Track Record · tested retrospectively, measured live
All coins: more survivorship bias – coins that crashed and disappeared are missing. Top 100: less distortion.
What works: our risk measure and on-chain growth. What doesn't: the old overall score.
On-chain growth has only been shown retrospectively and was selected after looking at the data – which is why it has been in a verifiable live test since 09/29/2026.
Risk measure
provenVolatility over the following 30 days: calmest fifth 32%, wildest 184% p.a. – the staircase rises at every step.
Largest interim loss: −10.5% to −37.6%.
Reliability: rank IC 0.72 with later volatility (t-value 42).
On-chain growth
in live test€100 in the top fifth became €184, in the bottom fifth €79. Bitcoin: €81, all coins: €118. Hypothetical, no fees.
Largest drop in the top fifth −34%, Bitcoin −43%. Top beat bottom in 29 of 46 weeks (63%).
Overall score v1
failed€100 in the top fifth became €99. Bitcoin: €81, bottom fifth: €258, all coins on average: €152. Hypothetical, no fees.
The score didn't find the winners.
Almost all groups rose because the market of today's coins rose (on average +52%). What matters is the order – and it was wrong.
Growth (v2): selected in the backtest, unchanged in a live test since 09/29/2026.
What €100 would have become
Re-split by Growth Score every week · all coins equally weighted · hypothetical, retrospective, no fees
Data as a table
| Date | Strongest growth | Weakest growth | Bitcoin | All coins |
|---|---|---|---|---|
| 11/10/2025 | €100 | €100 | €100 | €100 |
| 11/17/2025 | €90 | €88 | €90 | €88 |
| 11/24/2025 | €79 | €78 | €83 | €77 |
| 12/01/2025 | €86 | €78 | €86 | €80 |
| 12/08/2025 | €83 | €75 | €86 | €76 |
| 12/15/2025 | €81 | €72 | €84 | €73 |
| 12/22/2025 | €79 | €67 | €85 | €69 |
| 12/29/2025 | €81 | €68 | €84 | €71 |
| 01/05/2026 | €92 | €76 | €88 | €80 |
| 01/12/2026 | €96 | €84 | €87 | €85 |
| 01/19/2026 | €90 | €83 | €89 | €82 |
| 01/26/2026 | €84 | €78 | €83 | €76 |
| 02/02/2026 | €75 | €70 | €74 | €67 |
| 02/09/2026 | €68 | €68 | €67 | €64 |
| 02/16/2026 | €69 | €67 | €66 | €65 |
| 02/23/2026 | €66 | €66 | €62 | €61 |
| 03/02/2026 | €68 | €67 | €63 | €63 |
| 03/09/2026 | €69 | €69 | €63 | €63 |
| 03/16/2026 | €77 | €69 | €70 | €67 |
| 03/23/2026 | €71 | €70 | €65 | €65 |
| 03/30/2026 | €66 | €55 | €63 | €57 |
| 04/06/2026 | €71 | €57 | €66 | €59 |
| 04/13/2026 | €72 | €56 | €68 | €59 |
| 04/20/2026 | €75 | €59 | €71 | €61 |
| 04/27/2026 | €84 | €64 | €75 | €65 |
| 05/04/2026 | €81 | €61 | €75 | €64 |
| 05/11/2026 | €92 | €71 | €78 | €75 |
| 05/18/2026 | €80 | €63 | €74 | €65 |
| 05/25/2026 | €83 | €67 | €74 | €69 |
| 06/01/2026 | €85 | €67 | €70 | €69 |
| 06/08/2026 | €93 | €56 | €60 | €62 |
| 06/15/2026 | €103 | €59 | €63 | €65 |
| 06/22/2026 | €102 | €58 | €60 | €65 |
| 06/29/2026 | €99 | €53 | €57 | €61 |
| 07/06/2026 | €107 | €57 | €61 | €66 |
| 07/13/2026 | €103 | €57 | €61 | €65 |
| 07/20/2026 | €104 | €55 | €62 | €64 |
| 07/27/2026 | €109 | €55 | €62 | €65 |
| 08/03/2026 | €105 | €51 | €61 | €62 |
| 08/10/2026 | €110 | €51 | €62 | €63 |
| 08/17/2026 | €109 | €49 | €60 | €63 |
| 08/24/2026 | €143 | €59 | €74 | €81 |
| 08/31/2026 | €133 | €56 | €74 | €79 |
| 09/07/2026 | €150 | €63 | €77 | €92 |
| 09/14/2026 | €138 | €61 | €73 | €87 |
| 09/21/2026 | €163 | €69 | €78 | €105 |
| 09/28/2026 | €184 | €79 | €81 | €118 |
Calculated retrospectively with all rated coins. Caution: survivorship bias is strong here – small coins that crashed and disappeared are missing. All lines except Bitcoin are therefore probably too high.
