Docs

Reference

Whitepaper

Forgeus whitepaper, draft v0 (2026-11-17). The source of truth for product claims.

Note

This page mirrors the whitepaper draft v0. Fields marked TBD are not decided yet.

Status#

Draft v0 (2026-11-17). Fields marked TBD are not decided yet and must not be filled in with guesses: domain, chain and venue for $FOUS, contract address, holder tier thresholds, socials, date for limited real execution.

Name#

Forgeus, domain TBD, ticker $FOUS

Contract#

TBD — published at launch.

Short description#

A self-evolving AI that runs live experiments on on-chain markets. Strategies compete, mutate, and get culled every generation. All results are public and verifiable.

Tagline#

Intelligence under selection pressure.

  • Website: TBD
  • Docs: TBD
  • App: TBD
  • X: TBD
  • Telegram: TBD
  • GitHub: TBD

USP#

Most "AI trading" projects show a curated equity curve, retrain offline, and never publish what failed, so nobody can tell skill from survivorship. Forgeus runs its research in the open: a population of strategies is scored against live on-chain quotes, the weak half is culled every generation, and the survivors are mutated into the next one. Every generation, including the culled strategies and the generations where nothing beat holding, is published as a hash-chained record with its full input data, so anyone can download it and replay the result on their own machine.

Description#

The problem#

AI agents and "self-improving" trading bots are everywhere, and almost none of them can be checked. Backtests are picked after the fact, failures are deleted, and the only public number is the best one. A trader who wants to know whether a system actually learns has nothing to inspect.

What Forgeus does#

Forgeus is an experimental intelligence system that improves through continuous selection against real on-chain data. It keeps a population of specialist strategies. Each generation they are evaluated against live market conditions, the underperformers are eliminated, and the best are retained, mutated, and given more offspring in the next generation. The whole process is published as it happens: generation logs, the paper trading tape, mutation logs, lineage, and null results.

How it works#

  1. Quotes. Every minute the engine reads the price of fixed Uniswap V3 pools on Base (WETH/USDC 0.05% and USDC/cbBTC 0.05%) directly from the pool's slot0 at a recorded block number. Anyone with a Base archive RPC can re-read the same value at the same block.
  2. Population. Each strategy is a small "circuit": a signal (momentum, mean reversion, breakout or trend), a lookback, a threshold, a position size and a stop. A hold baseline per market is evaluated alongside as the bar to beat.
  3. Evaluation. A generation covers a fixed window of quotes (60 per market). Every strategy is paper traded over the window with a stated cost model (5 bps per unit of position change, equal to the pools' fee tier). Fitness = return minus half the maximum drawdown.
  4. Selection. The bottom half is culled. Survivors carry over. The top survivors get two offspring each, the rest one. Offspring are mutations of their parent (parameter changes, sometimes a new signal type).
  5. Records. Each generation is written as a canonical JSON record (inputs, population, results, culls, offspring, seed) and hashed together with the previous record's hash. The randomness for mutation is seeded from that previous hash, so the next generation is fully determined by public data.
  6. Null results. When no strategy beats both zero and its market's hold baseline, the generation is marked null and published like any other.

Paper and real. All evaluation today is paper evaluation against live quotes. No funds are traded. A strategy that meets the predefined evidence thresholds becomes "eligible" for limited real execution under strict risk controls; real execution is not live, and when it is, its results are reported separately. Simulated and real results are never mixed.

Why trust it#

  • Inputs are on-chain: every quote names the pool and block it was read at.
  • Outputs are reproducible: the evaluation engine is deterministic and the same code runs in the browser replay tool, which recomputes a generation from its archive and checks the hash.
  • The record is append-only: generation hashes chain to the previous one, so a deleted or edited generation breaks the chain.
  • Failures are kept: eliminated strategies, mutation logs and null generations are public and downloadable.
  • What is not claimed: no audit, no partners, no live trading, no performance promise. Paper results are not a forecast.

Business model#

TBD. Candidates: paid higher-rate API access and research exports for teams. Nothing is charged today; the public data and API are free.

The token#

$FOUS is not launched. Default plan (to be confirmed): fixed supply, fair launch, no presale or team allocation, no staking and no emissions. Holding $FOUS would unlock early access to new circuit docs and weight in research-direction governance. Thresholds are TBD. Until launch, governance is a one-wallet-one-vote signal and early access is off.

Utilities#

  1. Live generation logs & dashboards: see the current generation, the population, and the last results as they happen.
  2. Active / challenger / eliminated strategy views: filter strategies by whether they survived, were just born, or were culled.
  3. Lineage explorer: follow any strategy back through its parents to the genesis population.
  4. Public paper trading tape: every paper position change with its price, quote and block.
  5. Circuit visualization: see a strategy's signal, lookback, threshold, size and stop as a diagram.
  6. Evidence threshold tracker: see how close each strategy is to the thresholds for limited real execution.
  7. Downloadable generation archives: download any generation with all of its input quotes and outputs.
  8. Structured data API: read generations, strategies, trades, quotes and stats as JSON with no key.
  9. Generation completion notifications: get a signed webhook, an RSS item or a browser notification when a generation closes.
  10. Research notebook export: download a generation as a Jupyter notebook with the data embedded.
  11. Paper vs limited real-execution comparison: paper and real results side by side, never merged (real is not live yet).
  12. Strategy composition breakdown: the population by signal type and market, generation by generation.
  13. Historical cull & survival statistics: culls, survival rates and lifespans across all generations.
  14. Mutation logs: every offspring with the exact parameter changes from its parent.
  15. Null-result repository: every generation where nothing beat holding, with the reason.
  16. Community annotations on generations: signed-in wallets can leave notes on any generation.
  17. Priority access to new circuit docs: $FOUS holders read new circuit docs first (after launch).
  18. Governance on research direction: proposals and signed votes on what the system explores next.
  19. Independent replay / verification tools: replay a generation in your browser and check its hash and the chain.
  20. Public experiment data access: bulk export of all quotes and generations as CSV or JSON.

Competitors#

  • AI trading agents and vaults (AI agent tokens with trading claims): show results, rarely inputs. Forgeus publishes the inputs, the losers and the code path to reproduce every result.
  • Strategy marketplaces and copy trading: rank by past returns with survivorship bias. Forgeus keeps culled strategies and null generations in the public record.
  • Prediction and forecasting networks: reward forecasts scored offline. Forgeus scores against live on-chain quotes on a fixed public schedule.

Open items (TBD)#

  • Domain, X, Telegram, GitHub
  • Chain and launch venue for $FOUS, contract address
  • Holder thresholds for early access and governance weight
  • Evidence threshold values (initial values are in the docs and may change, changes are logged)
  • Risk limits and start date for limited real execution
  • Business model and pricing for higher-rate API access
  • Third-party review of the engine

Paper only. No funds are traded. $FOUS is not launched. Nothing here is financial advice.