Score breakdown

0x1966374A32a29D76b0ACbB9BFA68a95F86F4208b

Every number on the wallet page follows from the rows below. Nothing here is a summary of a calculation done elsewhere — the arithmetic is repeated on this page from the same inputs, so a figure that disagrees with the engine shows up as a visible discrepancy rather than a silent one.

Demo data

Judgment score

77/100

Weighted components

Deflated Sharpe

raw 1.723·weight 0.50 → 0.667·+66.0 to score

Normalised: clamp(DSR, 0, 1)
Evidence: Bailey & López de Prado deflated Sharpe over the measured window, computed from closed, priced fills. The haircut is a function of how many candidates were searched to surface this one — the raw Sharpe is not the ranking metric and never was.
Read as: Above 0.95 the record clears the conventional bar for "not explained by selection". Below it, the track record is consistent with having tried enough wallets to find this one by chance.

Drawdown

raw 29.5%·weight 0.15 → 0.200·+8.1 to score

Normalised: clamp(1 − min(maxDD, 0.5) / 0.5, 0, 1)
Evidence: Peak-to-trough decline of the equity curve the page plots. The floor means anything at or past a 50% drawdown scores zero here rather than going negative — a 60% and an 80% drawdown are both simply disqualifying.
Read as: A shallow drawdown is worth points because it is the part of a record a copier actually has to live through.

Time under water

raw —·weight 0.15·— to score

Normalised: clamp(1 − share of observations below the prior peak, 0, 1)
Evidence: Fraction of observations spent below the running high-water mark, from the same equity series.
Read as: Two records with the same drawdown depth are not the same record if one spent a month recovering and the other spent a year.
Missing: Not computable from what has been published for this address. Its weight was redistributed across the rows that were — which is why the effective weights above exceed the nominal ones.

Independence from BTC

raw —·weight 0.10·— to score

Normalised: clamp(1 − |corr(returns, BTC)|, 0, 1)
Evidence: Correlation of the return series against BTC over the same window. Both series come from the same observation timestamps.
Read as: A record that is a levered BTC position is not skill, however good the Sharpe looks. High correlation with an unremarkable deflated Sharpe returns the "Beta only" verdict outright.
Missing: Not computable from what has been published for this address. Its weight was redistributed across the rows that were — which is why the effective weights above exceed the nominal ones.

Consistency

raw 55.4%·weight 0.10 → 0.133·+1.4 to score

Normalised: clamp((share of positive observations − 0.5) × 2, 0, 1)
Evidence: Share of observations with a positive return.
Read as: Rewarded up to a point, and then a kill: at or above 99% positive the whole score is refused as an unverifiable curve, because a P&L series with no losing periods is the signature of accounting that is not marked to market.

One or more components could not be computed, so the remaining weights were rescaled to sum to 1. The score is therefore a judgment made on less evidence than the full model uses, and the effective weight column shows exactly how much more of the decision each surviving row had to carry.

Copy gate — thresholds, not weights

Gate	Observed	Bar	Result
Sample size Sample size wins over everything else. A short record with a perfect statistic is still a short record, and the alternative to refusing it is publishing a number a reader would reasonably treat as knowledge.	204 trades in 90d	≥ 60	clears
Max drawdown This is the cap on what a copier is asked to sit through, not a statement about the strategy. A record can be genuinely skilled and still fail here.	30%	≤ 35%	clears
Deflated Sharpe The raw Sharpe of the best wallet in a large search is high by construction. The deflated figure is the one that has already paid for the search.	1.72	≥ 0.60	clears
PSR Probability that the underlying edge is above zero. Below this bar the record is not distinguishable from luck at a confidence anyone should act on.	0.99	≥ 0.75	clears

Published signals naming this address · 0

No agent signal has been published for this address on this deployment. That is not a finding that the address is clean — it means no run has produced a verdict about it here. The feed says which.

How to interpret this

The score is a ranking device. It compresses five statistics into one number so that a hundred wallets can be put in an order. It is not a probability, not a price call, and not a recommendation — and a wallet scoring 80 is not “80% safe”.

Weights are published on purpose. A score whose formula is private cannot be argued with, and a score that cannot be argued with is a brand claim rather than a measurement. Every weight on this page is the same constant the engine uses.

Gates veto, they do not subtract. A wallet can score well and still fail the copy gate. The gate is about what a copier has to survive; the score is about whether the record is explained by skill.

Missing is not zero. An em-dash means the statistic could not be computed from public data. It is never rendered as 0, because a measured zero and an unknown are different facts and only one of them is evidence.

Nothing here is on-chain yet. Signals are hashed with the Alpha Ledger’s canonical payload format and queued. The contract is not deployed, so “queued” is the honest word and the page uses it.

Check it yourself

GET /api/signals — the feed, ungated, with a data_mode that tells you whether a run has happened at all.
GET /api/signals/verify/<id> re-derives every hash in a second implementation.
The normalisations above are the whole model. Recompute them from the raw column and you should land on the same score.
Read the post-mortem where we publish our own engine’s negative result.

Share this breakdown

Every judged address also has an embeddable badge that renders its current standing as an inline SVG.

Badge snippets

Raw figures on this page: deflated Sharpe 1.723, PSR 0.989, hit rate 55.4%, max drawdown 29.5% over 204 trades in the 90-day window.