The Live Track Record
Every week we run the same test and publish the result — good or bad. Coins whose Clever Score rose 5+ pointsover the week (“upgrades”) versus coins whose score fell 5+ points(“downgrades”): how did each group’s price do over that same week, against Bitcoin?
17
weeks measured — ordering held every week
72%
avg weekly beat-BTC rate, upgraded coins
16%
avg weekly beat-BTC rate, downgraded coins
| Week ending | BTC 7d | Upgrades: beat BTC | median return | Downgrades: beat BTC | median return |
|---|---|---|---|---|---|
| 2026-07-20 | +4.8% | 67% (81) | +8.1% | 9% (77) | -3.4% |
| 2026-07-26 | +0.2% | 67% (82) | +3.3% | 25% (63) | -3.6% |
| 2026-08-03 | -3.1% | 71% (94) | +0.7% | 20% (102) | -7.9% |
| 2026-08-09 * | +3.2% | 68% (221) | +6.7% | 18% (112) | -1.2% |
| 2026-08-12 * | -3.2% | 78% (128) | +3.6% | 26% (155) | -6.4% |
| 2026-08-22 * | +21.9% | 37% (314) | +17.5% | 6% (70) | -1.2% |
| 2026-08-29 | +1.0% | 61% (231) | +3.2% | 12% (82) | -6.2% |
| 2026-09-05 | +2.6% | 84% (295) | +9.4% | 17% (71) | -6.4% |
| 2026-09-13 | -3.8% | 91% (105) | +11.1% | 25% (150) | -8.3% |
| 2026-09-19 | +5.1% | 80% (76) | +14.0% | 20% (55) | +1.1% |
| 2026-09-26 | +3.6% | 88% (119) | +16.5% | 43% (69) | +2.8% |
| 2026-10-03 | +0.8% | 65% (79) | +5.0% | 11% (74) | -7.3% |
| 2026-10-10 | -2.1% | 77% (56) | +4.5% | 13% (71) | -7.6% |
| 2026-10-17 | -1.5% | 66% (56) | +2.7% | 3% (95) | -11.8% |
| 2026-10-24 | -1.2% | 89% (47) | +2.9% | 11% (185) | -7.6% |
| 2026-10-31 | -3.8% | 71% (28) | +3.7% | 8% (74) | -10.2% |
| 2026-11-07 | -1.8% | 67% (54) | 0.0% | 0% (35) | -9.0% |
* Hover the marked rows for measurement notes. (n) = coins in the cohort that week.
How to read this honestly
- This measures tracking, not prophecy. Score changes and price moves happen over the same week — the claim is that the score moves with the data as deterioration or improvement happens, not that it predicts tops or bottoms in advance.
- Medians, not averages.A handful of dead-coin repricing glitches print fake five-digit “returns”; we exclude anything beyond ±500% and quote medians so no single outlier flatters the numbers.
- Benchmark context matters. In weeks when Bitcoin itself rips (see Aug 22), beating BTC gets hard for everything — compare the upgrade and downgrade columns to each other, not to 100%.
- The early cadence was imperfect. Two of the first measurement windows overlap (marked *); the schedule has been a fixed weekly cron since. Rows are never edited after publication — the first nine are frozen in our source code.
The 90-day test — results
The Clever Score is built and backtested on a 90-day horizon, so before the first live cohort matured we committed to publishing its results whatever they said. Here they are: every coin scored at the v1.0 lock (July 3, 2026; 461 coins after 1 missing price series and 3 repricing glitches, both disclosed), split into quintiles by that day’s score, measured July 3 → October 1 against Bitcoin’s own +35.9%.
| July-3 score quintile | Median 90d return | Beat BTC |
|---|---|---|
| Q1 — lowest (39–50) | +10.9% | 24% |
| Q2 (50–54) | +22.4% | 40% |
| Q3 (54–57) | +32.6% | 47% |
| Q4 (57–60) | +21.9% | 38% |
| Q5 — highest (60–74) | +19.2% | 39% |
What the backtest predicted, and what happened: the pre-launch backtest put the top quintile at a ~40% beat-BTC rate and the bottom at ~31–33%. Live: the top quintile hit 39% — on target — and the bottom came in at 24%, worse than backtested. The top-vs-bottom spread was wider in reality than in the simulation, and the lowest-scored coins captured less than a third of Bitcoin’s move (bottom decile: median +11.4%, 20% beat rate).
What didn’t go to script:the ranking wasn’t perfectly monotonic — Q3 was the single best quintile. Two reasons, both disclosed: at ~92 coins per bucket the middle-quintile differences are within statistical noise, and the July cohort was scored by the early, compressed factor set (three history-dependent factors had not yet activated), which squeezed the middle quintiles into bands only a few points wide. The pre-registered claim — high scorers beat low scorers over 90 days — held. The next cohorts, scored by the full 30-factor system, will be published here as they mature.
Methodology, factor list, and the pre-launch backtest (misses included) are public on the methodology page. Clever Crypto is a research tool, not financial advice.