Two engines · MNQ futures + single-name stocks · one feed
Two autonomous engines watch dealer gamma walls and institutional sweep flow all session, and alert only when everything lines up — entry, stop, target, exit. Then they publish the outcome, costs and all.
how to read this — the walls are the price levels where options dealers concentrate; the engine watches how price behaves into them. This snapshot is a sample until the engine's feed is wired in.
02 / The two engines
One watches dealer gamma walls on MNQ futures. One watches clusters of $1M+ institutional options sweeps on single-name stocks. Both check nine conditions every minute of the session and stay silent until all nine read go — then hand you entry, stop, target, size, and an explicit exit. Not a hunch. The whole trade.
MNQ futures
Fades MNQ at 0DTE QQQ gamma walls, only when dealers are long gamma. One mechanism — nine hard boolean gates, all of which must read go before an alert exists.
Every fire hands you
Every fire gets a matching exit alert — no signal is left open-ended.
Kill at 40 · graduate at 100 · frozen criteria
Single-name stocks
Momentum after clusters of $1M+ same-direction institutional options sweeps. Same constitution: hard boolean gates, a one-minute loop, nothing fires unless every gate reads go.
Every fire hands you
Every fire gets a matching exit alert — no signal is left open-ended.
Kill at 40 · graduate at 100 · frozen criteria
03 / Daily briefs
Even on a silent session you get the read. A pre-market brief at the open and a full recap after the close — the kind of desk-grade levels report that usually runs three figures a month — land in Discord and on the cockpit, every trading day.
Before the session settles, the machine hands you the read: the gamma walls both engines are watching, the regime, the institutional sweep clusters carried in overnight, and where each engine is armed for the day.
After the close, every fire the engines published today — entry, bracket, the exit, and the cost-adjusted outcome — next to the stand-down count for the setups that didn't clear. No cherry-picking, no quiet losers.
Delivered to Discord and mirrored on the cockpit.
Pre-registered evaluation criteria — not a performance claim, projection, or promise of results. Risk Disclosure.
04 / The live cockpit
The engine writes a read-only dashboard — the gate board, the gamma walls, the fires, the day strip. Most minutes it’s standing down, and that’s the show: you’re watching a machine turn down hundreds of almost-setups so it can take the few that clear every gate. Nothing to babysit. Just the evidence, accumulating in the open.
Read-only. The engine writes it; nobody can touch it from here.
05 / How it works
This is the machinery under both engines — the reason the silence means something. No mechanism gets to fire on a hunch; it clears every gate, or it says nothing.
Each engine hunts exactly one thing — no kitchen-sink of indicators, no discretion. Wall Reject fades MNQ into 0DTE gamma walls; Sweep-Cluster Ignition rides momentum after institutional sweep clusters. One idea, expressed as code.
No scores, no weights, no vibes: each gate is a yes/no check. Every gate must read go before an alert exists. One no is a no — which is why most minutes are silent.
Every minute of the session, the engine polls the board. State changes land on the minute tick — nothing drifts in between, nothing fires on a whim.
Every minute the engine doesn't fire, it records which gate stood it down. The silence has a paper trail — that's what you're auditing on the cockpit.
06 / The evidence gates
Before a lane goes live, its kill and graduate thresholds are written down in public — dated, frozen, and out of our hands. Then the fires are counted.
Kill gate
40 fires
KILL at 40 fires if unprofitable after modeled costs. No appeals, no tuning-and-rerunning.
Graduate gate
100 fires · +0.10R
GRADUATE at 100 fires only at a cost-adjusted mean of +0.10R or better. Checkout stays closed until then.
Pre-registered evaluation criteria — not a performance claim, projection, or promise of results. Risk Disclosure.
07 / The graveyard
When you’re evaluating any alert service, ask one question: show me the strategies you killed.
08 / Costs
Commissions, fees, and slippage are modeled to the dollar and baked into every published figure — including the kill and graduate gates themselves. A strategy that only works before costs doesn’t work.
If you ever catch a pre-cost number anywhere on this site or in the journal, report it to hey@composite.fyi — we’ll fix it publicly and note the correction.
09 / What an alert looks like
Direction-coded, entry / stop / target, contract count, valid-until — and an explicit EXIT follow-up for every fire. The pair below will be a real fire from the journal, with its real date. We don’t publish mocks.
Every alert is an impersonal publication delivered to all subscribers identically and simultaneously — not advice, not tailored to anyone.
