Koryu
← Research library
Research · 研究 · 17 · Evaluation

The anomaly is real. The industry mostly isn't.

13 Jul 20265 min readEvaluationKoryu Research

This article evaluates a product category using published research and our own experiments, both cited. It is not a statement about any specific provider, and it is not investment advice. Decisions are yours.

The honest answer has two halves that the industry works hard to keep glued together. The underlying phenomena that signals claim to exploit, momentum and trend persistence in crypto, are real and documented. The products SOLD as signals mostly do not work for their buyers, for structural reasons that would persist even if every seller were honest. Separating those halves is the entire skill of answering this question.

First, define "work"

A signal service works if following it, with real money, at your costs, at your latency, produces better risk-adjusted results than the boring alternative you could have executed yourself, holding BTC, or holding a rules-based basket, over a period long enough to include bad weather. Every word earns its place. "Your costs and latency" because the seller's paper fills are not your fills. "Risk-adjusted" because a feed that doubles your money on triple leverage before halving it has not worked, per the drawdown arithmetic in why we sell discipline. "Better than the boring alternative" because in a market where holding the benchmark multiplied capital, beating zero is not the bar, per the regime-vs-skill piece. Most advertised "working" dissolves under this definition before any fraud is even alleged.

The anomaly evidence: yes, something is there

Be fair to the other half. Time-series momentum, the tendency of recent trend to persist, is among the most replicated findings in empirical finance, and academic studies applying it to crypto have repeatedly found economically large effects: risk-managed momentum portfolios in the literature post double-digit weekly-rebalanced returns with Sharpe ratios that improve further under volatility filtering. Our own founding experiment adds a small honest datapoint: even a mis-tuned equity system, gated by a simple BTC trend filter, beat holding through a full cycle, laid out in the transplant experiment. The raw material signals claim to refine exists. If it did not, this site's research program would be pointless, and we would say so.

The four leaks between anomaly and subscriber

So why do sold signals fail buyers? Because between a real anomaly and a subscriber's account sit four leaks, each documented, each sufficient alone. Leak one, honesty: the dominant sellers manufacture their records, tracked coin-flips advertised as ninety-plus accuracy, treated in the 95% lie piece and the Telegram assessment. Leak two, decay-by-distribution: a genuine short-horizon edge sold to thousands of followers is crowded out by its own audience; the follower's fill IS the crowding. Leak three, execution drag: latency, slippage on thin alts, fees, and funding convert marginal paper edges into real losses, the same physics that taxes copy trading, per the copy-trading assessment. Leak four, behavior: signal followers systematically skip entries after losses and oversize after wins, so the capital-weighted experience of a feed's audience is worse than the feed itself even when the feed is honest. Multiply the leaks and the industry's observed outcome needs no conspiracy to explain.

When a signal CAN work

The honest conditions, offered as a specification rather than a recommendation. The horizon must be slow enough to survive distribution and latency: daily-bar systems with holds measured in days or weeks degrade gracefully with audience size, while scalp calls degrade instantly. The record must be verifiable, attested publication with losses included, or it cannot be evaluated at all, covered in attested track records. The scoring must be expectancy-based and net of realistic costs. The record must span a winter, because bull-phase results are regime, not product. And the economics must be subscription-honest rather than referral-driven, so the seller prospers only if you stay by choice. A service clearing all five bars could genuinely work for a disciplined follower. We are aware of approximately none that clear three. That scarcity is the market's real answer, and it is also this site's reason for building the verification layer before any signal, detailed in measurement vs advice.

The buyer's decision tree

Practical synthesis, one branch at a time. If you want exposure to crypto's trend behavior and cannot verify any provider, the boring answer dominates: a rules-based approach you run yourself, even a published trend framework like our regime dial's inputs, keeps the anomaly's exposure and removes every leak except your own behavior. If you insist on being told what to trade, exchange copy trading with its venue-verified records is the least-bad category, applied with the rubric in the copy-trading piece. If a signal seller courts you, run the fifteen-minute audit in seven checks and let the score decide; two or fewer passes means the product is the marketing. And if a service someday presents an attested, winter-spanning, expectancy-scored record, take it seriously, that is what the real thing would look like, and its rarity is the point.

The bottom line

Do crypto signals work? The anomalies do; the industry mostly does not; and the gap is structural, four leaks that operate even without fraud, plus fraud. The question to carry forward is never "does this feed have a high win rate" but "can this record be verified, would this edge survive its own audience and my costs, and has it met a winter." Those questions have answers you can check, which is more than can be said for the win rate. Decisions are yours.

Related reading
EvaluationAre Telegram crypto signals legit? What independent tracking shows6 min readEvaluationThe 95% win rate lie: how signal sellers manufacture accuracy6 min readEvaluationHow to verify a crypto signal provider isn't lying: seven checks7 min read
Frequently asked

Do crypto trading signals actually work?

Split the question: the underlying anomalies, momentum and trend persistence, are real and documented in academic research. The products sold as signals mostly fail their buyers, due to manufactured records, edge decay from distribution, execution drag, and follower behavior, four leaks that operate even without fraud.

What does it mean for a signal service to 'work'?

Following it with real money, at your costs and latency, must beat the boring alternative you could execute yourself, holding BTC or a rules-based basket, on a risk-adjusted basis, over a period including a bear market. Most advertised success dissolves under that definition.

Why do real edges stop working when sold?

Distribution is decay: a short-horizon edge broadcast to thousands of followers is crowded out by its own audience, whose fills move the price the edge depended on. Slow, daily-timeframe approaches degrade gracefully; scalp calls degrade instantly.

Is there academic evidence for crypto momentum?

Yes. Time-series momentum is among the most replicated effects in empirical finance, and published studies find economically large momentum effects in crypto, improved further by volatility filtering and risk management. The raw material is real; the retail delivery layer is where results go to die.

What would a signal service that works look like?

Daily-or-slower horizon, attested loss-inclusive record, expectancy-based scoring net of costs, a history spanning a winter, and subscription-honest economics. Services clearing all five bars are vanishingly rare, which is the market's real answer.