The flow finds the edge. The structure makes it convex.
Convexity is selected, not predicted. The print stream nominates candidates; structure and sizing decide whether the trade is worth taking. The homepage argues that thesis. This page publishes the working procedure: seven milestones run in order, with a stop on failure at any gate.
Four gates, run in order.
Every candidate the flow nominates is forced through four sequential gates before a contract exists. The order matters: convexity is checked before edge, edge before size, and a failure at any gate stops the evaluation and names itself. There is no override that lets conviction outrank structure, and there is no partial credit. A trade that clears three gates and fails one is not a trade.
The fourth gate, the naked-short block, is currently disabled by operator policy. The rule stays published because the method is published: the blocking logic is preserved in the codebase and the re-enable path is documented. A disabled gate that remains visible is a different thing from a gate that quietly disappears.
| Gate | Rule | Condition |
|---|---|---|
| 01 · Convexity | gain ≥ 2× loss | Defined-risk structures only. The capped loss is known before the order is submitted. |
| 02 · Edge | signal precedes price | A specific, data-backed dark-pool or OTC signal that has not yet moved the lit price. |
| 03 · Risk | ≤ 2.5% bankroll | Fractional Kelly sizing with a hard per-position ceiling. The cap is enforced, not advisory. |
| 04 · Naked shorts | no naked shorts | Disabled by operator policy on 2026-04-30. The blocking logic is preserved for re-enable. |
Seven milestones, candidate to routed order.
The gates say what must be true. The milestones say when each check runs. A candidate enters at milestone one and either exits as a routed, journaled order or stops at the first milestone it cannot clear. Nothing skips ahead, and nothing that fails late gets to retry early.
Validate the ticker
Seasonality, analyst ratings, and news are pulled first, and all three are context, not gates. A favorable seasonal window never authorizes a trade on its own, and an unfavorable one never vetoes strong flow. The milestone exists to frame the candidate before the flow evidence is read.
context · seasonality / ratings / newsRead the dark pool flow
Off-exchange prints are aggregated for the candidate. During market hours the partial session is volume-weighted interpolated against the prior five-day average, and both the actual and the interpolated figures are always output: the math is published in full below. The read is also checked against the tail-risk regime, because a signal fighting the regime is downgraded, not ignored.
flow · actual + interpolatedThe Crash Risk IndexRead the options flow and OI changes
The options tape is read next: put/call balance, the side of the print, and the open-interest changes that show whether positioning actually followed the paper. The open-interest check is a required step in the method, not an optional one.
flow · options + open interestMake the edge decision
A binary PASS or FAIL on the edge gate. FAIL means stop: no structure is built, no Kelly arithmetic is run, and no trade is logged. The decision itself is still recorded as NO_TRADE, because a pass on discipline is evidence too.
gate · pass / failSelect the structure
A defined-risk options structure is selected to express the read with convexity. Reward-to-risk below 2:1 is a stop, not a discount, and the trade specification is written down at this milestone before any sizing happens.
convexity · R:R ≥ 2:1The defined-risk structure catalogSize with fractional Kelly
Fractional Kelly sets the position against bankroll with a hard cap of 2.5% per position. Conviction adjusts size within the ceiling and never past it. Sizing runs only on a structure that already cleared the edge and convexity checks.
risk · ≤ 2.5% bankrollThe fractional Kelly sizing methodLog the decision
Executed trades append to a journal that is append-only by design, so the record cannot be rewritten after the fact. NO_TRADE decisions are recorded as well. Routing itself runs through Interactive Brokers, and the audit trail keeps the full chain from print to order.
record · append-only journalHow orders route through Interactive BrokersReading a partial session.
Dark pool prints accumulate through the session, so a raw total read at 11:00 says little about the day. The method handles this with a published interpolation: project the full session from elapsed minutes, blend the projection with the prior five-day average, and weight the blend by how much of the session has actually printed. Early in the day the prior dominates; late in the day the tape speaks for itself.
