Whitepaper

This is the third public edition. It replaces the edition of August 22, 2026; where the two disagree, this document is correct. It describes an early, pre-revenue venture: nothing here is investment advice, an offer to sell securities, or a guarantee of future results.

1. Executive summary

Green Sand Strategies runs two things under one name. A private, pre-registered trading research program builds and tests a systematic strategy on the founder’s own hardware. A public product turns what that research produces into something people and software agents can use.

The research engine is a multi-timeframe trend-alignment breakout system. It runs privately, stays off the cloud, and never places a trade on its own. What it produces is a stream of timestamped evidence: every decision it makes, including every time it decides not to trade, gets written to a permanent, append-only record before anyone knows how that decision turns out. That record is the company’s real asset. The public product is how customers get access to what the record produces.

The live signal daemon has been logging real decisions since August 8, 2026, and the public site has been live since August 9, 2026, with working accounts and paper trading in place. The record is young, and most of what it has logged has not resolved yet. That is the honest headline: the machinery works, and it has not yet produced a result anyone should act on.

2. The thesis

Green sand is a foundry term: a mixture of sand, clay, and water, packed around a pattern to cast molten metal into a finished part. A foundry’s value lives in two places at once. There is the pattern shop and the metallurgy — the proprietary process that turns raw material into something with the right strength and finish. And there is the catalog of finished castings that customers actually buy. A customer needs the part to be reliably good. They do not need the furnace recipe to use it.

That split is the whole company. Two convictions sit underneath it.

Evidence has to be pre-registered to mean anything

A backtest optimized after the fact proves that the researcher had enough parameters to fit the past, nothing more. Our engine commits to a hypothesis, a scoring rule, and a falsification condition before seeing forward data, then logs every live decision — trade or no-trade — to an append-only file it cannot go back and edit. Several parameter changes that looked obviously correct have already been tested this way and rejected; the strategy configuration carries that falsified-fix record inline, on purpose, so nobody quietly re-adds a gate that already failed.

The research and the product are two businesses

The signal engine loses its value the moment its internals go public: the measures it uses, their parameters, position sizing, and the full scenario logic stay on private hardware, permanently. What reaches the public is a derived, sanitized signal contract — mapped states like signal, neutral, vetoed, and blackout — with the underlying risk math stripped out. The research stays in the foundry; the castings go out the door.

3. The engine

This is the part of the business that never ships directly to a customer. It is worth describing anyway, for the same reason a foundry describes its metallurgy without handing over the furnace: the quality of the process is what makes the catalog worth trusting.

Strategy

A trend-alignment breakout system built from three independently swappable modules: entry scoring, participation confirmation, and exit scoring. Which measures fill those slots, and the parameters behind them, are not published.

Regime gate

A higher-timeframe veto sits in front of every entry, with telemetry so the gate’s own performance gets measured over time, the same as everything else.

Signal daemon — live since August 8, 2026

Scores the watchlist on a fixed cadence, applies the regime gate, checks earnings blackouts, and logs every decision — including no-trades — to an append-only record. It places no orders and cannot reach an order endpoint. Any future execution layer will be a separate, separately-armed service.

Validation program

Monte-Carlo low-timeframe backtests, circular-block and cluster-robust confidence intervals, a reproducible go/no-go bootstrap gate, and an explicit overnight gap-through-stop risk measurement. This is instrumentation built specifically to make it hard to fool yourself.

The register

A dated, append-only list of every hypothesis the engine is tested against, with its falsifier written down in advance. Entries are never edited, only amended, and every amendment says what it changes and why.

A second research line

A complementary pre-open research line has cleared its own pre-registered validation gate and runs on an automated weekday schedule. It is a standalone toolset, not merged into the primary signal pipeline, and the newer readings it records are not allowed to influence its selections until forward evidence justifies letting them — a state the code enforces and a test protects.

4. How we decide whether a result is real

Five commitments.

Pre-registration

Hypotheses are written down, with their falsifiers, before the data that would test them exists. The register is append-only: entries are never edited, only amended, and amendments are dated and say what they change.

Append-only forward evidence

Every decision the live model makes is logged as it happens, before the outcome is known. That record cannot be edited after the fact, which is what makes it evidence rather than a recollection.

Honest uncertainty

Setups on the same stock are not independent events, and overlapping windows share outcomes. Our intervals resample whole stocks rather than individual setups, which widens them considerably. Where we have too few distinct stocks for that to be reliable, we say so and decline to publish a number.

