Glossary

Forven uses precise words for precise things. This is the A–Z reference. Where the marketing site uses a simpler name, the in-app term is the heading and the public synonym is noted in the definition.

A quick orientation first. The strategy pipeline runs quick_screen → gauntlet → paper → live_graduated, and strategies that didn't make it go to the graveyard (archived, rejected, backtest_failed). The public site compresses this to screen → gauntlet → candidate → live, where "candidate" means the paper stage.

Numbers below (thresholds and defaults) are the shipped defaults at the time of writing; they are tunable and illustrative of the discipline, not promises about outcomes. Forven is a research tool. Nothing here is financial advice.

A

Agent — One of Forven's six optional AI workers: the Brain (the coordinator), the Strategy Developer, the Simulation Agent, the Risk Manager, the Quant Researcher, and the Full-Stack Engineer. Agents never task each other and cannot place, size or close positions. They run only on models you have connected and selected. See Agents.

Agent run — One unit of agent work (writing an idea, building a strategy, validation, a post-mortem, a diagnosis), with a status, a priority, and a full transcript. Listed under Agents → Runs; older pages call it a task. See Tasks.

AI Drop Zone — The workspace and session model an external AI client uses to write, register, and test strategies. Every call made through it passes the same gates. See AI Drop Zone.

Approval — A pause on a transition or an agent action until you decide. By default gauntlet → paper is automatic, paper → live needs your typed GO LIVE with a per-asset ceiling, and code edits wait for you. Each approval category has a mode: manual, smart (a classifier model approves high-confidence items), or off. See Approvals.

Archived / rejected — Graveyard states for strategies that stopped earning their place or failed a gate. You can recover them to quick_screen.

Assistant — The in-app chat. It answers questions from live data and can propose actions — which you confirm on a card before anything changes. Each conversation has a spend cap ($5 by default). See The assistant & deep-dive.

Autonomy mode — See system mode.

B

Backtest — A run of one strategy over historical data through the execution kernel, split 70% in-sample / 30% out-of-sample, with fees, slippage, and funding applied. Research backtests end at the held-back test's cutoff. See Backtesting.

Baseline hurdle — A check on every walk-forward: the strategy's out-of-sample returns, after removing what buy-and-hold of the same asset and a fixed trend-following rule made over the same days. It is measured but doesn't block at gauntlet → paper, and is enforced at paper → live, where a fail keeps the strategy in paper. Shown as Beats baselines. See Promotion gates.

Bot Factory — Autonomous AI-driven paper-trading bots, each with its own persona and its own process; on Windows they run in a low-integrity sandbox by default. See Bot Factory.

Brain — The single hub-and-spoke coordinator: deterministic routing and policy enforcement. It receives all agent output, drives stage transitions, and enforces the gates. It keeps short working notes (capped at 2,000 characters) and a decision log. See The brain.

Breadth test — A test that runs a strategy's frozen code and parameters, unchanged, on 15 coins over sealed research data. It reads General, Home coin only, Mixed or Too few coins. The verdict ranks the paper queue, is required for live by default (any verdict but Home coin only), and picks the coins of a multi-coin paper strategy. The paper pool's hourly job starts it; you can also run it from the strategy page's Robustness tab. See The gauntlet.

C

Candidate — Public synonym for the paper stage.

Capital slice — In live trading, each live strategy sizes inside an equal slice of the account, so one strategy can't consume the whole book. Slices shrink to fit when the wallets can't hold every live strategy's margin at once. See Risk.

Circuit breaker — A CLOSED → OPEN → HALF_OPEN state machine per exchange dependency. Real outages trip it; a clean probe closes it again. Distinct from the kill-switch. See Circuit breakers.

Control strip — The bar at the top of every page: connection status, live P&L, regime chips for BTC, ETH and SOL, Mode: PAPER (or Live ×N when live strategies or armed bots exist), the TESTNET/MAINNET badge when the network is known, Trading Allowed or Halted, the MANUAL / SEMI / AUTO system mode, and the Emergency Halt and Reset Halt buttons. See Dashboard.

Cost stress — A gauntlet test that replays the baseline backtest's own window at 2× fees and slippage. The edge has to survive the higher costs.

Crucible — Retired. Crucibles and hypotheses were replaced by ideas and one strategy-creation loop.

