Edges Burn Out: How to Tell a Dead Edge from an Ordinary Drawdown When You Can't See a Cause

Edges Burn Out: How to Tell a Dead Edge from an Ordinary Drawdown When You Can’t See a Cause

Your bot was profitable for months, then the drawdowns stopped being made up — and the fee, the spread, and the BTC chart are all the same. Did the edge die, or are you just unlucky? Here’s how to separate them

A new article in the cycle. The last one showed how an edge vanishes when the regime shifts. But the most common and most agonizing case is different: the bot worked in the black for months, the drawdowns were within the norm, and then the black ran out — and all the visible causes are in place. The fees didn’t change, the spread is already tight, the bitcoin chart looks the same. Did the edge die, or is this an ordinary drawdown about to recover? The old advice “watch the fee and the regime” is useless here — there’s nothing to watch change. This article is about separating decay from drawdown in exactly this case, where nothing obvious changed.

---

Here today, gone tomorrow

Let’s state your scenario honestly, because it’s the real problem. The bot traded for months, and you roughly understood its behavior: income within such-and-such bounds, drawdowns of such-and-such depth, all explainable by its parameters. Then came drawdowns that stopped being made up by income. You go looking for a cause — and don’t find one: the fee didn’t change, the spread is still 1 cent, you look at the BTC chart — it seems the same. Now what? Turn the bot off because “something’s wrong”? And what if it’s an ordinary deep drawdown, and tomorrow it recovers, and you’ve already quit?

Here’s the real difficulty the earlier articles skirted: a dead edge and an ordinary drawdown look identical in the equity curve — both are a minus. You can’t tell them apart by eye, in principle. But you can with two tools that give numbers instead of feelings: the drawdown envelope (what even counts as normal) and edge decomposition (what exactly broke). Let’s take apart both, and at the end walk through your scenario step by step.

One remark up front, to kill a false hope. You can set the equity curve aside as a signal right away: it’s a sum of the entire past, with enormous inertia. It turns down only when the loss of the dead period outweighs the profit of the live one — and by then you’ve already given back most of what you earned. Watching it to catch the death of an edge is like judging a ship’s turn by its wake. You need finer tools.

---

Tool 1. “Within the norm” is not a feeling, it’s a distribution of drawdowns

You said the key phrase: the drawdowns were “within the norm,” and then they stopped being. But what is “the norm”? If it’s a feeling, it will deceive you: a real edge, as we saw in the article on expectations, is in the red a third of the time over short stretches, and a painful drawdown with a live edge is an ordinary thing. The “norm” has to be computed, not felt.

And you’ve most likely already computed it — during validation. Remember the order-permutation of trades and the Monte Carlo from the earlier articles? They give not a single number but a distribution of drawdowns that your edge should produce. That distribution is the objective norm. Let’s look at the numbers, for an edge of +2 cents per trade over 2000 trades:

It reads like this: a drawdown of up to ~630 cents your live, healthy edge produces regularly — it’s normal, nothing to panic about, even if it hurts. But a drawdown deeper than ~790 cents happens for a healthy edge less than 1% of the time — so if you see it, the process has most likely changed, this is no longer bad luck. There’s your objective line between an “ordinary drawdown” and a “suspicion of burnout” — not a feeling, but a percentile.

Let’s check that it works. Take a live run and a burned-out one (the edge dropped to −2 cents in the second half) and see where their drawdown falls in this distribution:

The live one sits in the middle; the burned-out one climbs past the upper boundary. There’s the first question in your scenario, with a number instead of a feeling: has the current drawdown breached the envelope your validation gave you? Inside — endure it, it’s the norm. Past the 99th — the process has changed, dig further.

---

Tool 2. The norm is breached — don’t guess, take the edge apart

Suppose the drawdown climbed past the envelope. “Something broke” — but the equity curve is an aggregate, and you can’t see what in it. Here’s the mistake most people make: staring at the total PnL and guessing. The right way is to decompose the edge and find which component shifted. There are three components, and each is checked separately.

