A “strategy” in Scienter today is a filter over signal kinds plus a conviction bar that you can backtest against the real emitted record — not a full expression language that places trades for you. This guide builds one from scratch.

​
1. Start from a question, not a signal kind

The eight prebuilt strategy baskets in the Farcaster mini-app are a useful reference for how this is meant to work — and for how honestly a “basket” should report what’s real. Of the eight, only two (“Blue Chip Momentum,” “Funding Rate Reversion”) are fully verified against real backtest data at the time of writing; the rest are explicitly labeled partial or pending in their own UI copy rather than showing an invented number. That’s the standard to hold your own strategy to.

​
2. Pick your signal kinds

The four independent groups are funding, liq (both read Hyperliquid positioning — count as one group), netflow (on-chain flow), and whale (scored wallet buys). Combining two signals from the same group manufactures conviction rather than adding real corroboration — see Signal tiers.

​
3. Backtest the combination before trusting it

Read three things before you believe the result:
  • biased — did you ask for a conviction floor below what actually gated live emission? If so, the sample is smaller and more selective than the numbers suggest.
  • One horizon only. Every outcome is marked at the ensemble’s fixed horizon (24h). There’s no stored price path for testing a different holding period.
  • Replay ≠ prediction. /edge/replay.json shows what was actually emitted on a date, using data available then — a genuinely different exercise from asking “what would this filter say about history today,” which the current evidence windows (funding expires in an hour, liq in 90 minutes) make impossible to reconstruct honestly.

​
4. Check per-kind decay before committing to a kind long-term

A kind that’s been auto-deprecated (status: deprecated) is being zeroed out of live ensemble weighting already — building a strategy around it is fighting the system’s own measured judgment.

​
5. Express it as a rule (parsed, not executed)

The rules DSL parses a user-defined strategy expression and validates it — unknown predicates are rejected at parse time rather than silently treated as always-false, so a saved rule that validates is a rule the engine actually understands. It does not automatically fire trades. Treat it as a structured way to save and share a filter definition, not an execution layer.

​
What this can’t do yet

  • No named technical-analysis strategies (Wyckoff, SMC, mean-reversion as labeled concepts) — only combinations of the four raw signal kinds.
  • No cross-horizon testing — everything is pinned to the 24h ensemble horizon.
  • No automatic rule → trade wiring. If you want execution, that’s a separate, explicit step covered in Non-custodial execution.
Scienter publishes impersonal market signals of general and regular circulation. Nothing here is investment, legal, or tax advice, or a recommendation to buy, sell, or hold any asset. Scienter is not a registered investment adviser or broker-dealer. Trading digital assets can result in the total loss of your funds — see Disclaimers for the full text.