LABS

I have ฿X a month. Help me start — without the spreadsheet.

WHAT

A guided single-page companion for someone who has never invested. Takes income, monthly surplus, target retirement age. Returns: a one-paragraph recommendation, a bucket split (cash / feeders / dream-trades), and the next concrete action this week.

DATA

Plan wizard logic + savings rate + goal ladder (we already compute it on /plan).

DIFF

Not a screener, not a backtest. A 'where do I start' companion. No jargon unless you turn it on.

Stocks trading below the cash on their books — Graham's most extreme test.

WHAT

Graham's 1934 'cigar butt' screen: a stock is interesting when the price is below net current asset value per share (cash + receivables, minus all liabilities). Live filter for SET names currently meeting the bar, with the discount % shown honestly.

DATA

ncavPerShare, ncavDiscountPct (already in MOCK_STOCKS, see src/lib/graham.ts).

DIFF

Most Graham screens start with P/E or P/B. This one starts deeper: with the working capital itself.

What happened the 30 days after the bell rang? Replay the moment 3+ signals aligned.

WHAT

Replays every historical moment a SET ticker triggered a 3+ family confluence. Shows the next 30 trading days of price action, side by side with the SET baseline. Honest about the win rate, the average move, and the worst drawdown.

DATA

anomaly.confluence + prices.json. We already build this graph for /signals; this lab shows the post-event window.

DIFF

Most 'signals' tools show the trigger and stop. This one shows the outcome — even when the outcome is uncomfortable.

Two names that always moved together — until they didn't.

WHAT

Finds pairs of SET names with high historical correlation that recently decoupled. The 'break' is the moment a relationship snapped — sometimes mean-reverting, sometimes a regime change. Outputs a ranked watchlist with the current correlation vs the 90-day baseline.

DATA

rolling correlation on prices.json (we compute it for the signal web).

DIFF

Most correlation tools show 'current correlation.' We show the BREAK — the moment a relationship snapped, and which way it went.

Greenblatt's two ranks on SET50 — earnings yield × return on capital. Not Graham. A complement.

WHAT

Ranks SET50 by Joel Greenblatt's 2006 'Magic Formula' — first by earnings yield (cheap), then by return on capital (good business). The top 5-7 names are the candidates. Pure orthogonal to the Graham shelf.

DATA

pe, roe, marketCap — already aggregated by magicFormulaRanks() in graham.ts. We just need the UI.

DIFF

Not a Graham value screen (we already have that). Greenblatt's complement: cheap AND good, ranked, not picked.

Real-time list of names currently meeting all 7 of Graham's defensive tests.

WHAT

A live tape of SET names whose defensive score (Graham's 7-criteria binary test) is currently 7/7. As prices move, names enter and leave. The conveyor is the watchlist the defensive investor never has to think about.

DATA

defensiveScore (0–7), already in MOCK_STOCKS via defensiveScore() in graham.ts.

DIFF

Not a screener snapshot. A LIVE list — names enter and leave as their defensive tests change day to day.

When the signal system said STRONG BUY, how often was it right? Most tools never tell you.

WHAT

Honest backtest of the signal system. For every confluence event in our history, did the next 30 days actually deliver the direction the signal implied? True positives, false positives, win rate, and a confusion matrix. The system stands or falls by the numbers — no marketing.

DATA

anomaly.confluence + prices.json — same as the PARLOR but with an outcomes overlay.

DIFF

Most signal services never publish their track record. We do. If the numbers are embarrassing, the numbers stay.

Seven signals say seven different things. What does the desk actually think?

WHAT

Every forward-looking signal on the desk — Forward View, Kronos, Mr. Market Mood, Market States, and (when on) Polymarket — votes BULL, BEAR, or abstains. Votes are bucketed by horizon (now / 20d / 60d / 90d / event) so a 20-day forecast cannot cancel a 90-day baseline. Each horizon gets a verdict, a confidence, and a roll call of who voted what.

DATA

forward-view.json (90d) + kronos-signals.json (20d) + market-mood.json (now) + market-states.json (now) + /api/belief (event). One function, one output, no fetching.

DIFF

Most dashboards either show every signal on the same chart (the user has to weight them) or they pick one. This gathers them, weights them, and surfaces the disagreement honestly — the cautionary note names any high-weight dissenter. A unanimous 2-vote verdict and a unanimous 5-vote verdict look the same on a chart; here they don't.

A daily first-person voice for the market — built from the actual mood + signal data.

WHAT

A deterministic template that turns today's mood score, the active confluence list, and the top signal into a 200-word first-person paragraph from Mr. Market's perspective. Not a chatbot — a voice grounded in real numbers, refreshed at the same cadence as the data.

DATA

market-mood.json + signal web state + the top confluence subject. The voice never invents — every claim is grounded in one of those three sources.

DIFF

Not a chatbot. A voice that only knows what the data knows — and a click on any sentence shows you which number it's grounded in.

What would SET be doing today if [event] hadn't happened?

WHAT

The structural breaks in SET's price history are dated. Roll back to the day before a chosen break, replay the model forward, and the chart shows ACTUAL vs COUNTERFACTUAL paths. Honest about the limits: structural breaks are detected, not caused, so the counterfactual is a 'what if no shock' world, not a 'what if no cause' world.

DATA

structural breaks timestamps + prices.json + the drift model. We have the breaks; we have the model; we don't have the engine yet.

DIFF

Most 'what if' tools are made up. This one would use the real breaks we already detect. The honest version says what it can't do.