Component 1 — the win rate at your entry price. This is the heart of the signal. You enter with a limit at, say, 55 cents; at validation, at that price, you won 58% of the time — that’s your edge, three points. Now measure that same win rate over a fresh window of trades and compare it to the noise band:

If the fresh win rate stays in the band [53…63], the signal is alive, the minus is bad luck. If it dropped to 51%, below the band — that’s no longer noise, it’s a component shift: the predictive link died. Notice the 51%: it’s nearly your entry price (55 cents ≈ 55%), meaning the signal no longer gives an advantage, you’re buying a fair coin. That’s burnout, and you saw it in the component while the aggregate was still noise.

Component 2 — the quality of your execution. And this is right about your “1-cent spread.” You look at the spread on the ticker — it’s 1 cent, unchanged, so everything’s fine? No. The nominal spread can be the same while your fills degrade. Competitors showed up — and even at the same 1-cent spread, adverse selection rises (the very one from the article on strategy types): you’re filled more often on the unfavorable side, and the good entries go to whoever’s faster. What you have to measure is not the spread on the ticker but your execution: the realized entry price against the target (have you started, on average, entering worse than you posted your limit) and the share of missed good entries (fill rate). The spread is 1 cent, but your effective execution may have gone negative — and that’s a component that dies invisibly, because on the ticker everything is “as it was.”

Component 3 — the follow-through after the signal. The signal fires as often, but the move behind it stopped reaching the target? Then it’s not the frequency of the signal that died, but its strength. Checked as: under the same entry condition, the realized move has become smaller.

The point of decomposition: if some component shifted coherently and persistently — you don’t just have a “minus,” you have a mechanism, and with it the answer to whether it’s reversible. If all components are in their bands and the drawdown is within the envelope — it’s bad luck, sit tight. The aggregate PnL won’t show this; the components will.

---

Tool 3. “Regime” is not the bitcoin candle, it’s the variable your edge depends on

You looked at the BTC chart and said “the same.” But here’s a perceptual trap. The regime for your strategy is almost certainly not the shape of the daily candle you’re looking at with your eyes. It could be the level of short-horizon realized volatility (the daily candle is the same, but the five-minute swings changed), the strength of intraday mean reversion, the number of competing bots (visible in the rate of order-book updates, not in the price), the time-of-day profile. Bitcoin is “the same by eye,” while the variable your signal actually exploits may have crept away.

From the article on regime: split your trades by the variable your edge actually depends on (you found it when you checked the strategy for regime dependence), and watch exactly that. If the edge lived in high short-horizon volatility, and it quietly sagged — that’s a regime shift, even if the daily bitcoin candle didn’t budge. And an important difference from burnout: a regime shift is reversible (the regime returns, the edge returns), while burnout from competition is not. So on a regime shift you mute the bot temporarily, on burnout you turn it off.

---

The honest boundary: the window of uncertainty

Now the thing the article has no right to keep silent about. There’s a period when burnout and a deep-but-normal drawdown are statistically indistinguishable — this is straight from the article on expectations: to tell a shift in the edge from noise you need a sample, and while there are few trades in the new state, both explanations are equally likely. No tool gets around this: the envelope isn’t breached instantly, a component doesn’t leave its band on the first trade.

So the right response to “the drawdown climbed past the norm, but I haven’t caught the cause yet” is not a binary “turn off or leave on,” but cut the size and dig into the components. You reduce the bet (limiting the damage if it’s burnout), keep measuring the win-rate-at-price, the execution, and the regime variable (accumulating distance to tell them apart). As soon as a component leaves its band or the envelope is breached confidently — you decide. Demanding instant certainty means either cutting a live edge on a normal drawdown or waiting too long; cutting the size is the honest compromise between them.

---

Your scenario step by step

Let’s put it all together for your specific case — the bot was in the black, then the drawdowns stopped being made up, the fee/spread/candle are the same:

1. Breach the envelope. Pull the distribution of drawdowns of a healthy edge from validation. Is your current drawdown inside the 95th percentile? Then it’s most likely the norm — sit tight. Climbed past the 99th? The process has changed, go to step 2.
2. Take the edge apart into components. The win rate at your entry price — did it drop below the band? Your real execution (entry price against the target, fill rate) — did it worsen, despite the 1-cent spread? The follow-through after the signal — did it weaken? Found a shifted component — that’s burnout, and you know the mechanism.
3. Check the regime variable, not the candle. Short-horizon volatility, the number of bots in the order book, the intraday profile — the variable your edge depends on, not the daily BTC chart. Sagged — that’s a (possibly reversible) regime shift.
4. In the window of uncertainty — cut the size and dig. While the cause isn’t found and the envelope is breached, don’t turn it off flat and don’t sit on your hands: reduce the bet and keep measuring the components until the picture clears.