D

Daemon — The long-running loop, hosted in the backend, that streams prices, enforces the hard risk controls, and reconciles exchange state. See The research daemon.

Daily-loss halt — When same-day loss passes max_daily_loss_pct (2% by default), new live entries are blocked for the rest of the UTC day. It clears itself at UTC midnight. Paper is never blocked.

Data lake — Forven's local store of market data — candles plus enrichment streams — as Parquet files. See Managing market data.

Decay kill-switch — A live-stage guard that retires a live strategy whose Sharpe degrades by more than decay_kill_switch_pct (30%) over its rolling window.

Deflated Sharpe ratio (DSR) — The probability that a Sharpe is genuine after accounting for how many variants were tried, including failed sibling strategies. Always computed; the Deflated Sharpe gate is on by default, with a bar of 0.90. The research door records it at gauntlet → paper, where it ranks the paper queue, and the paper → live gate enforces it on the backtest plus forward trades. Under Strict paper it rejects at gauntlet → paper. See Metrics.

Demotion — A strategy sent from the gauntlet back to quick_screen. The third demotion archives it as untestable (untestable:max_demotions).

Dethrone — Replacing a weaker strategy with a stronger challenger on the same market. A new paper strategy is protected for a soak period first. Dethrones are approved automatically by default. See Promotion gates.

Diagnostics — The page of point-in-time health checks, the 24-hour cost roll-up, MCP server status, resumable runs, and the notification inbox. See Diagnostics.

Direction × regime gate — A system-wide veto on entries whose direction fights the market regime (by default: longs in TREND_DOWN or HIGH_VOL, at confidence 0.6 or more). Modes: off, observe (the default — log only), enforce. See Market regimes.

Door into paper — The gate between the gauntlet and paper: the research door by default, or Strict paper. Set in Settings → Lab → Door into paper. See Promotion gates.

E

Enrichment streams — Market data beyond candles — funding, open interest, long/short ratios, taker flow, liquidations, sentiment, macro — merged into the backtest frame without look-ahead. See Data sources.

Execution check — A panel on paper and live strategies whose settings no longer match what was validated. It shows a Setting / Validated / Now table and offers Restore validated settings, or, for live strategies, Accept a backtest as the live baseline (a reason is required). It appears under the strategy header and in the trading desk's Why panel.

Execution kernel — The one piece of code that turns signals into positions — entries, exits, stops, sizing, fees, slippage, and funding — shared by backtest, paper, and live. See Execution modes.

Execution mode — paper (simulated fills on live market data; the default and supported mode) or live (real Hyperliquid orders; unsupported, and only settable through the environment).

Execution settings — Stops, take-profits, trailing stops, and sizing. Kept separate from strategy parameters by design: a risk field inside params is ignored by the engine.

F

Fair-sample quick screen — The default quick-screen rule. The screen window widens (doubling from the 730-day default up to 1,460 days, data permitting) until it holds 20 trades, and the whole window is judged: Sharpe above 0, profit factor at least 1.0, a non-negative return and drawdown at most 30%. The out-of-sample slice only has to avoid losing badly (Sharpe above −0.5), and the in-sample/out-of-sample gap and ratio are recorded, not enforced. Strict quick screen restores the fixed window and the split in-sample/out-of-sample checks. See Promotion gates.

Fitness score — A 0–100 composite used to rank strategies: Sharpe (30%), win rate (20%), profit factor (20%), max drawdown (15%), trade count (15%), with penalties for a weak shape. See Promotion gates.

The Forge — The strategy lab (/lab): every strategy in the pipeline and the graveyard, a Pipeline rail with counts and pass rates, + Submit idea, and a peek drawer for each strategy that links to its full page. See The Forge.

Funding flags — funding_applied and funding_complete on a backtest record whether funding was charged and whether its data was complete. The capital gate refuses funding-blind evidence.

G

Gate — A check a strategy must pass to change stage, also called a promotion gate. The door into paper is deliberately achievable: by default it stops strategies that are broken, leaky or clearly losing, and records the selection-bias checks. The paper→live gate is strict and enforces them. See Promotion gates.

Gauntlet — The robustness battery before paper: walk-forward analysis and parameter jitter are required by default, with cost stress, Monte Carlo, and regime split alongside, distilled into a robustness score. Every test must save a verdict before paper; under the research door a FAIL is recorded and enforced at paper → live. See The gauntlet.