10 / Built for / not for
Built for
Not for
11 / The offer
Checkout opens when Wall Reject clears its graduate gate — 100 fires at a cost-adjusted mean of +0.10R or better — and not a day before. Selling signals ahead of the evidence is the exact thing this project exists to not do. Join the waitlist and you lock the founding rate for when — if — that gate clears. Until then you pay nothing, and you can watch the evidence accumulate on the live cockpit like we do. If the lane dies at 40fires instead, you’ll get the post-mortem, not an invoice.
One paid tier
$49/mo founding
All live lanes’ real-time alerts plus full cockpit access — you subscribe to the discipline, not to a strategy. Lanes come and go by pre-registered rules; the price doesn’t split into tiers when they do. Rate locked while continuously subscribed. Standard rate after graduation: $89/mo. Monthly, cancel anytime, one-click cancel from the account page.
Free tier — the trust engine
The public journal and stand-down feed (delayed or end-of-day) plus the end-of-day fire recap with brackets and cost-adjusted outcomes. Audit us for free as long as you like; pay only for real-time. There is no free trial of real-time alerts — the delayed tier is the evergreen trial.
The evidence-gated waitlist
Email only. No card. Locking the founding rate costs nothing and commits you to nothing.
Email only — no card, no counter, no drip. One email if the gate clears; one post-mortem if it doesn’t.
Pre-registered evaluation criteria — not a performance claim, projection, or promise of results. Risk Disclosure.
12 / Engineering floor
Nothing below is a promise — each line is a property of the system you can check.
2,700+ automated tests
The engine ships behind a test suite in the thousands — run, not estimated.
No order-placing code path
The engine publishes alerts. There is no code path that touches a brokerage account — nothing to fat-finger, nothing to run away.
Read-only cockpit
The engine writes the dashboard; nothing on the web can write back. Nobody can touch the numbers from here.
No analytics, no trackers
This site sets no analytics, no pixels, no fingerprinting — verifiable in the page source and the repository.
13 / The honest disclaimer
Our first-generation engine was retired after its evaluation window finished under breakeven once costs were counted. That retirement is why the kill and graduate gates above are now written down before any lane goes live.
A win rate.
Accuracy numbers sell subscriptions, not discipline. What gets published is the journal: every fire, costs included, with the sample size attached.
A track record before a gate clears.
Until a lane graduates, the only number in our marketing is the pre-registered threshold itself. Evidence-period results live in the journal, labeled as what they are.
How many alerts you'll get.
The engine fires when every gate reads go, and not otherwise. Some weeks that is rarely. Frequency promises manufacture signals.
That a strategy is 'live' when it isn't.
Every lane carries a status chip — live, launching, or killed — and dead lanes move to the graveyard in the past tense.
That you're running out of time.
No countdowns, no seat counters, no 'almost full.' The founding rate waits for the evidence, same as you.
What to trade.
The engine fired; that's the whole message. Impersonal publication, identical for every subscriber, decisions entirely yours.
14 / FAQ
Because most minutes fail at least one gate. The engine evaluates every minute of the session and publishes only when every hard boolean gate reads go — one holding gate is a stand-down, journaled and silent. That silence is the product working, not the product being broken.
It gets killed, publicly, by criteria written down before it launched: 40 fires unprofitable after modeled costs and the lane dies. We have already done this once — Wall Break was scrubbed at its pre-build study, before a single alert shipped, and post-mortemed in public. Dead lanes move to the graveyard in the past tense.
No. Every alert is an impersonal publication delivered to all subscribers identically and simultaneously. It is not tailored to any person's situation, and it is not a recommendation — the engine fired; the decisions are yours. Trading futures and options involves substantial risk of loss.
Checkout is closed until Wall Reject clears its graduate gate — 100 fires at a cost-adjusted mean of +0.10R or better. Selling signals ahead of the evidence is the exact thing this project exists to not do. The waitlist locks the founding rate and takes no card.
The engine itself never trades — there is no order-placing code path in the system, which you can verify. What anyone does with a published alert, including us, is a personal decision and is never used as marketing here.
The journal. Every fire — including suppressed ones — is recorded with its timestamped alert price, the cost model applied, a PAPER or LIVE label, and its outcome. Every published figure is cost-adjusted and carries its sample size. If you catch a pre-cost number, report it and we fix it publicly.