Why a High Win Rate Does Not Always Mean Profitable Signals
Winning often helps only if the size of the wins is enough to cover the losses. A fixed August sample from TraderStat shows how a high win rate and a negative sum can coexist.
By TraderStat Research
- Data as of
- 2026-09-13T14:50:31.744889+00:00
- Observation window
- Signal publication dates: 1–31 August 2026 UTC; outcomes as recorded on 13 September
- Sample
- 160 source-defined closed crypto signals classified as free; One Million Challenge
The short answer
In this August sample, 145 of 160 recorded outcomes were positive, yet the sum of their price-return percentages was negative. The larger average loss outweighed the more frequent small wins.
In this article
| Measure | Archived result |
|---|---|
| Positive / negative / zero outcomes | 145 / 15 / 0 |
| Win rate: positive outcomes / all included outcomes | 90.625% |
| Average positive signal result | +0.574416% |
| Average negative signal result | −6.291600% |
| Sum of positive signal percentages | +83.290267 percentage points |
| Sum of negative signal percentages | −94.373994 percentage points |
| Net sum of signal percentages—not account return | −11.083727 percentage points |
The missing half of an accuracy claim
Win rate counts outcomes; it does not measure their size. A provider can record many small positive moves and fewer, much larger negative ones. The headline percentage can therefore look strong while the sum of the recorded signal results is below zero.
We found that pattern in a defined historical sample for One Million Challenge. Of 160 qualifying August signals, 145 had positive results and 15 had negative results. There were no zero outcomes. That is a 90.625% win rate, but the sum of the individual price-return percentages was −11.083727 percentage points. This is an illustration of the arithmetic, not a claim about subscriber losses or a judgment on every strategy offered by the provider.
Exactly which signals are included?
The sample covers signals published from 1 August 2026 at 00:00 UTC through 31 August, with an exclusive cutoff of 1 September at 00:00 UTC. It includes public, source-linked, deduplicated crypto records classified as free, with an eligible closed result under the source-defined first-target verification rules and a public Telegram post link. Modelled outcomes are excluded from this case study.
In the snapshot collected on 13 September, One Million Challenge had the largest qualifying sample among the provider groups returned by our query for that August window. We selected it by sample count, then examined the result. That does not make it representative of Telegram signals generally, and a common calendar filter does not prove that every provider's history is complete. The live profile uses its own broader available history and may include different totals.
Follow the gains and losses, not just the count
The 145 positive signal percentages sum to +83.290267 percentage points. The 15 negative signal percentages sum to −94.373994 points. Adding the two gives −11.083727 points. Dividing that sum by 160 gives an average signal result of approximately −0.069273%.
The average positive result is approximately +0.574416%, while the average negative result is approximately −6.291600%. The difference in size explains how the losses outweighed the more frequent wins. Calculations use the stored six-decimal values without additional rounding. Multiplying the rounded averages back by the counts will produce a small rounding difference.
The underlying posts explain the first-target convention
The linked LTC/USDT source lists a first entry of 44.33 and first target of 44.58. The corresponding first-target move is approximately +0.563952%. The linked INJ/USDT post lists a first entry of 4.776 and a stop of 4.475112; the move from entry to stop is −6.3%. Both levels match the archived source-defined records used in the sample.
Those examples show how the recorded outcome size is obtained. They are not a reconstruction of the provider's entire execution strategy. The posts also include other entries, targets and advertised leverage. Our case study uses one first-target outcome per eligible record and does not multiply returns by that leverage. The original price sequence comes from the stored verification; we did not newly audit every underlying candle for this editorial explanation.
Why the sum is not an account return
The sum gives each recorded signal one observation. A real account also has position sizes, overlapping trades, deposits, withdrawals, fees, funding, execution delays and a changing capital base. Without those facts, adding signal percentages cannot establish how much money an account earned or lost.
For the same reason, we do not compound this sample into a hypothetical equity curve. The records do not establish that every signal was taken sequentially with a fixed share of capital. The negative sum is enough to demonstrate the limitation of win rate; it is not enough to estimate a subscriber's balance.
What to ask before comparing two providers
Compare the average positive and negative outcomes alongside win rate and sample size. Use the same observation period, market, units and calculation policy. A Forex pip total and a crypto percentage sum cannot be combined into a meaningful leaderboard. A source-defined result should also be distinguished from one that depends on modelled entry or stop assumptions.
Read what is absent. A high closed-signal accuracy can omit unresolved ideas, missing history or outcomes that cannot be calculated. A free public sample does not validate claims about a paid channel. An advertised accuracy figure may use later targets, different entries or a different denominator from TraderStat's measure.
The conclusion this evidence supports
For this dated sample, a 90.625% win rate coexists with a negative sum of recorded signal percentages because the average loss is much larger than the average win. Accuracy alone would hide that difference.
The practical next step is to inspect the size of outcomes and the definition of the sample before acting on a headline win rate. The evidence here supports that reading habit. It does not predict future performance, identify the best provider or establish whether a paid subscription is worthwhile.
How we checked the evidence
We froze a sequential, paginated MAIN public-record projection on 13 September 2026. We filtered by August signal publication time, crypto/free classification, public source and post link, no duplicate reference, closed/verified status, and eligible ohlc-v2-first-target results. We grouped by provider and selected the largest qualifying group by count. The archived record IDs, values, filters and code reproduce all figures. Win rate includes zero outcomes in the denominator; this sample has none. Gains, losses and averages are calculated from stored six-decimal values without additional rounding. The two example source previews were checked against the recorded levels. A separate status check on 13 September found all 511 candidate records and no source-review rejections; archived outcome values were left unchanged. On 22 September, a scoped suitability check found all 160 retained case-study records unchanged and both linked source examples accessible. This later check did not repeat the full 511-candidate selection or reconstruct every market candle.
What this research cannot establish
- One provider and one month do not represent all Telegram providers or future results.
- The sample contains only eligible closed source-defined outcomes, so it excludes unresolved ideas and modelled outcomes.
- A sequential read is not an atomic database snapshot; source repairs and outcome rechecks can alter later counts.
- Stored historical outcomes were used; the editorial check did not newly verify every source post and every market candle.
- No subscriber account, paid-service results, leverage-adjusted return, fees, funding, slippage or capital allocation was measured.
- A later source-status check on 13 September found no rejected sources among the 511 August candidates; it is not an atomic reconstruction of the original snapshot.
Prepared with GPT-6 Astra from retained public evidence. The provider is a dated example selected by sample size, not an endorsement or a complete provider review.
Sources and further reading
- One Million Challenge profile and recorded history
Live profile totals cover a broader history than this frozen August cohort.
- LTC/USDT public source example
First entry and target support the +0.563952% price-move calculation.
- INJ/USDT public source example
First entry and stop support the −6.3% price-move calculation.
- How signal performance is compared
- TraderStat methodology