- progress
- = minutes since 9:30 ET / 390
- projected
- = actual / progress
- blend
- = (projected × progress) + (prior 5d avg × (1 − progress))
- pace
- = actual / (avg prior × progress)
| Session progress | Confidence | Prior weight |
|---|---|---|
| 0–25% | VERY_LOW | 75%+ |
| 25–50% | LOW | 50–75% |
| 50–75% | MEDIUM | 25–50% |
| 75–100% | HIGH | <25% |
- The interpolated figure feeds the edge decision; the actual figure is reported next to it, always.
- LOW or VERY_LOW confidence: re-evaluate after 2 PM ET rather than force a call on thin data.
- Pace above 1.2× the prior average says the flow is real, not an artifact of the projection.
- Actual flow running opposite the prior average suggests a reversal, not a continuation.
The output carries its own confidence grade, and the grade has teeth: a LOW or VERY_LOW read is not traded on, it is held for re-evaluation in the afternoon. Both figures, actual and interpolated, are always reported together, because a blended number that hides its inputs is a number you cannot audit.
Thresholds, stated.
A method that says bearish flow without saying where bearish begins is a mood, not a method. The bands below are the published thresholds the evaluation uses. They make any single read checkable after the fact: the inputs are on the tape, the bands are on this page, and the conclusion either follows or it does not.
| Signal | Bands |
|---|---|
| P/C ratio | >2.0 BEAR1.2–2.0 LEAN_BEAR0.8–1.2 NEUTRAL0.5–0.8 LEAN_BULL<0.5 BULL |
| Flow side | ask-dominant BUYINGbid-dominant SELLING |
| Analyst buy% | ≥70% BULL50–69% LEAN_BULL30–49% LEAN_BEAR<30% BEAR |
| Seasonality | >60% FAVORABLE50–60% NEUTRAL<50% UNFAVORABLE |
One caveat is part of the spec rather than a footnote: seasonality and analyst ratings are context, not gates. They frame a candidate at milestone one, but strong flow overrides weak seasonality, and no ratings consensus can pass or fail an edge decision on its own.
Six repeatable expressions.
Run the method long enough and the outputs cluster. The registry names six recurring expressions: each is a specific edge source mapped to a defined-risk structure with a convexity floor the trade must clear at entry. They are worked outputs of the seven milestones, not independent strategies. Every one of them runs the same gates, the same sizing, and the same journal.
| Play | Edge source | Structure | Convexity floor |
|---|---|---|---|
| Dark Lead | Off-exchange accumulation | Call debit spread | ≥ 3.0× |
| Gamma Pin | Dealer gamma walls | Iron condor | ≥ 2.2× |
| Panic Gap | Widening VCG-R | Put backspread | ≥ 4.0× |
| Rotation Front | GRG sector migration | Call calendar | ≥ 2.5× |
| Crash Hedge | Elevated CRI | Put debit spread | ≥ 5.0× |
| LEAP Mispricing | IV term mispricing | Long LEAP call | ≥ 2.0× |
The floors differ because the edges differ. A tail hedge bought when the Crash Risk Index is elevated demands more convexity than a pin structure sold into a dealer gamma wall, because the hedge pays rarely and must pay large when it does. The structures themselves come from the same defined-risk catalog that milestone five selects from.
Asked about the method.
Four questions that come up when the procedure is read closely, answered plainly.
What happens when milestone 4 fails?
The evaluation stops. FAIL at the edge decision means no structure is built, no Kelly sizing is run, and no trade is logged. The decision itself is still recorded as NO_TRADE, so the record keeps evidence of discipline as well as evidence of trades.
How long is the lead window?
It is measured per ticker, not assumed. The scanner surfaces the lead window alongside the flow score, and the window is treated as a property of the name and its flow history rather than a constant of the method.
Why interpolate intraday flow at all?
Because a raw partial-session total is misleading at noon and nearly meaningless at 9:45. The method projects the session from elapsed minutes, blends the projection with the prior five-day average, grades its own confidence by session progress, and always outputs both the actual and the interpolated figures.
Does the method predict direction?
No. It measures positioning: where off-exchange volume is accumulating, which side of the tape the options flow prints on, and whether open interest followed. Structure selection and Kelly sizing then price the risk of expressing that read. Convexity, not prediction, carries the trade.
Run it against the live tape.
The demo instance runs the same scanner, the same gates, and the same journal on seeded data.