Market Intelligence by @CoachZubair

The numbers on this dashboard come from a deterministic walk through SignalsHistory.csv, the audit log we append to on every run. Below is exactly how each metric is computed so you can decide what weight to give it.

What counts as a signal

• Entry: the first day a ticker enters a Buy or Sell streak (verdict on a given horizon).
• Entry price: the most recent settled daily close at the time the signal fired — not the intraday live tick we displayed on the dashboard. Stocks/ETFs/commodities use the prior session's Yahoo close; crypto uses the prior UTC daily close (CoinGecko's history.at(-2)). Both /swing.html and this page now use the same basis (BT-17, 2026-05-28): finalize.js writes settled_close from fetch_prices.js into SignalsHistory.csv, and compute_indicators.js reads history.at(-1). Using the live tick had a small upward bias on win rate (you would tend to enter near intraday lows but exit on EOD closes); removing it makes the two pages agree on the same trade.
• Forward walk: we scan settled daily closes after the entry day until something resolves the trade.

What counts as an exit

• TP1 Hit: any forward bar closes at or above TP1 (for Buy) or at or below TP1 (for Sell). TP1 is treated as a full exit — TP2 is not separately tested here. This keeps the Performance tab and the swing-performance tab consistent on what a "win" means for the same trade.
• Stopped: any forward bar closes at or below the original stop (for Buy) or at or above it (for Sell). For strategies with a trailing stop, this is the breach of the original below-entry stop — exits on a ratcheted-up stop above entry get the "Trailed" label below.
• Trailed: (strategies with trailing stops only — e.g. FM Inspired) the active stop ratcheted ABOVE entry as the trade moved up, then a forward bar breached it. Booked as a winning exit because by construction the stop has cleared the entry price.
• Time-stop: 15 trading days have elapsed since entry with neither TP1 nor stop hit. The trade is force-exited at that bar's close. This catches level-less trades (universe candidates without a watchlist TP/stop) and slow-moving signals that would otherwise sit indefinitely in "Open."
• Exited: the verdict changes (e.g. Buy → Hold) before TP1 or stop is touched. We book the exit at the first non-streak day's close, since that is the earliest day you could have known to exit. Exited trades are excluded from the win-rate denominator.
• Open: still in a Buy/Sell streak with fewer than 15 forward bars accumulated. Return is shown for context but excluded from the win-rate denominator.

Exit price

When TP1 or stop is hit, the exit is booked at the breaching bar's close, not at the TP1/stop level itself. Booking at the level always favours the win column (the bar that breaches typically closes past it). Using the bar's close removes that one-sided bias and reflects what you would actually realise.

A slippage haircut is then applied to every closed exit. The magnitude is asset-class-aware: 50 bps for crypto (mid of major-exchange retail spot — Coinbase ~149 bps round-trip, Kraken ~52 bps maker / 80 bps taker; single-side ~50 bps is a defensible "what a careful retail trader pays"), and 10 bps for stocks / ETFs / commodities (commission-free US equities with minimal spread). Buy exits get slightly less, Sell exits pay slightly more. Open positions are mark-to-market and not slipped. The knob keeps a "+0.1%" trade from being read as a win when it would actually be a wash after costs — and the per-class calibration prevents crypto win-rates from systematically overstating by ~140 bps per trade compared to equities.

What about "Hold" verdicts?

Hold signals are not tracked in the win-rate table. If the system says Hold and the price subsequently falls 50%, no penalty appears anywhere on this page. Holds are deliberately measurement-free — they are the system's default state ("no opinion either way"), not an active position to evaluate. If you want to assess Hold performance, look at the raw per-ticker price column on the dashboard, not the win-rate metric here. This is a known accountability gap; the alternative (treating Hold as a continuous short or long) introduces its own distortions, so the current honest framing is "we only score directional calls."

Sample-size flag

Months and tickers with fewer than 5 closed signals are dimmed and tagged "n=<5". A 1-signal ticker showing 100% win rate is noise; treat it that way.