Key figures like a fund
From the 46 weekly returns of the curves above · hypothetical, no fees · tap a figure for an explanation
| Metric | Strongest growthTop | Weakest growthBottom | BitcoinBTC | All coinsAll |
|---|---|---|---|---|
| +84% | −21% | −19% | +18% | |
| — | — | — | — | |
| 65% | 61% | 46% | 64% | |
| 1.31 | −0.21 | −0.39 | 0.54 | |
| 2.47 | −0.31 | −0.58 | 0.94 | |
| −34% | −51% | −43% | −43% | |
| — | — | — | — | |
| 1.11 | 1.00 | 1.00 | 1.16 | |
| 0.77 | 0.74 | 1.00 | 0.83 | |
| +106% | +5% | — | +55% | |
| 54% | 41% | 50% | 48% | |
| 65% | 48% | — | 52% |
Risk-free rate for Sharpe, Sortino and alpha: avg. 3.78% p.a. over the period (3-month US Treasury bills, FRED DGS3MO). Annual return and Calmar only from 52 weeks – so far 46. Sharpe and Sortino are annualised from weekly values (× √52). Caution: we chose the Growth Score after looking at this data. Retrospective figures therefore probably look better than they will live.
Average weekly return per fifth
Coins split into five equal groups by Growth Score each week
Every week individually – no cherry-picked examples
Each week: top fifth minus bottom fifth
29 of 46 weeks the top fifth was ahead
What works: risk
Our risk measure: how much a coin has swung over the last 30 days
High risk values did indeed signal stronger swings.
Volatility over the following 30 days
Volatility, annualised
Largest interim loss over the following 30 days
Average per fifth
Calculated retrospectively, average over 42 weekly reference dates, all rated coins. The risk measure says nothing about whether a coin will rise or fall – only how much it swings.
Live test
Real scores, stored daily – not calculated retrospectively
7 days reliableDay 300
30 days reliable
First reliable statement over 7 days after about 70 days, over 30 days after about 300 days – by then there are 10 non-overlapping periods each. Until then we only show live values as provisional.
Growth Score (v2-fundamentals)
Fixed on Sep 29, 2026 · since Sep 30, 2026 the score Kairos shows
- Only the fundamentals pillar of the model score, unchanged: TVL growth 30 days (30%), fee growth 30 days (30%), value capture revenue/fees (15%), TVL size (15%), fee volume 30 days (10%).
- Each factor as a percentile within the rated top 250; missing factors drop out of the weighting.
- Coins without DefiLlama data (TVL or fees) get no value and don't count in the live test.
Overall score v1 (retired)
Fixed on Sep 29, 2026 · still stored and measured daily, but no longer shown
- Weighted average of five pillars: valuation 35%, momentum 20%, fundamentals 20%, liquidity 10%, sentiment 15%.
- Each pillar consists of factors, rated as percentiles within the rated top 250 (excluding stablecoins and wrapped tokens).
- Macro adjustment of ±4 points, the same for all coins – doesn't change any ranking.
Verifiable
Every daily snapshot gets a Bitcoin timestamp – nobody can change the scores afterwards, not even us
The last few days
- Oct 27, 20266e5274e3e7…c8e3adwaiting for a Bitcoin blockCSVProof (.ots)
- Oct 26, 20266535ebbd90…8c37baanchored in Bitcoin· block 973,158CSVProof (.ots)
- Oct 25, 2026fdffcfbfda…882087anchored in Bitcoin· block 973,001CSVProof (.ots)
- Oct 24, 20262539166aea…33b0bbanchored in Bitcoin· block 972,858CSVProof (.ots)
- Oct 23, 20265be39f5c24…1e9000anchored in Bitcoin· block 972,685CSVProof (.ots)
- Oct 22, 20269a0e98373b…2c22b5anchored in Bitcoin· block 972,551CSVProof (.ots)
- Oct 21, 2026873743b90b…b76656anchored in Bitcoin· block 972,401CSVProof (.ots)
- Oct 20, 2026da41474ffa…6eba97anchored in Bitcoin· block 972,237CSVProof (.ots)
- Oct 19, 2026d3d4b98920…75698fanchored in Bitcoin· block 972,089CSVProof (.ots)
- Oct 18, 2026c9f5ba3b34…471aa4anchored in Bitcoin· block 971,932CSVProof (.ots)
A new timestamp waits a few hours until public OpenTimestamps servers write it into a Bitcoin block. Only the file's hash is sent to these servers, no data.
