PULSE INTELLIGENCE
Structure, captured before it becomes obvious.
Structure, captured before it becomes obvious.
Structure, captured before it becomes obvious.
Pulse records the option surface quietly and continuously, then looks for the rare moments when price, volatility, skew, and positioning move together with intent.
Pulse records the option surface quietly and continuously, then looks for the rare moments when price, volatility, skew, and positioning move together with intent.
Pulse records the option surface quietly and continuously, then looks for the rare moments when price, volatility, skew, and positioning move together with intent.
01 / THE OBSERVATION LOG
Where the data comes from
Where the data comes from
Everything in Pulse starts from one table: the observation log. Every fetch writes a single row per instrument capturing the things you can’t see on a price chart — ATM price, ATM implied vol, 25-delta skew (both the raw skew and the call-side/put-side pair), ATM gamma, CVOL, and downside/upside variance. One row is a snapshot of both the price and the shape of the option surface underneath it. That same table feeds every Intel chart, and it’s what makes alerting possible: you can only reason about structure if you’ve been recording it, quietly, for months, whether or not anything interesting was happening.
Everything in Pulse starts from one table: the observation log. Every fetch writes a single row per instrument capturing the things you can’t see on a price chart — ATM price, ATM implied vol, 25-delta skew (both the raw skew and the call-side/put-side pair), ATM gamma, CVOL, and downside/upside variance. One row is a snapshot of both the price and the shape of the option surface underneath it. That same table feeds every Intel chart, and it’s what makes alerting possible: you can only reason about structure if you’ve been recording it, quietly, for months, whether or not anything interesting was happening.
02 / COMBO LOGIC
From a spike to a combination
From a spike to a combination
The first version of alerting was rate-of-change: one metric moving fast enough in a fixed window. Useful, but it produced exactly the failure mode you’d expect — a single metric jumping is mostly noise, and on illiquid or thin-surface instruments it’s often a bad print. The change in thinking was that a genuinely informative read isn’t one metric moving, it’s several moving together in a specific geometry. Vol and price and gamma don’t usually reprice in the same direction with the same intensity by accident. So combos were built as conjunctions: a price condition, a vol condition, and a positioning condition, all measured against the same baseline. Individually each would be a shrug; together they’re a state change.
The first version of alerting was rate-of-change: one metric moving fast enough in a fixed window. Useful, but it produced exactly the failure mode you’d expect — a single metric jumping is mostly noise, and on illiquid or thin-surface instruments it’s often a bad print. The change in thinking was that a genuinely informative read isn’t one metric moving, it’s several moving together in a specific geometry. Vol and price and gamma don’t usually reprice in the same direction with the same intensity by accident. So combos were built as conjunctions: a price condition, a vol condition, and a positioning condition, all measured against the same baseline. Individually each would be a shrug; together they’re a state change.
The families are: a sell setup (vol bid fading while gamma builds and price goes nowhere — the market paying up for movement that isn’t arriving), a buy setup (the mirror — gamma unwinding, vol rising, price already moving), expansion (vol and skew repricing while price holds, i.e. the surface front-running the tape), gamma collapse (vol expanding as positioning unwinds — the more dangerous cousin), capitulation and confirmed capitulation (a fast flush with a rising vol bid, read long as a mean-reversion candidate), and a blow-off fade (the same read inverted, into exhaustion). Direction matters and is carried on the record.
The families are: a sell setup (vol bid fading while gamma builds and price goes nowhere — the market paying up for movement that isn’t arriving), a buy setup (the mirror — gamma unwinding, vol rising, price already moving), expansion (vol and skew repricing while price holds, i.e. the surface front-running the tape), gamma collapse (vol expanding as positioning unwinds — the more dangerous cousin), capitulation and confirmed capitulation (a fast flush with a rising vol bid, read long as a mean-reversion candidate), and a blow-off fade (the same read inverted, into exhaustion). Direction matters and is carried on the record.