Analyst direction-correctness

Per-firm "% correct" only counts calls made after we started recording prices (2026-05-05). Calls older than that have their "Price at Call" backfilled to the first day we ever observed — turning the metric into a window-return rather than analyst skill. The per-call table still shows every call we tracked, but the firm summary is restricted to the post-tracking window so the percentage is honest.

Swing setup scoring (used on /swing.html)

Six signals, one point each — the same set for watchlist and halal-universe tickers:

1. Uptrend — EMA50 > EMA200 and the EMA200 is rising over ~1 month (skipped if < 200 bars of history are available).
2. RSI pullback — RSI(14) < 48 (the zone where uptrends tend to find support; below 25 is vetoed as a falling knife).
3. EMA50 support — price within ±3% of the 50-day EMA, with support holding.
4. MACD improving — histogram rising 2 bars from a recently negative reading.
5. Volume healthy — the pullback is on contracting volume, or the latest up-day came on >1.2× average volume.
6. Relative strength — the ticker's 3-month return beats the SPUS (halal S&P 500) benchmark.

A ticker surfaces as a candidate when it fires at least 60% of its available signals (minimum 3) and is not vetoed. In buy zone (price inside the watchlist 1M zone) and reversal confirmed (an up-close that reclaims the EMA50) are shown as context but do not add to the score. Veto: if RSI > 70 OR Bollinger %B > 85, the row is dropped regardless of score — overbought is overbought, not a buy.

What this isn't

Forward testing & position size (since 2026-06-10)

Every strategy sleeve runs as a funded forward paper-test: a model account of $30,000 initial + $1,000 added on the 1st of each month, allocated 45% Core DCA / 20% All-Weather / 15% Quality / 10% Momentum / 5% Trend / 5% Swing. Core DCA contributions buy the 8 halal ETFs immediately, pace-weighted (Accumulate 1.5× / Normal 1× / Ease 0.5×), and are recorded as real-dollar lots that never sell. Other sleeves size positions as pool capital ÷ open positions (the per-signal tables still use a flat $1,000-per-position convention for comparability). This is not a historical backtest — signal history only begins May 2026, so each strategy must earn its track record in real time, with settled-close entries and a slippage haircut, before real capital follows it.

How verdicts are made (since 2026-06-10)

Daily 1M Buy/Hold/Sell verdicts are mechanical: trade levels are computed each morning from price structure (an ATR-sized band around the 50-day EMA, a pivot-low/ATR stop, weekly pivot-high targets), and the verdict comes from trend × zone location × market regime × risk guards (earnings within 14 days, hard negative analyst-revision shifts, falling-knife and overbought vetoes). A Buy can come from a pullback into the zone or a breakout close above the prior 20-day high; a Sell needs persistence — a settled close at least 1% below the stop, or two consecutive settled closes below both moving averages — so a single bad close can't whipsaw the call. The AI may only downgrade a verdict for a nameable news risk — never upgrade. The 12M column is a DCA pace label (Accumulate / Normal / Ease), set by how deep the asset sits in its own drawdown history — a long-term accumulation pace, not a trade call. All positions across every strategy sleeve resolve in one shared trade ledger, so this page, the swing page, the portfolio page and the P&L reports cannot disagree about the same trade. Full plain-English definitions live in the glossary.

• Not advice. These numbers describe how the dashboard's own signals played out, not how a real portfolio would perform. Fees and taxes are not modelled; slippage uses the fixed haircut above, and the funded model's per-position sizing is a stated approximation.
• Not parameter-optimised. The thresholds (RSI 48, ±3% EMA50, 3-month relative strength, big-move %, stop/TP levels) are set once and not tuned against historical data. There is no in-sample / out-of-sample split.
• Halal-universe survivorship. The halal universe is a curated list (a few hundred names drawn from the official holdings of major halal ETFs, re-screened quarterly). Tickers that were once in but later failed Sharia screening are not present, so performance numbers on universe tickers are slightly biased upward by construction.
• RSI flat-period quirk. When 14 consecutive sessions have zero net movement (rare, but it happens on sukuk ETFs), this dashboard returns RSI = 50 rather than the canonical 100. Cross-checking against TradingView/Yahoo on such days will diverge.