). One-click Scalp/Day/Swing profile presets (Pro) apply lookback, correlation, sensitivity, and R:R together.
The Popup (Toolbar Icon)
Click the extension icon for a quick-glance view without opening the full Command Center: today's signal count, win rate, open trades, an active-session strip (Asia/London/NY, UTC clock), the current chart's live signal card, and buttons for Open Command Center, Settings, ⏸ Pause (halts new signals without losing history), Journal, and API Keys. If the current site needs a one-time permission grant, an "Enable on this site" button appears; PatternForge never auto-requests access to a domain you haven't visited. Pro/Pro+ license activation (Gumroad key) and status also live here.
Keyboard Shortcuts
Global (Chrome-level, work anywhere)
| Shortcut | Action |
|---|---|
| Ctrl/Cmd+Shift+F | Open the PatternForge AI popup |
| Ctrl/Cmd+Shift+H | Toggle the signal overlay on/off |
| Ctrl/Cmd+Shift+S | Check the current signal |
| Ctrl/Cmd+Shift+X | Emergency pause — halts all signals immediately |
Reassign any of these at chrome://extensions/shortcuts.
On-chart (single key, while the overlay has focus and you're not typing in a field)
| Key | Action |
|---|---|
| M | Minimize / expand the overlay card |
| C | Copy the current signal as text |
| A | Open AI Analysis in the Command Center |
| R | Remarks — written BUY/SELL/HOLD summary |
| S | Share the setup (downloadable share card) |
| P | Position sizer panel |
| L | Lock/unlock the card's position on screen |
| E | Economic calendar |
| O | Strategy finder (walk-forward parameter optimizer) |
| T | Market sessions clock |
| D | Your performance stats |
| W | Watchlist scan panel |
| N | Order-ticket / alert-push helper for the current signal |
| ? | Card options & the full shortcut list |
| Esc | Close the topmost open panel |
The card can also be dragged from anywhere on it (buttons still click normally), and it shows the detected candlestick formation, a bar-close countdown, and a live session clock.
Manual Symbol Pinning
Most charts expose their symbol and timeframe in the page DOM and PatternForge reads them automatically. Some platforms — canvas-rendered MetaTrader 5 web, parts of Forex.com WebTrader, and similar terminals — paint the whole chart into one canvas with no readable symbol node. For these, pin the symbol manually:
- Open the chart. If auto-detection fails, the card shows "No symbol detected" with a pin form.
- Type the instrument exactly as your data provider knows it (e.g.
EURUSD,XAUUSD,US30, or a provider-specific form likeEURUSD.pro). - Select the timeframe that matches the candles on screen.
- Click Start on this symbol (or press Enter).
- The pin is saved per hostname and reused automatically on your next visit.
- Click Unpin on the card any time to return to auto-detect.
EURUSD (no slash), Alpha Vantage/some providers expect a split form. Enter the symbol as it appears on your chart; the extension converts formats internally where it can.Daily Risk Gate (Overtrading Brake)
Press ? on any chart to open card options, including risk limits that default to:
When you approach or cross one of these self-set limits, a warning panel appears before you can log another trade. It can be overridden (this is a personal guard-rail, not a broker-enforced lock), but the override is recorded.
Supported Platforms & Data Providers
Trading & charting platforms
PatternForge AI is permitted to run on 60+ broker, exchange, and charting domains, including TradingView, MetaTrader 4/5 web terminals (MQL5, MetaTrader Web), OANDA, FXCM, IG, Forex.com, Interactive Brokers, Charles Schwab / thinkorswim, TD Ameritrade, Webull, Robinhood, eToro, Plus500, CMC Markets, Pepperstone, IC Markets, Eightcap, Fusion Markets, Saxo, XTB, Admiral Markets, Tickmill, FBS, XM, HFM, AvaTrade, Libertex, Capital.com, Trading 212, Revolut, Binance, Binance.US, Coinbase, Kraken, Bybit, KuCoin, Bitfinex, Bitstamp, Gate.io, OKX, MEXC, Dukascopy, Yahoo Finance, and charting tools like ExoCharts, NinjaTrader, Sierra Chart, MotiveWave, TrendSpider, TrendLyne, and ChartIQ.
