1 · The thesis
The quiet divergence
When a large buyer wants a big position, they don't lift the offer on the open market — that moves the price against them. They work the order quietly: in dark pools and other off-exchange venues, in small tranches, over weeks. FINRA publishes how much of each stock's weekly volume trades in those venues. That's the raw material.
The tell we're looking for is a divergence: off-exchange share climbing steadily while the price stays flat. Heavy buying that should lift the price, but doesn't — because it's being absorbed. That's the whole idea.
2 · Reading the table
What each column means
Tier
The verdict for the week. STRONG, WATCH, or —. It's a screen, not a trade signal: a shortlist of names worth a human look.
Score
The notice score — a weighted count of which flags fired. Used only to rank names within a tier. Higher = more of the important flags lit up.
Flags
Which of the six confluence signals fired this week. A single flag is noise; the tool cares about agreement. ★ divergence is the one that matters most.
Dark Share · Δ Price · CMF · CUSUM
The underlying measurements the flags are computed from. Click any row to open the drawer and see every input, the exact rule for each flag, and whether it fired.
Confluence over conviction. No single number decides anything. A name rises to the top only when several independent measurements — level, trend, persistence, money flow — point the same way at once.
3 · The six signals
The flags, one by one
Each flag is a plain, inspectable rule over the last 16 weeks of data. No black box — here's exactly what each one measures and when it fires.
The whole thesis in one rule. Dark-pool share is in a clean uptrend and the price has gone essentially nowhere. Someone is building; the tape isn't showing it.
dark_slope > 0 AND dark_r² > 0.30 // clean uptrend in off-exchange share
price_4w ≤ 2.0% // price essentially flat-to-down
A z-score answers "how unusual is this week?" by measuring how many standard deviations the latest dark-pool share sits above the stock's own recent average. It's a level check — a single anomalous week.
# fires when z > 1.5
A z-score catches one odd week; a trend catches a steady climb. We fit a straight line to the weekly series (ordinary least squares) and ask two things: is the slope positive, and is the fit tight? R² measures tightness — how much of the movement the line actually explains. High R² = a real trend, not scatter.
slope = Σ(xᵢ−x̄)(yᵢ−ȳ) / Σ(xᵢ−x̄)² R² = explained variance ÷ total variance
# fires when slope > 0 AND R² > 0.30
Patient buyers add a little each week — too small to trip a z-score. A CUSUM (cumulative sum) is built for exactly this: it adds up how far each week runs above the baseline, ignoring small wiggles (the slack k), and only accumulates when the excess is real. A persistent drift, however gentle, eventually pushes the running sum past the decision line.
# fires when the running sum peaks above h = 4.0σ
The dark-pool signals live off-exchange; Chaikin Money Flow is the on-exchange confirmation. Each day it asks where in its range did the stock close? Closing near the high (buyers won the day) is positive; near the low is negative. Each day is weighted by its volume and summed over 20 days. Positive CMF = net accumulation in plain sight, backing up the quiet story.
CMF = Σ(MFM · volume) / Σ(volume) # over 20 days
# fires when CMF > 0.05
When unusually heavy volume arrives but the price stays put, a big seller is being met by an equally big buyer at one level — the demand is absorbing the supply. We flag it when recent volume is a full standard deviation above its own baseline while the 4-week price move stays inside ±5%.
4 · Score & tiers
From flags to a verdict
The notice score weights the flags by importance
Not every flag carries equal weight. The signature divergence and the persistence of CUSUM count for the most; a lone level anomaly counts for less. The score is just their weighted sum — used to rank names inside a tier.
How the tier is decided
divergence fired alongside dark_z or cusum, or 3+ flags agree.
Exactly 2 flags fired. Something is stirring, but not yet confluence.
Fewer than 2 flags. Nothing to see this week.
5 · Under the hood
The original v1 composite
Before the trend layer above, Quiet Money scored each name with a simple weighted sum of three sub-signals (a fourth is context-only). The dashboard now ranks on the richer trend flags, but the composite is the same intuition distilled — and still lives in the codebase.
S3 uses EDGAR filings where the name is the subject of a 5% stake (someone crossing 5% of it), not stakes it took in others. S4 counts insider Form 4s for context but isn't yet buy/sell-filtered, so it doesn't move the score.
6 · Read with care
What this is — and isn't
- •A screen, not a signal. Tiers surface names worth investigating. They are not buy or sell recommendations.
- •Thin history early on. The statistics need weeks of data to mean anything. A symbol with only a handful of stored weeks shows a caution note in its drawer — read those with a grain of salt.
- •Not yet backtested. The tool self-reports whether its notice score actually predicts forward returns, but that check stays gated until enough weekly cohorts accumulate and their forward returns resolve. Until then, treat every flag as a hypothesis.
- •Dark-pool share isn't direction. FINRA reports off-exchange volume, not whether it was bought or sold. The whole point of confluence — pairing it with money flow and flat price — is to infer the direction the raw number can't give you.