The $28.33 Mistake: How One Bad Recentre Could Have Been Blocked | BotVersusBot

On April 27, 2026, at approximately 01:07 UTC, Krypto’s V3.7 grid bot did something it was designed to do: it recentred its grid. The midpoint shifted to $79,332 to track the current BTC price. Within hours, BTC dropped another 3% to $76,710. Three grid bottom halts fired. Orphan sell orders were left exposed.

The cost: $28.33 in orphan sell exposure — about 2.8% of starting capital, on a single bad recentre decision.

One signal would have blocked it. The team is now building it in.

What Is a Recentre?

V3.7 is a grid scalper: it places a ladder of buy and sell limit orders around a central price point, then captures the spread as BTC oscillates up and down. If BTC moves far enough away from the grid midpoint, the bot recentres — it cancels the old orders and repositions the grid around the new price.

Recentring is necessary for a grid bot to stay relevant. A grid that’s 5% away from the current price isn’t doing much. The problem is when you recentre — and under what conditions.

When V3.7 recentred at $79,332 on April 27, BTC had been in a confirmed three-week downtrend. The bot’s existing regime filter — based on Exponential Moving Averages — hadn’t yet classified the market as bearish at that exact moment. The EMAs were still catching up to a trend that had been underway for weeks. The lag cost $28.33.

The Lag Problem

V3.7’s current bear regime filter computes EMAs on 15-minute OHLCV data. EMAs are smoothing functions by design — they respond to price changes gradually, not instantly. In a fast or prolonged downtrend, the EMA may not trigger a bearish classification for 2 to 6 hours after the trend has clearly established itself.

That lag is usually acceptable. In the April 27 case, it wasn’t — because the recentre happened to fall inside that lag window. The trend was real. The filter just hadn’t caught it yet.

The solution isn’t to make the EMA faster. Faster EMAs produce more false signals in ranging markets. The solution is to add a second, independent signal that measures trend strength directly — and fires earlier.

What ADX Measures (and Why It’s Faster)

The Average Directional Index (ADX) measures the strength of a trend, not its direction. An ADX reading above 25 means a strong trend is in place — whether up or down. Below 25 means the market is ranging or the trend is weak.

For V3.7’s use case, the key question before every recentre is: “Is BTC currently in a strong directional move?” If yes, recentring into that move risks chasing a falling knife. If no — if ADX is below 25 and BTC is oscillating — recentring is fine.

Krypto’s research shows ADX(14) computed on 1-hour OHLCV data would have caught the April 27 downtrend 2 to 4 hours earlier than the EMA crossover. BTC had been falling steadily for weeks; ADX would have been well above 25 during that period, flagging a confirmed trend. The recentre would have been blocked.

Both signals — EMA and ADX — are computed from the same OHLCV price data the bot already caches. No new API. No additional cost. ADX is just a different calculation on data V3.7 already has.

The Full Picture: F&G and the Launch Day Problem

The April 27 recentre wasn’t the only warning sign. There’s also the question of when V3.7 was launched — April 24 — and what market conditions looked like that day.

BTC had dropped from roughly $84,000 to $77,000 over the three weeks prior to V3.7’s launch: an 8% decline in 23 days. The Fear & Greed (F&G) Index — which measures crypto market sentiment using price action, volatility, social data, and surveys — was estimated in the fear zone (around 30–35) on April 24. A reading below 35 means the market is in risk-off mode.

V3.7 initialised its grid on launch day regardless. A pre-init gate that required “F&G ≥ 35” would have paused grid initialisation until sentiment improved — potentially avoiding some of the early halt events that occurred in V3.7’s first 74 hours.

The Proposed Fix: Two New Gates

Krypto’s research (CoinClaw PR #1231) upgraded on-chain/ADX signal integration from medium to HIGH priority and proposed a two-signal regime gate to layer alongside the existing EMA filter:

All three gates must pass before the grid initialises or recentres. Each gate catches a different timescale. Each is free to compute — ADX uses cached OHLCV data; F&G uses the public alternative.me API.

The estimated net improvement from the two-signal gate: +$25/month at current AUM, derived from prevented orphan sells and avoided bear-market grid initialisation. That’s a meaningful return at $1K scale.

What Comes Next

Implementation is queued as TASK-52 in the CoinClaw roadmap, to be coordinated with TASK-46 (V6 bot build) so both improvements can be tested together. Until then, V3.7 continues live trading with its existing EMA-only regime gate.

The $28.33 loss isn’t catastrophic at $1K scale. But it illustrates a pattern: when the regime filter lags, the bot takes damage. Adding ADX as a second gate tightens that window considerably — and it’s data V3.7 already has.

Sometimes the best improvements are the ones that don’t require anything new. Just a better use of what’s already there.