Market data (for pattern history)
| Provider | Free tier | Notes |
|---|---|---|
| Yahoo Finance | Keyless | Enabled by default; covers major indices, gold, and many symbols with no signup. |
| Binance | 1,200 req/min, keyless | Crypto pairs. |
| Twelve Data | 800 requests/day | Broad forex/stock/index coverage. |
| Finnhub | 60 requests/min | Live quotes. |
| Alpha Vantage | 25 requests/day | Forex + equities. |
| EODHD | 20 requests/day | End-of-day + intraday. |
| Massive / Polygon.io | 5 requests/min | US equities/indices aggregates. |
| Financial Modeling Prep | 250 requests/day | Optional. |
| Marketstack | 100 requests/month | Optional. |
| Tiingo | 1,000 requests/day | Optional. |
| Alpaca | 200 requests/min | Optional; also supports Pro+ broker trade-history sync. |
| CoinGecko | 10,000 requests/month | Crypto reference data. |
| Databento | $125 in starting credits | Optional, paid-tier data. |
AI providers (bring your own key)
| Provider | Approx. cost |
|---|---|
| DeepSeek | ~$0.14 / 1M tokens (cheapest) |
| Morph | ~$0.14 / 1M tokens |
| Google Gemini Flash | ~$0.25 / 1M tokens |
| OpenAI GPT-4o Mini | ~$0.50 / 1M tokens |
| Alibaba Qwen | ~$0.50 / 1M tokens |
| Moonshot Kimi | ~$0.95 / 1M tokens |
| Anthropic Claude Haiku | ~$1.00 / 1M tokens |
| Z.AI GLM | Free tier available |
Notification webhooks: Telegram (bot token + chat ID), Discord (webhook URL), and Slack (webhook URL) are all optional and configured in API Keys.
Free, Pro, and Pro+
| Limit / feature | Free | Pro ($9/mo) | Pro+ ($79/yr) |
|---|---|---|---|
| Stored signals | 20 | 200 | 500 |
| Accounts/tags | 1 | 50 | 100 |
| Active price alerts | 3 | Unlimited | Unlimited |
| Watchlist size | 5 | 50 | 100 |
| AI analyses / day | 5 | Unlimited | Unlimited |
| Trading Profile presets | — | Yes | Yes |
| Auto-AI on new signals | — | Yes | Yes |
| Journal CSV import | — | Yes | Yes |
| Unlimited history depth | — | Yes | Yes |
| Broker trade-history sync | — | — | Yes |
| Trade-log API endpoint | — | — | Yes |
Activate with a Gumroad license key in the popup or Settings tab. Confirm current pricing on the live Gumroad listing, since in-product pricing text can lag a price change.
Worked Example — EUR/USD H1
H1 signal bar: Entry (close) = 1.08500, High = 1.08640, Low = 1.08420.
If 12 matched past analogues replay this exact setup and 7 hit target before stop, 4 hit stop first, and 1 times out: Historical Win Rate = 7/(7+4) × 100 = 64%, based on 11 resolved trades. With avgSimilarity 88% and 12 matches out of a topN of 20 (sampleFactor 0.60): Confidence = round(88×0.40 + 64×0.40 + 60×0.20) = round(35.2+25.6+12) = ≈73%.
How Far Back Does PatternForge AI Check Chart Patterns?
There is no single fixed number of years. How far back PatternForge looks depends on timeframe, data provider, and a bar limit—not a hard “always 10 years” rule.
