Factor Lab 上的每个实验都是一个已完成的研究对象，带有假设、闸门记录、净表现、硬指标得分或 DSR 背景、容量状态，以及可 Fork 的数据。把你信任的保存下来，Fork 到你的 Build 工作区。只要你一登录，这一切都会同步到 iOS app。Every experiment on Factor Lab is a finished research subject with a hypothesis, gate record, net performance, hard-metric score or DSR context, capacity status, and forkable data. Save the ones you trust, fork them into your Build workspace. It all syncs to the iOS app the moment you sign in.
每张卡片都是一个完成的实验对象：假设、闸门、净夏普、硬指标得分、容量，以及一个清晰的 Fork 操作。先为证据而逛，其次才是灵感。Every card is a finished experiment subject: hypothesis, gate, net Sharpe, hard-metric score, capacity, and a clear fork action. Browse for evidence first, inspiration second.
大多数平台只孤立地给你看一条回测曲线和一个夏普。这里的每个实验也都把统计上的注意事项摆在最前——不是因为结果不好，而是因为这道闸门是动真格的。Most platforms show you a backtest curve and a Sharpe in isolation. Every experiment here leads with the statistical caveats too — not because the results are bad, but because the gate is for real.
FDR 校正后的 p 值和紧缩夏普比率应对多重检验的代价。PBO 给过拟合风险划定上界。没有任何东西能靠运气蒙混过关。FDR-adjusted p-values and the deflated Sharpe ratio pay the multiple-testing bill. PBO bounds the overfitting risk. Nothing sneaks through on luck.
跨多个预测时域的 IC 和 rank-IC、信号衰减曲线、按行情分条件的一致性。我们要这个 edge 先讲得通，才会去信那条曲线。IC and rank-IC across forecast horizons, signal-decay curves, regime-conditioned consistency. The edge has to make sense before we trust the curve.
taker 手续费、随订单规模放大的滑点、合约的资金费率成本、容量估计。如果你的 Fork 用了不同的成本，闸门会重新评估。Taker fees, size-scaled slippage, funding costs on perps, capacity estimates. If your fork uses different costs, the gate re-evaluates.
每个实验都带着它的研究数据。只有生命周期可发布、且目标 App build 完整支持规范化策略的实验才显示 Fork；其余会明确显示 Requires App Update 或 Research Preview。Every experiment carries its research data. Fork appears only when the lifecycle is publishable and the target App build supports the complete canonical strategy; otherwise the surface says Requires App Update or Research Preview.
fork 会把它们全部保留。Factor IDs, parameters, market, timeframe, sizing mode, cost model — everything the backtest engine needs. fork preserves it all.在已实现波动率行情扩张之后，ETH 1h 在扣除资金费率后仍带有持续的方向性动量。原假设：扩张之后的前向收益期望为零。After realized-volatility regime expansions, ETH 1h carries persistent directional momentum net of funding. Null: expected forward return after expansion is zero.
factor_ids: ["ret_1h", "realized_vol_24h", "funding_8h"] method_id: "vol_scaled_momentum_v2" params: { lookback: 24, vol_target: 0.35, decay: 0.92 } symbol: "ETHUSDT" market: "um_futures" interval: "1h" exit: { stop_loss_pct: 0.04, max_hold_bars: 48 } cost_model: { taker_bps: 5, slip_base: 1, slip_k: 25 }
你在这里的每一个操作都与账号绑定。登录一次——同样的保存和 Fork 会在你打开 iOS app 的那一刻就出现。Every action here is account-bound. Sign in once — the same saves and forks appear the moment you open the iOS app.
Me · 已保存 下。用合集整理，按你自己的评分排序。Star an experiment or a whole group. Saves land under Me · Saved on iOS within seconds. Organize with collections, sort by your own rating.
Build · 草稿，血缘完整保留。A fork preserves factor IDs, parameters, market, timeframe, sizing, cost model, metric snapshot, and the evidence panel. It lands in Build · Drafts with full lineage.
JSON 是完整数据载荷（可重新导入）。CSV 是成交记录加每根 K 线的盈亏。报告是一份可分享的 HTML 摘要，带可信度信息。JSON is the full data payload (re-importable). CSV is the trade log plus per-candle P&L. Report is a shareable HTML summary with the trust info.
你浏览归档。这是填满它的自主研究程序——它的实时状态、什么已经通过闸门，以及最近归档的幸存者。只读。没有旋钮。You browse the archive. This is the autonomous research program that fills it — its live status, what has passed the gates, and the most recently archived survivors. Read-only. No knobs.
研究程序在服务端按固定计划运行。网站在设计上是只读的——你浏览并 Fork 幸存者，但从不去调搜索本身。每个幸存者都带着一份可复现的清单。The research program runs server-side on a fixed schedule. The site is read-only by design — you browse and fork survivors, but never steer the search itself. Every survivor carries a reproducible manifest.
Factor Lab 是已验证原材料的来源。Build 则是你打磨它的地方——加约束、用你的成本、你的仓位。两边互通。Factor Lab is the source of validated raw material. Build is where you shape it — with constraints, your costs, your sizing. The two talk to each other.
假设被生成、优化、闸门把关、归档。每周发布新批次。每个实验都能从它的清单复现。Hypotheses are generated, refined, gated, archived. New batches ship weekly. Every experiment reproduces from its manifest.
Fork 带着完整数据载荷到达 Build。改参数、用你的成本重跑、作为你自己的变体发布回去——或推到 Monitor 作为仅观察的信号。Forks arrive in Build with the full data payload. Change params, re-run with your costs, publish back as your own variant — or push to Monitor as a watch-only signal.
无需登录即可浏览。当你找到喜欢的，登录一下，它就会跟着你到手机上。Browse without signing in. When you find one you like, sign in and it follows you to your phone.