Philosophy

Forven is opinionated about a few things. Reading this page is the fastest way to tell whether the product is for you. Everything in the app — the pipeline, the gauntlet, the gates, the agents — exists to enforce the beliefs below, not to decorate them.

The only durable edge is process

A retail trader cannot out-speed, out-inform, or out-capitalize an institution. The one edge that remains is enforcing discipline on your own research process more aggressively than most people are willing to.

So Forven makes the discipline structural instead of optional. A strategy does not become tradeable because it looked good once. It earns each stage by passing the gate in front of it:

researching → backtesting → quick_screen → gauntlet → paper → live → retired

(The public site simplifies these names to screen → gauntlet → candidate → live; "candidate" there means the paper stage. The app uses the real names, and so does this wiki.)

You cannot skip a stage, and you cannot bribe one. The gates are arithmetic, not vibes.

In-sample metrics lie

Almost every published Sharpe ratio you have ever seen is computed in-sample. Almost every backtest that "works" across the full history is overfit to that history. A curve that only ever climbs is usually a curve that has read tomorrow's prices.

Forven treats in-sample numbers as a hypothesis, not a result. The backtester splits each run into in-sample and out-of-sample data and reports the two separately. When you read a strategy, look at the out-of-sample column. The gauntlet's walk-forward analysis pushes this further: it re-tests on data the parameters never saw, and a strategy is allowed to decay only so far before it fails.

The platform is also openly suspicious of good numbers. Metrics that are too clean trip lookahead auto-rejects — a profit factor at or above 8.0, or a Sharpe at or above 5.0, are treated as evidence of a future-bar leak (think a stray .shift(-1)) and the strategy is rejected rather than celebrated.

These ceilings are illustrative guardrails, not targets. A high score is not a promise of profit — it is a reason to look harder for a bug.

Kill your darlings

You will build strategies you are attached to. Most of them will fail, and the ones that fail quietly are the expensive ones. The people who survive this game kill a strategy the moment its out-of-sample edge disappears — not the ones who give it "one more week."

Forven automates the killing so your attachment never gets a vote:

• quick_screen rejects candidates on basic math before they waste compute — negative in-sample edge, drawdown over its ceiling, an in-sample/out-of-sample ratio that screams overfitting, or too few trades to mean anything.
• The gauntlet discards anything that cannot survive walk-forward, parameter jitter, and cost stress.
• In research, a hypothesis is judged disproven by a verdict math floor, and the floor overrides any softer opinion. An LLM verdict cannot rescue a strategy the arithmetic has already failed.
• A graduated hypothesis's best child per cell is marked canonical and protected — but only so the next idea has to genuinely beat the canonical, not so a favorite is spared scrutiny.

You can override a kill. The app will let you. It will also argue with you, and the audit trail will remember that you insisted.

The agent is a researcher, not a trader

Forven's agents propose strategies, write and debug code, run simulations, and explain their reasoning. They do not have capital-allocation authority. The orchestrator — the brain — routes their output through the same gates a human would face; the agents never place trades directly and never task each other.

Crossing from paper to live is a separate, deliberate act. It requires an explicit human approval gate by default, and live/mainnet is unsupported regardless — paper (optionally on the HyperLiquid testnet) is the supported mode. The execution mode lives in one setting, FORVEN_EXECUTION_MODE, and it defaults to paper:

# Paper is the default and the supported mode.
$env:FORVEN_EXECUTION_MODE = "paper"

Autonomy is a dial you turn, not a default you discover. If automated capital allocation ever ships, it will be an explicit toggle that defaults off.

Transparency over confidence

Credibility here comes from showing the work, never from a performance claim. The lab exposes every metric, every trade, every parameter change, and every stage transition is logged for full lineage. Beta rough edges are surfaced honestly rather than hidden — the kill-switch can flatten positions on its own, the hypothesis pool silently evicts its weakest member under pressure, regime classification can lag a fast market by a few minutes. None of that is a secret, because a tool you cannot audit is a tool you cannot trust.

Forven would rather look uncertain and be honest than look decisive and be wrong.

What this means for you

If you want a system that confirms your conviction, this is the wrong tool. If you want one that tries to disprove every idea you have — including the ones you like — and only lets the survivors through, you are in the right place.

A note on results. Forven is a research tool. Out-of-sample survival is evidence of robustness, not a prediction of future returns. Any numbers in the app or in these docs are illustrative. Nothing here is financial advice.

Related

• The pipeline — the full lifecycle these beliefs are built into.
• The gauntlet — the robustness battery that makes discipline unavoidable.
• Agents — the researcher-not-trader agent layer and its guardrails.