Generation pause — Freezes the jobs that create and test new strategies (strategy creation, auto-intake and the testing cycle) while scanning and paper keep running. An API control.

GO LIVE — The typed confirmation, together with a required per-asset notional ceiling, that a promotion into live needs. The only way around it is the Allow automatic go-live setting, which is off by default and needs a typed LIVE to save.

Graveyard — Where strategies that left the pipeline go: archived, rejected and backtest_failed. The Forge's Graveyard bucket groups them by cause (for example Held-back test, Untestable, Losing edge) and can recover them to quick_screen.

H

Held-back test — Settings calls it the research holdout. The most recent market data, one quarter back (3–6 months) by default, is sealed away from all research; each new strategy gets one walk-forward on it, and by default a pass is required. It runs as the last check when the strategy applies for live (paper → live) by default, or before paper under Strict paper. Strategies created before it was switched on are exempt. See Promotion gates.

High-water mark (HWM) — Peak equity since the last reset. Drawdown is 1 − current_equity / HWM. Resetting the kill-switch re-baselines it.

Hyperliquid — The only exchange the (unsupported) live path can route to. Its testnet is where the live order path is verified, by a daily test harness. Paper never touches it. See Hyperliquid.

I

Idea — The starting point of every new strategy: a thesis about what the market does, the mechanism behind it, and what would disprove it. The Strategy Developer writes ideas up and builds strategies linked to them; you submit your own with + Submit idea in The Forge. A strategy's Idea chip opens the idea and its other strategies. See Ideas.

Implausible-metrics reject — Sharpe ≥ 5 or profit factor ≥ 8, in-sample or out-of-sample, is rejected as a likely look-ahead leak rather than celebrated.

In-sample (IS) — The first 70% of a backtest's data. Optimistic by construction; trust out-of-sample instead.

K

Kill-switch — The drawdown circuit breaker for real capital. When drawdown from the HWM reaches max_drawdown_pct (30% by default), it closes every open live position and halts new live entries until you reset it. Paper is never affected. See Risk.

L

Live-evidence test — See sequential test.

live_graduated — The live stage: a strategy placing real orders, after the strict gate and a typed GO LIVE. Public synonym: live. Live/mainnet is not a supported feature.

M

Manual Backtest — The page (/backtest/new) for running any strategy against history with settings you choose. Blank fields resolve the way the strategy does. See Backtesting.

MCP server — Forven's Model Context Protocol server: 21 tools that let an external AI client design, backtest and validate strategies through the real gates.

Monte Carlo — A gauntlet test that resamples a strategy's realized trades to measure tail risk. It tests the stability of the trade sequence, not whether the edge is real — that's what walk-forward, regime split, the Deflated Sharpe gate and the held-back test are for.

Multi-coin paper strategy — A paper strategy whose breadth test reads General also paper-trades the coins where its unchanged rule made money in that test (home coin first, up to 6 coins), as one record. Each coin sizes on an equal share of the strategy's paper equity, and trades closed the same UTC day count as one day of evidence. The coin list is chosen by the breadth test and frozen. Paper only: going live with several coins is held (multi_coin_live_review) and the strategy keeps paper trading. Strict paper turns it off. See Promotion gates.

O

Operator — You. Operator-gated actions (promotions, risk and system controls, settings) need the operator key when one is set.

Orphan strategy — A strategy whose type has no registered class. The backtest refuses to run it rather than silently produce zero signals.

Out-of-sample (OOS) — The last 30% of a backtest's data, unseen when the signals were fitted. The number the gates read, and the one you should read first. See Metrics.

Overfitting guardrails — The quick-screen checks. Under the default fair-sample rule: a trade floor, Sharpe, profit-factor, return and drawdown floors on the whole window, an out-of-sample slice that mustn't lose badly, the high-win-rate trap and the implausible-metrics reject. Under Strict quick screen: separate in-sample and out-of-sample floors, an IS/OOS gap and ratio ceiling, and a profit-factor floor on both slices.

P

Paper pool — The managed set of paper strategies: at most 40 at once, 3 per market and 10 per strategy family by default (0 = no cap). A strategy that passes the door while the pool is full waits in the paper queue; paper strategies whose forward record shows them losing are retired. Set in Settings → Lab → Paper pool. See Promotion gates.

