PatternForge AI — Chart Pattern Signals & Trade Journal

PatternForge AI — Historical Chart Pattern Signals & Trade Journal PatternForge AI is a chart research extension for traders who work on web-based charts. It answers one narrow, measurable question: when the current price shape appeared on this instrument and timeframe before, what actually happened next? Every BUY, SELL, HOLD, or NO SIGNAL you see comes from comparing the current price pattern against historical bars and replaying what followed. There is no fallback estimate, no seeded score, and no forced call. If the engine cannot test a direction from history, it shows HOLD — deliberately. This is analysis and planning software. It is not a broker, does not connect to your broker for trading, and never places, modifies, or closes orders. WHAT IT DOES • Detects the symbol and timeframe of the chart you have open, or lets you pin them manually when the page does not expose them. • Fetches historical bars from the market-data source you choose (or a keyless source where available). • Matches the current 30-bar shape against every earlier, non-overlapping window in that history. • Replays each historical match bar-by-bar with the same stop and target the card shows — a stop at one risk unit, a target at the configured R multiple (default 2R). • Decides BUY or SELL only when the replay says one side paid; otherwise returns HOLD. • Shows entry, stop-loss, take-profit, first target (1R), second target, similarity, historical win rate, confidence, and an optional pair-quality score. • Shows three plan boxes (Free, Pro, Pro+) pinned under the card header, in the popup and in the Command Center; the box for your current plan shows a blinking "Active". • Names the most recent candlestick formation when it recognises one, and flags disagreement with the quant direction (Pro). • Provides a position sizer with per-instrument contract metadata (forex, metals, indices, crypto) that converts to your account currency — or withholds the lot size when it has no exchange rate — plus a daily risk gate, pre-trade checklist, and post-trade review tags. • Keeps a local trade journal that closes its own trades when price reaches the stop or target, and reports expectancy, profit factor, drawdown in R, streaks, and breakdowns by session, weekday, symbol, direction, and timeframe (journal analytics: Pro). A printable weekly report opens in its own tab. • Runs a walk-forward parameter sweep on demand (Pro), using only bars available at each historical anchor — no lookahead. • Provides a keyless economic calendar, a session clock, a watchlist scanner, and price alerts. • Includes AlamToolKit, a built-in directory of calculators and tools from AlamToolKit.com, with Trading Calculators listed first. Every number on the card is either measured from history or labelled as an estimate. WHAT IT DOES NOT DO • It does not place, modify, or close broker orders. The extension is read-only and blocks outgoing requests that resemble order placement — including via a custom endpoint. • It does not guarantee profit. Historical comparisons describe what happened; they do not predict what will. • It does not invent price levels. If no live quote is available, no levels are shown, and the card tells you why. • It does not force a call every candle. HOLD is a designed outcome, not a failure. • It does not send your data to the publisher. There is no PatternForge backend. HOW THE ENGINE WORKS 1. History is fetched from the first enabled provider that supports the instrument class, requiring at least 120 bars before continuing (5,000 preferred). 2. The most recent 30-bar window is z-normalised, so the shape can be compared fairly regardless of price level or scale. 3. That shape is compared against every earlier, non-overlapping window. The similarity threshold is relaxed step-by-step until at least five matches are found, and whichever threshold was actually used is reported on the card. 4. Every match is replayed forward, bar by bar, under the same stop and target the card draws. If a single bar's range spans both the stop and the target, it is counted as a loss — the conservative reading, because OHLC cannot say which came first. 5. Each match is replayed as a long and as a short. Whichever side had positive expectancy is used. If neither did, HOLD is returned. If fewer than three analogues resolved to a stop or target, NO SIGNAL is returned. 6. Confidence blends average similarity (40%), win rate (40%), and sample size (20%), and is reduced when the similarity threshold had to be relaxed. When there is no directional call, it rests on similarity and sample size only. The optional AI layer never overrides any of this. It can describe a setup in plain language; it cannot move entry, stop, target, or size. KEY FEATURES Chart overlay A floating card on the chart shows the direction, levels, similarity, historical win rate, confidence, session context, news status, a recent-close sparkline, and a live bar-close countdown. Multi-timeframe confluence is a Pro feature. Drag the card anywhere; lock it in place; minimise it to a header row. Command Center A full side panel with sections for Signals, Written Analysis, Journal, Scanner, Alerts, Risk Calculator, API Keys, and Settings. Includes an emergency pause and a hotkey panel. Journal that closes itself Log a trade and the background service worker watches price until the stop or target is reached — even if the chart tab is closed. The stop is checked before the target; a bar that could plausibly have touched both is scored as the worse outcome. Risk tools A position sizer using per-instrument contract metadata (not a single guessed contract size), a daily loss cap in R, max-open and max-trades-per-day caps, directional exposure warnings across pairs, a high-impact news window, and prop-firm-style presets. Limits warn at log time and can be overridden; the override is written to the journal entry. Walk-forward strategy finder (Pro) Tests parameter combinations by m…