If after this all components are in their bands, the regime variable is in place, and the drawdown hasn’t climbed past the 99th percentile — you most likely have an ordinary deep drawdown of a live edge, and the right move is to endure, not to turn off.

---

What would refute this — and where the limits are

The condition on which the approach would collapse (and didn’t). If a burned-out edge’s drawdown did not breach a healthy one’s envelope — the envelope would be useless. But the burned-out one regularly climbs past the 95th–99th percentile, while the live one sits at the 52nd. And if the signal’s decay were not visible in a component before the aggregate — decomposition would be empty. But the win-rate-at-price falls below the band while the aggregate PnL is still noise. Both attempts confirmed the approach.

The honest limits. The envelope and decomposition move the moment of recognition earlier, but don’t make it instant — the window of uncertainty is irreducible (see expectations). And sometimes an edge burns out without a clear shift in one component — everything decays a little at once; then only the envelope and patience remain. The tools reduce the uncertainty, they don’t abolish it.

---

What I don’t know, and where the limits are

• The numbers are illustrative, from synthetics. Your envelope, your component bands are computed on your own validation data and will be different; but the logic is the same.
• The response threshold is your risk choice. React at the 95th percentile or wait for the 99th, cut the size by half or by a third — it depends on what you value more: not giving back the profit, or not cutting a live edge. There’s no objective point, there’s a conscious choice.
• Decomposition requires the components to be logged. The win-rate-at-price, the realized execution against the target, the strength of the move after the signal — this has to be written as you trade, or there’ll be nothing to take apart.
• The window of uncertainty can’t be removed by computation — only by caution and a small size until distance accumulates.

---

What to do tomorrow

1. Replace the feeling of “within the norm” with the envelope. Pull the distribution of drawdowns from validation; know your 95th and 99th percentile in advance, before you need them.
2. Log the components of the edge, not just the PnL. The win rate at the entry price, the realized execution against the target, the strength of the move after the signal — so there’s something to take apart when the norm is breached.
3. Don’t trust the ticker spread — measure your fills. At an unchanged spread the execution can degrade from competition; that’s a separate component of death.
4. Watch the regime variable, not the bitcoin candle. The quantity your edge depends on; it may have shifted invisibly to the eye.
5. In the window of uncertainty, cut the size instead of guessing. You can run and monitor the components and the envelope without risking capital, in paper mode — free, before any subscription.

---

FAQ

The bot stopped earning, but the fee, spread, and chart are the same — what do I check? Not what didn’t change, but two numbers. First: has the drawdown breached the envelope from validation (deeper than the 99th percentile of a healthy edge). Second: take the edge apart — did the win rate at your entry price drop below the band, did your real execution worsen (despite the 1-cent spread), did the move after the signal weaken. A shifted component is burnout.

The spread didn’t change — what does execution have to do with it? The ticker spread and your execution are different things. At the same 1-cent spread, competitors can worsen exactly your fills: you’re taken more often on the unfavorable side, the good entries go to the fast. Measure your realized entry price against the target and your fill share, not the spread.

How do I tell it’s just a deep drawdown and not death? If the drawdown is inside the envelope (no deeper than ~the 95th percentile of a healthy edge), all components are in their bands, and the regime variable is in place — it’s most likely the norm of a live edge, and the right move is to endure. Death gives itself away by a breached envelope plus a shifted component.

How long do I wait before deciding? There’s no fixed number — there’s a window of uncertainty where burnout and drawdown are indistinguishable. In it, don’t wait idly and don’t cut flat: reduce the size and keep measuring the components until the picture clears.

The bitcoin chart is the same — so the regime didn’t change? Not necessarily. The regime for your edge isn’t the daily candle but the variable the signal exploits: short-horizon volatility, the number of bots, the intraday profile. The candle is the same, but it may have shifted. Watch it, not the overall chart.

---

Disclaimer

This material is educational and is not financial advice. Past results do not predict future results. Trading on prediction markets carries real risk, up to and including the total loss of capital. Access to Polymarket is restricted or prohibited in some jurisdictions — verify legality where you live before trading.