Paper queue — Where a strategy that passed the door waits for a paper slot (paper_queue_full), still in the gauntlet stage. It is ranked by the evidence the door recorded: deflated Sharpe, robustness score, out-of-sample Sharpe and the breadth verdict. Waiting is capacity, not a rejection, and never counts toward the auto-archive.

Paper trading — Forward trading on the live Binance feed with fills simulated inside Forven and no capital at risk. The public site calls it the candidate stage. Real-capital halts never block it. See Paper trading.

Parameter jitter — A gauntlet test that nudges the chosen parameters by up to ±10% and re-runs the backtest many times; enough reruns must still pass, which exposes parameter cliffs.

Phantom position — A filled exchange position with no matching local record. Recovery creates the missing record, unlinked to any strategy.

Pipeline — The whole strategy lifecycle, and the page (/pipeline) that explains why each strategy is where it is. See The pipeline.

Profit factor (PF) — Gross profit ÷ gross loss.

Promotion gate — See gate.

Provenance — Every backtest and verdict records the engine version, code identity, parameters, and a fingerprint of the data it ran on, so results stay auditable and stale evidence can be detected.

Q

Quant skill — A versioned, confidence-scored research insight — what works and what doesn't, for a regime, family, or failure mode — that feeds back into new ideas. See Quant skills.

quick_screen — The first gate: cheap triage before the expensive gauntlet. By default it judges a fair sample, the whole window widened until it holds the trade floor, and it rejects implausible metrics. Public synonym: screen.

R

Regime — A market-condition label — TREND_UP, TREND_DOWN, RANGE_BOUND, or HIGH_VOL — from ADX, EMA alignment and ATR ratio. The app's chips read TREND UP, TREND DOWN, RANGE and HIGH VOL, with TRANSITION marking a stretch the model is unsure about. A strategy can declare the regimes it is built for. See Market regimes.

Regime split — A gauntlet test that splits trades by regime and requires profit across enough of them, proving the edge doesn't depend on one market state.

Research door — The default door into paper. It admits a strategy that isn't broken or clearly losing: 20+ confirmation trades, a non-negative return, an out-of-sample profit factor of at least 1 and Sharpe above 0, the drawdown, implausible-metrics and Monte Carlo tail-drawdown floors, and no loss under 2× costs. The deflated Sharpe, walk-forward fold rate, regime split, parameter jitter, Monte Carlo and cost-stress verdicts and the robustness score are recorded there and enforced at paper → live. See Promotion gates.

Research loop — The scheduled jobs that keep the pipeline moving without you. In Auto mode the strategy-creation loop asks the Strategy Developer for new ideas and strategies within a daily budget; in Semi and Auto, testing jobs drive candidates through the gauntlet and retire losers. Manual mode freezes it all. See The research daemon.

research_only — Retired. The old "Parked" lane survives only as a compatibility alias; strategies that can't be fairly tested are now archived as untestable.

Retirement — Archiving a strategy that stopped earning its place. The paper pool's hourly job retires a paper strategy that is clearly losing (15+ closed forward trades, a negative average and a t-stat below −1), reaches a 15% forward drawdown, goes 60 days without a trade, or fails its sequential test; it is archived as paper_retired:<why>. A live strategy is retired by the decay kill-switch instead.

Robustness score — The gauntlet's 0–100 composite of how consistently a strategy held up across its tests. The bar is 30 by default: recorded at the door into paper, enforced at paper → live (at the door under Strict paper).

Routine — A scheduled agent workflow you define: a prompt for the brain, a schedule, a tools context, and optionally a Discord channel for the result. See Routines.

S

Safety floors — Absolute rails clamped onto the promotion gates, so no preset or custom value can soften them past a minimum. Editable in Settings. See Promotion gates.

Scanner — The loop that runs paper (and, if wired up, live_graduated) strategies forward on every bar and hands their signals to the execution kernel.

Sequential test — The default paper → live evidence test, also called the live-evidence test. Once a paper strategy has 20 closed forward trades and 30 days in paper, it asks whether the forward trades track the backtest, whether backtest and forward evidence combined clear a bar corrected for how many strategies were in paper, and whether the forward Sharpe is positive. It reads collecting, continue, go or fail. Go never promotes on its own; a fail retires the strategy from paper. Fixed floors (pre-plan) restores the old forward minimums. See Promotion gates.

