Build Log · Research

Built in plain sight.

Every strategy change we ship — the idea, the backtest, and the honest caveats, including what we don't yet know. The operator trades these signals with real money, so the homework is public too.

Note. Backtests below are historical simulations, not live results. They are how we decide what to ship — not a promise of what it will do. New strategies are flagged as such and run small until they prove out live.

Capitulation buy — buy blood, not a falling knife

A contrarian book for the crashes our trend filter is built to sit out.

The problem

RegimeEdge sizes down — to zero — in high-volatility STRESS, and our trend filter blocks longs in downtrends. That's the discipline that protects capital. But it also means that when a market is selling off hard, the bot does nothing — it sits in cash and waits. Good for not losing money; it leaves the recovery on the table.

The idea

Add one narrow, contrarian entry that only wakes up in a crash: buy when there's blood in the streets — but don't catch a falling knife. Three things must line up:

Blood  — price at least 15% below its 200-day average (deep capitulation, not a normal dip).
Fear  — the Fear & Greed index in its Extreme-Fear band.
Not a knife  — price has already bounced ≥3% off the low (it has stopped falling before we step in).

What the backtest said

BTC + ETH, daily, 2019–2026. The guard is the whole point:

−90%
Buying deep dips raw (no guard)
+58%
+ bounce guard (wait for the turn)
+152%
+ extreme-fear confirm
52%
win rate, full version

Catching the knife loses (−90%). Waiting for the bounce flips it positive (+58%). Adding the fear filter compounds it (+152%). The edge isn't "buy the dip" — it's "buy the dip after it stops dropping."

What we deliberately checked — and changed

We re-ran it on the 1-hour timeframe the bot trades on. The bounce guard held (win rate 40% → 73%), but the fear filter did not transfer to a short-horizon average. So the signal evaluates blood against the daily 200-day average, where the evidence is strongest — not the intraday one. We changed the design because the data told us to.

Honest caveats

Status

Enabled live on BTC/ETH, 2026-06-03. Reversible with a single config flag. When it first fires, the entry, exit, and P&L land on the Track Record like every other trade — win or lose.

This is what "built in public" means.

We publish the wins, the losses, and the homework — including the backtests that talked us out of things. Watch the receipts.

See the live track record →