Pipsgrowth EX12015 MultiIndicatorConfluence

Pipsgrowth EX12015 is not a trading EA. It is an AI Market Analyzer dashboard that ships market context to three large-language-model providers in parallel, parses their structured replies, and renders the verdict on a 550x750-pixel chart panel so a human trader can decide what to do. There is no order open, no SL/TP submit, no magic-number account management — every function in the source either prepares data, makes an HTTPS call to an AI endpoint, or draws a label on the chart. The three TryClose/TryModify stubs that exist in the file are dead code carried from the broader MultiIndicatorConfluence family; nothing in OnTick or PerformAIAnalysis() calls them.

The data pipeline starts in PrepareMarketData(). It assembles a single text payload containing the current symbol, bid/ask, spread in pips, and the active chart timeframe, then appends up to CandlesToAnalyze (default 30, hard-capped at 200) OHLCV bars formatted as YYYY-MM-DD HH:MM, open, high, low, close, volume. If SendIndicatorData is true it then layers in the current and two prior values of each enabled indicator: EMA(9) and EMA(21) with a BULLISH/BEARISH cross label, RSI(14) with OVERBOUGHT/OVERSOLD/NEUTRAL flag at 70/30, MACD(12,26,9) main and signal lines, ATR(14) plus a pips conversion, and Bollinger Bands(20, 2.0) upper/middle/lower with a price-position tag (ABOVE UPPER / UPPER HALF / MIDDLE / LOWER HALF / BELOW LOWER). It finishes with a price-action summary giving the last-candle, 5-candle, and 10-candle pip deltas. All seven indicator toggles (Include_EMA, Include_RSI, Include_MACD, Include_ATR, Include_BB) are independent inputs, so the prompt can be trimmed for cost.

The calendar leg is the second input vector. FetchForexFactoryCalendar() GETs https://nfs.faireconomy.media/`ff_calendar_thisweek`.xml every CalendarMinHoursBetweenFetch hours (clamped 1–24). The strict cooldown is enforced by a two-tier check — one on the last successful fetch, one on the last attempted fetch — so the EA will never re-hit the URL inside the cooldown window even if a fetch failed. When called inside the window it returns the cached string from CalendarDataCache instead of calling out. ParseCalendarXML() walks the XML event tags, filters by ShowLow/Medium/High Impact and Holiday toggles, and reformats each event as [Impact] Country DateTime - Title (F: forecast) (P: previous), then splits the result into TODAY and UPCOMING (next 48h) sections.

The three AI endpoints are called sequentially from PerformAIAnalysis() with the prepared market payload (and calendar payload, for Perplexity and OpenAI). UsePerplexity, UseDeepSeek, and UseOpenAI are independent bools; at least one must be true or OnInit returns INIT_FAILED. Each call follows the same shape: build a prompt requesting the model to populate a TREND_TABLE with BULLISH/BEARISH/NEUTRAL plus 0–100% confidence for the 1M / 5M / 15M / 1H horizons, then SIGNAL: LONG/SHORT/WAIT, ENTRY: price, SL: pips, TP: price, CONFIDENCE: X/10. Perplexity is additionally prompted to web-search for breaking news and central-bank statements. SendHTTPRequestWithHeaders() POSTs the JSON with a 30s timeout; if the response contains "error" the function pulls the message field and returns API Error: <message>. Reasoning-model leakage (<think>...</think> blocks) is stripped by the CallPerplexityAI wrapper before the response is stored.

ParseAIResponse() then runs over the cleaned text. It does flexible substring matching — if a line contains 1M: and : and the m1 slot is empty, it captures the line — so the parser tolerates slight format drift from the models. The 3-way consensus block in UpdateDashboard() is the visual centerpiece: it counts LONG/SHORT/WAIT tags across the three providers and prints ★★★ UNANIMOUS (3/3) ★★★ in lime when all three agree and the average confidence is at least 8, ★★ STRONG CONSENSUS (3/3) ★★ in green for confidence 6–7, ★ WEAK CONSENSUS (3/3) ★ in yellow for confidence below 6, MAJORITY CONSENSUS (2/3) for a 2-vs-1 split, NO CLEAR SIGNAL (Wait) when 2+ models returned WAIT, and SPLIT DECISION (1-1-1) otherwise. The per-timeframe mini-table (1M / 5M / 15M / 1H rows × PERPLEXITY / DEEPSEEK / GPT-5 columns) lets you eyeball agreement horizon by horizon.

The dashboard itself is a dark-slate-gray rectangle (color clrDarkSlateGray, 550×750 by default) anchored at Dashboard_X=20 / Dashboard_Y=30, with a green ANALYZE button (clrGreen / clrDarkGreen border) at the top-right. Every OnTick, UpdateDashboard() re-renders up to 100 OBJ_LABEL objects named AI_Label_0 … AI_Label_99, refreshing the bid/ask/spread/indicator lines. There is a hard 60-second gate between manual analyses (TimeCurrent() - LastAnalysisTime < 60 short-circuits with UpdateStatusLabel("Wait 60s between analyses")), and an optional auto-refresh path that fires PerformAIAnalysis() every RefreshRateSeconds seconds when that input is non-zero. Because the EA only calls chart-object and WebRequest APIs, it has no concept of trading session or weekend — it works on every symbol the broker exposes, including XAUUSD on M5 and H1.

The hard prerequisite that breaks this EA more often than any other setting is the WebRequest whitelist. OnInit actually probes https://api.openai.com and prints the precise error code: error 4060 means WebRequest is disabled in Tools > Options > Expert Advisors; error 4002 means the URL is not on the allow list. The three URLs that must be whitelisted are https://api.perplexity.ai, https://api.deepseek.com, https://nfs.faireconomy.media, and https://api.openai.com. The Input string slots hold sample API keys; replacing them with real keys from your own Perplexity / DeepSeek / OpenAI account is mandatory. API costs are metered by token count and the prompts run up to 1500 max_tokens at temperature 0.3, so a 3-model analysis with calendar data will typically cost a few cents per call — running auto-refresh at sub-minute intervals can burn real money fast. Treat RefreshRateSeconds=0 (manual click only) as the default mode, and only enable auto-refresh with a budget you have set in advance.

In practice the workflow is: drop the EA on the chart, set the three API keys, leave RefreshRateSeconds at 0, click ANALYZE after a London or New York open. Read the consensus block first — UNANIMOUS LONG at confidence 8 or higher is the only signal that warrants acting on its own; everything below that is a context read. Use the indicator readout to sanity-check what the models are saying (if the dashboard says EMA 9 > 21 BULLISH and the models agree, the alignment is real; if the dashboard says EMA 9 < 21 BEARISH and the models call LONG, treat that as a contradiction and click ANALYZE again in 60 seconds rather than trade). The CALENDAR block at the bottom of the panel flags any high-impact event in red within the next 48 hours — pause all discretionary trades during the [High] lines and let the post-news volatility settle before re-querying. The structured SL/TP numbers the AIs return should be treated as the model's first opinion, not a committed stop — apply your own position sizing and your broker's minimum-stop distance before any order.