calibration
window: 24 h48 h168 h· machine readable: /api/calibration · how all of it is computed: methodology
324 of 1,926 setups were already expired when they were written down: the engine finds a cluster after the fact, and lifetime counts from the first buy. Correct as a record, useless as an offer — such a setup never reaches the executor at any score. Detection takes 45 min, at worst 120 min. States: expired: 1318, rejected: 604, trade setup: 3, watch: 1.
what holds setups back
gate table, holds alone
Holding back means the condition was the only one not passed — one failed among five holds nothing. The last column decides what to do: measured and did not fit is about the threshold, not measured is about the data.
| condition | passed | failed | not measured | holds alone | of those, measured |
|---|---|---|---|---|---|
| entry not late | 1,623 | 303 | 0 | 52 | 52 · not measured 0 |
| market cap | 1,539 | 387 | 0 | 39 | 39 · not measured 0 |
| second source | 1,903 | 23 | 0 | 6 | 6 · not measured 0 |
| 24h volume | 1,736 | 180 | 10 | 5 | 5 · not measured 0 |
| taxes | 1,789 | 1 | 136 | 5 | 0 · not measured 5 |
| risk level | 1,688 | 14 | 224 | 3 | 2 · not measured 1 |
| liquidity | 1,681 | 212 | 33 | 0 | 0 · not measured 0 |
| audit | 1,822 | 3 | 101 | 0 | 0 · not measured 0 |
| consensus | 1,926 | 0 | 0 | 0 | 0 · not measured 0 |
| no honeypot | 1,822 | 0 | 104 | 0 | 0 · not measured 0 |
| entry not late: why | price moved #% against a limit of #% × 303 | ||||
| market cap: why | $# against a minimum of $# × 387 | ||||
| second source: why | #% of wallets left the Binance signal against a limit of #% × 12Binance closed its own signal (exitRate), the move already happened × 5Binance closed its own signal (outDecline), the move already happened × 4tags against a buy: Smart Money Remove Holdings, Insider Wash Trading, Wash Trading × 1tags against a buy: High Tax Token, Smart Money Remove Holdings, Insider Wash Trading × 1 | ||||
| 24h volume: why | $# against a minimum of $# × 180Token Info did not answer, no turnover visible × 10 | ||||
| taxes: why | the audit did not name the taxes × 136buy #%, sell #% against a limit of #% × 1 | ||||
| risk level: why | the risk level did not arrive × 224# against a limit of # × 14 | ||||
| liquidity: why | $# against the $# needed (# positions of $#) × 212Token Info did not answer, pool depth unknown × 33 | ||||
| audit: why | the audit answered but gave no result × 101mechanisms found: #, High Risk Found, Contract Code Not Verified, Newly Created Token × 2mechanisms found: #, Contract Not Renounced, Newly Created Token × 1 | ||||
| no honeypot: why | upstream did not declare the safety check ready, and there is no audit either × 104 | ||||
what the price did after the setup
three horizons, medians
Not profit and not trades: this is the token price movement from the moment the setup appeared — the price we could have entered at, not the price the wallet paid. A move counts as a win above 2.0% of costs. Three horizons because one always flatters: fifteen minutes says whether we were in time, four hours whether the token survived its own pump.
Read measured together with the median. The longer the horizon, the fewer tokens live to a second price point — a dead token has no price — so the four-hour median is computed over survivors and flatters. We cannot remove that bias, only name it.