How to check it yourself
- Download a day's CSV and proof and put them in the same folder.
- The file is unchanged if its SHA-256 hash matches the one shown:
sha256sum kairos-2026-10-27.csv(Windows:certutil -hashfile kairos-2026-10-27.csv SHA256). - Check the timestamp:
ots verify kairos-2026-10-27.csv.otswith the OpenTimestamps client (pip install opentimestamps-client), which verifies against your own Bitcoin node. Without installing anything: drag both files into the verify box on opentimestamps.org.
What this proves: the scores for that day were fixed no later than the time of the Bitcoin block. What it doesn't: that they are right or useful – only the Track Record shows that.
For pros: key figures, top 100 and limits +
Backtest
Overall score v1: The score was off during the test period.
Nov 10, 2025 to Sep 28, 2026 · 46 weekly reference dates · on average 204 coins · no signals published at the time, calculated afterwards with the same model
Growth (v2, fundamentals pillar)
Selected after looking at these figures – so only a hypothesis
| Horizon | Rank IC avg. | t-value | Top − bottom | Top better | Top 100 only: IC (t) |
|---|---|---|---|---|---|
| 7 days | 0.020 | 1.2 | +1.9% | 63% | 0.005 (0.2) |
| 30 days | 0.061 | 1.8 | +8.8% | 79% | 0.058 (1.6) |
| 90 days | 0.150 | 3.8 | +39.2% | 94% | 0.154 (5.5) |
Overall score v1 (retired)
This is how it was live until 09/29/2026
| Horizon | Rank IC avg. | t-value | Top − bottom | Top better | Top 100 only: IC (t) |
|---|---|---|---|---|---|
| 7 days | -0.053 | -2.4 | −1.8% | 35% | -0.047 (-2.3) |
| 30 days | -0.068 | -2.4 | −4.2% | 26% | -0.028 (-0.9) |
| 90 days | -0.110 | -3.3 | −21.0% | 18% | -0.033 (-0.8) |
Rank IC: link between score and later return (−1 to +1, 0 = none). t-value per Newey-West; only reliable from about |2|. “Top 100 only”: the 100 largest coins per reference date – survivorship bias distorts less there.
Live since Oct 18, 2026
Growth Score (v2-fundamentals) · average per reference date, no survivorship bias
After 7 days
too earlyNo statement possible yet: only 1 of 10 independent periods.
3 reference dates · 290 observations
After 30 days
too earlyNo statement possible yet: There is no completed period yet.
0 reference dates · 0 observations
After 90 days
too earlyNo statement possible yet: There is no completed period yet.
0 reference dates · 0 observations
Overall score v1 (retired) · average per reference date, no survivorship bias
After 7 days
too earlyNo statement possible yet: only 1 of 10 independent periods.
3 reference dates · 581 observations
After 30 days
too earlyNo statement possible yet: There is no completed period yet.
0 reference dates · 0 observations
After 90 days
too earlyNo statement possible yet: There is no completed period yet.
0 reference dates · 0 observations
Limits
- Survivorship bias: today's top 250 were tested – coins that crashed and disappeared are missing.
- Selection after the fact: we picked growth (v2) after we knew the results. Retrospective figures for it therefore look better than they should – only the live test counts.
- Short period: the free data access only goes back one year; the periods overlap.
- Missing factors: trending without history (neutral); the macro adjustment doesn't affect rankings.
- Only one market regime – the test doesn't show how the scores behave in other phases.
- No fees or slippage.
How we calculate
- Every day Kairos stores the scores of all rated coins in the top 250 (excluding stablecoins and wrapped tokens) – per model version.
- Every week we split the coins by score into five equal groups and measure how each group performed over the following 7 days – all coins equally weighted, no fees.
- The curve “What €100 would have become” chains these weeks together. That works because the weeks don't overlap. The fund figures come from the same weekly returns.
- Coins that later drop out of the top 250 stay in live with their price – otherwise the result would look better than it is.
- Every week counts equally. We don't show cherry-picked examples.
Model-based assessment from public data – not investment advice. Cryptocurrencies are highly volatile; a total loss is possible.
What we tested, in plain language →And the well-known chart signals? Indicator Check →