METHODOLOGY
How it's read
Every few hours the pipeline pulls a year of daily closes for the whole book and computes the technicals deterministically. No model is involved at this step — the math is fixed and reproducible.
RSI(14)
Wilder's smoothing of average gains vs. losses over 14 sessions.
Bollinger
20-day SMA, ±2σ bands, with %B (position in the band) and band width.
Trend
Price relative to the 50-day and 200-day simple moving averages.
Drawdown
Distance from the trailing 52-week high.
Momentum
Rolling 1-month, 3-month and 1-year returns.
Money Flow Index (14)
Volume-weighted RSI — tracks whether buying or selling pressure is backed by real volume. >80 = flow overheated (distribution risk); <20 = flow washed out (accumulation). A price high not confirmed by MFI is a classic turn tell.
Each name is then scored across three buckets:
SETUP
- Oversold (low RSI)
- Flow washed out (low MFI)
- Thesis intact
- Catalyst near
CROWD
- Overbought (high RSI)
- Flow overheated (high MFI)
- At 52-week high
- Above upper Bollinger
SQUEEZE
- Bollinger width at multi-month lows
Turn detection & the peer-review rule. A live ping on the conviction map only fires when a turn is corroborated — the Technician leads, and a signal must clear two or more independent conditions before it shows. A bright-red SELL needs at least two of: TD9 sell ripe · crowd ≥ 90 · RSI ≥ 72 · MFI ≥ 80 — so no single indicator can paint a name red on its own. The Act Now bar is stricter still (three-plus conditions aligned). This is the same discipline applied to the live feed: every signal must come from a live source and be peer-reviewed by at least one panel seat before it reaches the screen — modeled-but-unconfirmed reads never ping.
The top-scored names go to the editorial layer, paired with the chokepoint thesis for each ticker. The model writes the narratives. It does not compute the numbers.
Two data modes. The formulas above are real and run in your browser. Synthetic (default): each name's history is a deterministic series anchored to its provided price & 1-day move — illustrative, not live. Live: click the data chip (top-right), paste a free Twelve Data key, and the dashboard pulls a real year of daily closes per name and recomputes every number for real. DeMark TD setup, Weinstein stage, Bollinger, RSI and the correlation matrix all become true readings; names the free source doesn't carry (thin OTC / pre-IPO) stay synthetic and are tagged ○. Nothing is stored unless you tick “Remember key.” Daily refresh: the headline, investor letter and the per-member Panel reading are generated from whatever data is loaded each time the page renders, and they carry today’s date — so with a remembered key the site auto-pulls fresh closes on each visit and the whole brief updates itself. These narratives are rule-based (deterministic templates over the live metrics), not an LLM writing fresh prose; a true model-written daily letter would need a small scheduled backend.