| horizon 15m | setups | measured | median | above cost | best | worst |
|---|---|---|---|---|---|---|
| trade setup | 1,172 | 1,167 | -0.1% | 16% | +27.5% | -58.1% |
| rejected | 604 | 585 | -1.5% | 22% | +990.0% | -95.2% |
| watch | 149 | 145 | -0.3% | 21% | +220.4% | -88.5% |
| high conviction | 1 | 1 | -40.2% | 0% | -40.2% | -40.2% |
| horizon 1h | setups | measured | median | above cost | best | worst |
|---|---|---|---|---|---|---|
| trade setup | 1,172 | 1,162 | -0.1% | 25% | +92.9% | -66.4% |
| rejected | 604 | 598 | -5.4% | 19% | +2028.7% | -97.3% |
| watch | 149 | 144 | -5.5% | 17% | +188.7% | -94.1% |
| high conviction | 1 | 1 | -31.2% | 0% | -31.2% | -31.2% |
| horizon 4h | setups | measured | median | above cost | best | worst |
|---|---|---|---|---|---|---|
| trade setup | 1,172 | 1,074 | +0.9% | 46% | +91.4% | -69.5% |
| rejected | 604 | 244 | -1.6% | 32% | +1235.6% | -86.1% |
| watch | 149 | 42 | -0.9% | 19% | +161.8% | -69.7% |
| high conviction | 1 | 0 | · | · | · | · |
does the score rank
score in five parts
Setups with a measured outcome, split by score into five equal parts. If the formula works, the median grows from top to bottom — the main question to ask of a score.
| score | setups | median at one hour | above cost |
|---|---|---|---|
| 28 to 50 | 381 | -10.9% | 16% |
| 50 to 55 | 381 | -0.6% | 23% |
| 55 to 60 | 381 | -0.1% | 29% |
| 60 to 64 | 381 | +0.0% | 29% |
| 64 to 81 | 381 | -0.3% | 15% |
wallets in clusters
addresses and their clusters
One cluster has one outcome and several participants, so crediting it to each counts one event many times. The modest question here: are there addresses that turn up only in bad clusters. A reason to look, not to retire.
| address | score | clusters | with outcome | median at one hour |
|---|---|---|---|---|
| 0xdcc6…bc61 | 62.0 | 861 | 857 | +0.0% |
| 0x92de…e2f8 | 60.0 | 848 | 844 | +0.0% |
| 0x532f…f290 | 64.8 | 623 | 621 | +0.1% |
| 0x144a…48b2 | 62.5 | 202 | 200 | -0.5% |
| 0x9df2…e831 | 72.3 | 115 | 115 | -0.9% |
| 0xc7f6…e3ad | 73.1 | 115 | 113 | -1.1% |
| 0x4873…4828 | 68.5 | 109 | 107 | -0.5% |
| 0xc8db…9361 | 67.9 | 109 | 109 | -0.8% |
| 0x7590…7538 | 70.2 | 100 | 100 | -0.9% |
| 0x7875…f340 | 54.3 | 99 | 98 | -29.7% |
| 0x95c4…a260 | 64.9 | 99 | 99 | -1.6% |
| 0x9a2a…8def | 66.8 | 99 | 99 | -1.1% |
| 0x2d97…75c5 | 74.3 | 95 | 95 | -0.8% |
| 0x962c…fa25 | 69.2 | 95 | 95 | -0.3% |
| 0xdd6c…2a34 | 69.8 | 93 | 93 | -1.1% |
| 0x2f7a…bfa6 | 66.9 | 88 | 86 | -1.7% |
| 0x1046…2a25 | 73.0 | 86 | 82 | -0.4% |
| 0x207b…3be2 | 68.1 | 86 | 86 | -1.8% |
| 0x3d93…13b8 | 73.8 | 86 | 85 | -0.5% |
| 0xe737…0547 | 72.2 | 86 | 85 | -1.0% |
universe
subscription, bench, silence
A subscription socket holds a hundred addresses. An address that gave no trades takes a slot for nothing — but dropping it helps only if there is a replacement. Bench at zero means a cleanup would shrink the universe, not improve it.
what follows from this
conclusions and report time
- "entry not late" holds 52 setups, and in 52 cases the value was measured and did not fit — here the threshold decides.
- "market cap" holds 39 setups, and in 39 cases the value was measured and did not fit — here the threshold decides.
- "second source" holds 6 setups, and in 6 cases the value was measured and did not fit — here the threshold decides.
- Most setups (432 of 1926) are held by several conditions at once: tuning any single threshold will not give one new trade.
- 324 of 1926 setups were born already expired: found later than their lifetime ran out, so they reach the executor at no score.
Report built 2026-10-16 20:04:09 UTC. No number here was picked by hand — this page and npm run calibrate call the same function. What it deliberately does not do is name the number you should set: that is a human choice, and advice fitted to one week is not a calculation.