03 / SIGNAL DISCIPLINE
Making them survivable
Making them survivable
The part that took the most iteration is suppression. A combo that fires eleven times in a week is worse than no alert. Three mechanisms fixed it: a hard cap on how often one type can fire per symbol in a rolling window; an oscillation guard that blocks a signal that returns to the level a previous alert started from (A→B then B→A is the alert unwinding, not a new one); and a reset-at-fire baseline — once a combo fires, the counter restarts at zero and the fire’s own readings become the new anchor, so a re-fire needs the full move measured from the fire rather than the same stale move re-measured. That last one is why a single regime can’t chain-fire.
The part that took the most iteration is suppression. A combo that fires eleven times in a week is worse than no alert. Three mechanisms fixed it: a hard cap on how often one type can fire per symbol in a rolling window; an oscillation guard that blocks a signal that returns to the level a previous alert started from (A→B then B→A is the alert unwinding, not a new one); and a reset-at-fire baseline — once a combo fires, the counter restarts at zero and the fire’s own readings become the new anchor, so a re-fire needs the full move measured from the fire rather than the same stale move re-measured. That last one is why a single regime can’t chain-fire.
Settled combo records retain setup, direction, window status, verification quality, and the resolved regime context.
04 / DELIVERY
Getting them off the screen and into a phone
Getting them off the screen and into a phone
Alert delivery is deliberately independent of the UI: a workflow listens for any new alert record and hands it to a sender. Recipients are only users who opted in on the SMS page — consent is checked, not assumed — and messages route through a registered messaging service rather than a raw number, which is what keeps carriers from blocking them. Format is plain: symbol, setup name, the readings that triggered it, and the price. No charts, no links to open.
Alert delivery is deliberately independent of the UI: a workflow listens for any new alert record and hands it to a sender. Recipients are only users who opted in on the SMS page — consent is checked, not assumed — and messages route through a registered messaging service rather than a raw number, which is what keeps carriers from blocking them. Format is plain: symbol, setup name, the readings that triggered it, and the price. No charts, no links to open.
05 / VERIFICATION
The honest part: scoring them independently
The honest part: scoring them independently
Combos are graded on price alone, and deliberately kept out of the prediction and verification records. Every one freezes its alert-time price permanently on first evaluation, then grades three checkpoints — 12, 24 and 36 hours — against the observation nearest each mark, win defined purely as price on the correct side of that frozen base.
Combos are graded on price alone, and deliberately kept out of the prediction and verification records. Every one freezes its alert-time price permanently on first evaluation, then grades three checkpoints — 12, 24 and 36 hours — against the observation nearest each mark, win defined purely as price on the correct side of that frozen base.
Three rules give it integrity. Windows only settle while observations are actually flowing for that symbol; if fetching is down, the window stays open rather than grading against a stale price. Each graded print carries a quality label based on how far it sits from the intended checkpoint — verified, delayed, or unavailable — and unavailable checkpoints are excluded from the win/loss tally while their evidence print and the next session’s read are kept alongside for transparency. And settled results are permanent: they are never regraded, even when a later observation lands closer to the mark.
Three rules give it integrity. Windows only settle while observations are actually flowing for that symbol; if fetching is down, the window stays open rather than grading against a stale price. Each graded print carries a quality label based on how far it sits from the intended checkpoint — verified, delayed, or unavailable — and unavailable checkpoints are excluded from the win/loss tally while their evidence print and the next session’s read are kept alongside for transparency. And settled results are permanent: they are never regraded, even when a later observation lands closer to the mark.
That’s the piece worth emphasising in the article — combos aren’t scored by the same intelligence that produced the prediction record. They’re scored by price, after the fact, with the base frozen at the moment of the alert. The regime label is stamped on only once the 36-hour window is fully settled, so it reflects the macro state at resolution rather than the state at firing. The Intel page surfaces all of it: hit rates by setup type and direction, by symbol, win/loss by window, the verification-quality split, and the settled totals — including the ones that lost.
That’s the piece worth emphasising in the article — combos aren’t scored by the same intelligence that produced the prediction record. They’re scored by price, after the fact, with the base frozen at the moment of the alert. The regime label is stamped on only once the 36-hour window is fully settled, so it reflects the macro state at resolution rather than the state at firing. The Intel page surfaces all of it: hit rates by setup type and direction, by symbol, win/loss by window, the verification-quality split, and the settled totals — including the ones that lost.
The Intel page makes every settled outcome inspectable, including the read quality and the results that lost.