Sharpe ratio — Risk-adjusted return, computed from per-bar returns of the equity curve (flat bars included) and annualized. Flagged unreliable under 20 trades.

SLA tier — How fresh Forven keeps a data series, set by what uses it: live, paper, pipeline, research universe or idle. Each tier allows a number of missed bars, from 1 for live data to 24 for idle series, and a series well past its allowance counts as a breach. See Managing market data.

Stale evidence — A verdict produced by an older engine or on different data. It is marked stale and re-validated automatically rather than trusted. See Promotion gates.

Stance preset — One setting that sets every gate threshold: Relaxed, Default, Strict, or Custom. The Strict stance only moves thresholds; it isn't Strict paper or Strict quick screen. See Promotion gates.

Strategy Creator — The page for building strategies as Rules or in Python, with an AI assist panel that can draft a strategy or propose a change, a chart preview, in-sample vs out-of-sample vitals, a stress test, and Send to Forge →. See Strategy Creator.

Strategy-creation budget — The daily cap on autonomous strategy-creation tasks (Daily strategy-creation budget, 40 a day by default). Ideas you submit don't count against it. Formerly the research budget.

Strict paper — The door-into-paper option that restores the door from before the October 2026 gate plan: every check blocks paper, the held-back test runs before paper, a missing source-reconciliation reading blocks, and multi-coin paper strategies are off. It doesn't change the live gate's evidence rule. See Promotion gates.

System mode — How much runs on its own: Manual (the fresh-install default; every autonomous job frozen), Semi (the pipeline runs what you submit, but agents write no ideas of their own), or Auto (agents also write ideas within the daily budget). Set in the control strip. See Autonomy modes.

System pause — A halt on new live entries. It doesn't freeze research and never blocks paper.

T

Throughput preset — One dial — Trickle, Conserve, Balanced, or Max — over agent cadences, worker counts, and the daily strategy-creation budget (4, 12, 40 or 120 tasks a day). See Cost controls.

Trade mode — A strategy's direction: long_only (the default), short_only, or both. long, short and long_short are accepted as shorthands.

Trading desk — The layout of the Paper Trades and Live Trades pages: a strategy rail, a chart with fills and signals, a market strip, Position / Why / Performance / Details tabs, a Manual order ticket, and a blotter of fills, decisions and risk. See Paper trading.

U

Universe books — An experimental, paper-only feature: one pre-registered rule held across 15 coins at once and rebalanced at each daily close. Hidden unless you turn on the portfolio layer in Settings → System → Experimental features. See Universe books.

Untestable — A strategy that couldn't be fairly tested (no trades in the screen window, code that won't run, too little history) is archived with an untestable:<code> reason. It isn't a merit verdict, so it skips failure post-mortems; The Forge marks it Untestable.

V

Verdict — The judgement on a test or a strategy. Each gauntlet test records pass, fail or blocked (an error is a non-result, never a verdict). The strategy page's verdict card sums a strategy up as Viable, Promising · not proven, Not viable or Not enough evidence.

Vectorized path — A strategy's all-at-once signal method, much faster than bar-by-bar on long windows. See Writing custom strategies.

W

Walk-forward analysis (WFA) — Out-of-sample validation over rolling folds: fit in-sample, judge out-of-sample, fold after fold. In the gauntlet the whole unseen part of the window is tiled into folds. The cornerstone of the gauntlet.

Workspace — Two things: the agents' markdown files in FORVEN_HOME/workspace (identity, your profile, tool docs, journals), and the folder where strategy files are written for registration (forven/strategies/custom/).

Z

Zombie run — An agent run that stopped making progress. Forven flags it so the stuck work can be cleared or resumed; Diagnostics lists resumable runs.

---

A note on numbers: every threshold and default above is configurable and reflects the shipped defaults, not a guarantee. Forven is a research tool for building and killing strategies before real capital is involved. Out-of-sample survival is evidence, not a prediction, and nothing here is financial advice.

Related

• The pipeline — how the stages fit together
• The gauntlet — the robustness battery in detail
• Ideas — where new strategies come from
• The brain — the coordinator and the agent layer