Stated power limits

Every result carries the smallest effect that sample could have detected. A finding of “no difference” from a sample that could only ever have seen a large one is not evidence of anything, and we label it that way.

A threshold set in advance

What counts as publishable is fixed before the numbers are looked at. Results that miss it are reported as missing it, rather than reframed.

5. What the evidence does and does not show

We have not established that a higher score means a better trade. Nor the opposite. Against our own publication threshold, every timeframe we measure currently falls short — mostly for having too few distinct stocks to support a claim about a whole market — and we say that plainly rather than publishing the subset of numbers that would look better.

This is why the signal board’s order — entry signals first, higher scores first within each group — is presented as a reading order and nothing more. A higher card is easier to find, not a better trade.

The measurements that would move us, stated before we have them: a relationship between score and outcome that holds across enough distinct stocks to be worth reporting, after correcting for how far each setup was required to travel; evidence that the setups the model refuses would in fact have lost; and results that survive on stocks we did not choose in advance. Our current evidence comes from a curated research library, which is a reasonable place to develop a model and a poor place to prove one.

6. The product

If the engine is the foundry floor, the app is the storefront and the catalog combined. There is an analogue-style paper-trading app for people, and a parallel API for agents at api.greensandstrategies.net, both live since August 9, 2026.

The stack is single-operator-friendly and all-Cloudflare: Workers for the API and auth, Pages for the frontend, D1 for accounts and the paper-trading ledger, KV for cached signal snapshots, WAF rate limits for abuse protection, and cookieless analytics. Zero-egress object storage is reserved for datasets but still switched off pending billing, so a published dataset ships as a static asset. The private signal daemon pushes sanitized snapshots into this stack over an authenticated ingest endpoint. Engine internals never travel as code or as data. Cost discipline is structural: a hard $50/month infrastructure ceiling, no autoscaling dependency, no payment method on file with any provider that could run away on cost, and billing alerts well before the ceiling.

The free board is delayed 15 minutes. One exception, stated plainly: the two owners read the same board with no delay, through a gated route that is not advertised and cannot be bought, and their paper fills price at that same undelayed snapshot. The rule we hold ourselves to is that no account ever fills at a price it was not shown, and every fill records the delay it priced at, permanently, so the two regimes can never be pooled into one track record.

Shipped and live today

• The public signal board — the sanitized, delayed feed on two timeframes, no account required
• Passkey accounts — no password, no email
• An event-sourced paper-trading ledger — a full trade history model, no real money in the loop

7. How to check us

• The live board and the public JSON feed show every decision, including the refusals, delayed 15 minutes and free.
• The score-reliability table we published as a downloadable dataset was withdrawn on August 25, 2026. Nothing in it was wrong: it carried the required-move figure that makes each hit rate readable, and it shipped the thin rows suppressed rather than estimated. What it could not do was be the only evidence pack on offer, because it was measured on the curated research library — the thing section 5 says is a poor place to prove a model. A pack that answers that standard replaces it; until then we would rather offer none than let this one stand in for evidence we do not have.
• Anything we cannot support, we mark as unsupported rather than omitting. If you find a claim on this site that the evidence does not carry, tell us and we will correct it in public.

8. Roadmap

Near term

Research access and the first dataset-shop listings. Research access is engine results for the 4-Hour and 1-Day timeframes, plus a daily fifteen-name list from the pre-market scan, alphabetical and unranked — not a faster view of the free board, which stays delayed and stays free. One price has been set: $17/month, opening January 2027. It is gated on the account you sign in with, not an API key; no customer-key infrastructure exists. The gate for turning this on: the free signal contract needs enough forward history that a paying customer is not the first person testing it.

Following — agent marketplace

Ship the MCP server and the metered agent API. This is the bet that a meaningful share of future demand for market signals will come from other software rather than people, and that being agent-native from the start is a real advantage over incumbents retrofitting an API onto a human-first product.

Ongoing — engine side

Keep the validation program running: every strategy change stays pre-registered, every forward log stays append-only, and falsified ideas stay documented rather than quietly retried. Grow the dataset library and bring the second research line up to the same standard as the primary engine.

Commitments that hold regardless of phase

The private engine stays off the cloud, permanently. Any future execution layer stays separate from the research engine and gets armed on its own, deliberately. Every public signal surface keeps its entertainment-and-education framing as the product grows.

Not advice

Everything here is educational and entertainment statistics. It is not investment advice, not a recommendation, and not a forecast. The model does not know you or your circumstances. Nothing on this site is a record of real trading; the paper trading desk fills with play money, at the prices your own board shows you.

New here? Start with the plain-English guide