What the engine requests
| Setting | Typical value | Meaning |
|---|---|---|
| History request size | Up to about 5,000 bars | Cap on how many candles it tries to load |
| Match window | Often ~20–30 bars | Shape of the recent pattern |
| Forward horizon | Often ~5–20 bars | How far each past match is replayed |
| Cache | About 12 hours | Same symbol/TF may reuse stored bars |
Matching only uses history that was successfully downloaded for that symbol and timeframe.
Rough calendar depth by timeframe
Same 5,000 bars is a different number of years on different TFs:
| Timeframe | If you have ~5,000 bars | Ballpark span |
|---|---|---|
| M15 | 5,000 × 15 min | ~1.5–2 months of sessions (not years) |
| H1 | 5,000 hours | ~7–8 months of 24h data (less if market closed) |
| H4 | 5,000 × 4h | ~2–3 years order of magnitude |
| D1 | 5,000 days | up to ~13–20 years if the provider allows |
So: intraday = months of history, daily = multi‑year when the API allows it.
Provider limits (what you actually get)
| Source | Typical depth |
|---|---|
| Binance | About 1,000 candles max per request |
| Yahoo (keyless path) | Often ~60 days on low TFs, ~2 years on H1-style, max on daily |
| Twelve Data / EODHD / Polygon-style | Depends on plan; engine asks for large outputsize / multi-year ranges on daily |
| Alpha Vantage | “Full” daily series when available; intraday is thinner |
If the provider returns only 200 bars, the engine can only search those bars—even if another provider could go deeper.
What you see on the card
When a match works, coverage is reflected in fields like coverage start / end and match dates/years on analogues. That is the real lookback for that run—not a marketing “always 20 years.”
Practical answer
- M15 / M5: weeks to a couple of months of history, not multi‑year.
- H1: often months up to ~1–2 years depending on provider.
- H4 / D1: can reach several years if the data source returns a long daily/H4 series.
To go deeper, use a timeframe that stores more calendar time per bar (H4/D1) and enable a market-data provider with a larger history quota—not more poll refreshes on M15.
Common Mistakes to Avoid
- Ignoring HOLD. HOLD means either not enough matched history existed to test a direction, or both BUY and SELL lost money on replay — forcing a trade re-introduces the bias the math already rejected.
- Mixing pip-value currencies. If your balance is in EUR but Pip Value is entered as USD-per-lot, the position size is wrong. Keep one currency consistent in the sizer.
- Pinning the wrong symbol code. Mismatched provider symbols return no history or NO_DATA — check the code your enabled provider actually uses.
- Treating high confidence as a Kelly-style bet size. Confidence is not a position-sizing input — keep risk % independent and small regardless of the score shown.
- Comparing win rates across different R multiples. Changing the Take-Profit R:R ratio changes what "win" means in the replay — compare like-for-like settings only.
- Forgetting the emergency pause is on. Ctrl+Shift+X or the popup's Pause button blocks new signals — check the status dot before assuming nothing is happening.
- Expecting a signal on every chart. The engine needs enough bars of history (roughly 500+, and specifically window+horizon+60 bars for the default 30-bar window) to have anything to match against.
Frequently Asked Questions
Does PatternForge AI place trades automatically?
What does "SIM" vs "LIVE" mean on a signal card?
Why is Historical Win Rate sometimes blank?
Why does the same symbol sometimes show HOLD instead of BUY or SELL?
Do I need an API key to use PatternForge AI?
Is my API key sent to a PatternForge server?
chrome.storage.local) and are used directly by the extension's own service worker to call the provider you configured. The chart page itself never receives your keys.What happens on a platform I use that isn't in the supported list?
How do I activate Pro/Pro+ after purchasing?
Permissions, Privacy & Disclaimer
Why each permission is requested
chrome.storage.local. Nothing leaves your browser except calls to the third-party market-data, AI, or webhook providers you explicitly enable, and chart pages never receive your API keys. Uninstalling the extension removes its local storage per Chrome's normal behavior. Export or delete your data any time from Settings → Data & Privacy.
