How we score ourselves.
Every public performance figure on XLumience is produced by the method on this page: fixed before the outcomes were known, applied to every read, misses included. Most providers never publish their methodology. This is ours, in full.
00 · The scoreboard, today
Every morning we publish a price range for the next 24 hours. That range is the last step of a morning of measurement: price structure, our own reading of the XRP Ledger, regulation at the source, the catalyst agenda, news and sentiment (how the range is made). When the window closes, the real price decides whether it landed inside. This table is updated daily from that counter; nothing here is edited by hand, and the misses stay in.
| Band width (tercile) | n | Coverage |
|---|
| Day | Published range | Width | Real close | Inside |
|---|
00b · What happened after each reading
Each reading also leans a way: up, down, or neutral. We do not compress that into one hit rate, because a one-cent move is not insight. Three honest ways to read the same days, side by side; the numbers behind all three are below. Pick the rule you trust.
| Reading | n | ■ Held | ■ Did not hold | ■ Could not be judged |
|---|
01 · What counts as a read
Once a day, at 06:15 UTC, the XLumience Oracle™ publishes one read: a projected price range for the next 24 hours, a baseline, and a written reading of the market that explains how the range was arrived at. The read is published every day, whatever the picture: an uncertain day is named as such, never suppressed. There is no publication filter, so there is nothing to cherry-pick.
When the window closes, the read is resolved automatically against the real market price (source: CoinGecko, at the moment of measurement). Resolution is mechanical and runs independently of the system that made the read.
02 · How the range is made
The range is not a volatility formula with a name on it. It is the last step of a morning in which the engine first measures, then reads, and only then commits to a band. In order:
| Step | What is measured, and where it comes from |
|---|---|
| Price and structure | Spot price, 24-hour change, 14-day realised volatility (ATR), RSI, MACD, moving averages; the current market regime (trend, consolidation or transition) is classified from these. |
| On-chain, measured by us | Exchange net flow across ~740 labelled exchange wallets on the XRP Ledger, scored against its own baseline (z-score); whale-sized transfers; escrow, AMM and RLUSD figures. Our own scanner reads the ledger; nothing here is a third-party opinion. |
| Quantitative model | A gradient-boosting model trained on 365 days of daily data gives an independent lean. It is weighed only as far as its measured performance allows, never trusted blindly. |
| Regulation, at the source | Official feeds from US, Japanese and EU regulators and legislatures, read in their own language; the CLARITY Act's status is verified against Congress.gov on every run. |
| Agenda | A hand-maintained catalyst calendar (Fed, ECB, BoJ, escrow releases, votes, ETF flow days), each entry checked against its primary source. Near a binary event the engine deliberately holds conviction back and widens the band toward the side that carries the tail risk. |
| News and sentiment | A curated news archive where every item keeps its source and a working link, a Claude-based sentiment scan of that archive, the Fear & Greed index, and live OSINT (prediction-market odds, fresh headlines) for what the archive has not caught yet. |
| Our own record | The engine reads its own recent misses before writing. A system that does not look at yesterday's score has no business publishing today's. |
All of this goes into one reasoning pass. The Oracle writes the reading you receive as a member, weighs the factors against each other, and sets the low and high of the range, anchored on realised volatility and shaped by the regime and the agenda. A safety layer then checks the read for sanity (impossible prices, inverted bands, stale feeds) before it is logged and published. Nothing is edited after that moment.
What we do not compress into a single score: the direction the reading leans. The range is scored as one number because it can be; the lean is shown in section 00b as a distribution under three stated rules, so that a one-cent move is never sold as insight.
03 · The record (archived measurement, 25 August 2026)
Nine out of ten days, the price stayed inside the range we published in advance. Measured range coverage: 88.9% (96 of 108 resolved reads with a published range), against a nominal target of roughly 90%. Measured on 25 August 2026 over the full series; the live scoreboard above carries the same counter, updated daily.
The counter behind this figure is the full series of resolved daily reads, one row per day, no exclusions. Every read is timestamped when published; no outcome is ever edited afterwards.
04 · The range, and what it promises
Each read carries a projected range. Our target is that the real price finishes inside the range on roughly 9 out of 10 days (nominal coverage ~90%); the measured coverage is published next to the target, so you can see whether the range means what it claims.
A range can always be made wide enough to be never wrong. That is why we also measure coverage per band width: if wide ranges scored systematically better than narrow ones, the range would be buying safety instead of encoding uncertainty. In our measurements coverage is flat across widths: wide ranges appear on volatile days and earn the same coverage narrow ranges earn on quiet days.
| Band width (tercile) | n | Coverage |
|---|---|---|
| Narrow (4.5 to 7.5%) | 36 | 88.9% |
| Middle (7.5 to 10.5%) | 36 | 86.1% |
| Wide (10.7 to 27.8%) | 36 | 91.7% |
Measured 25 August 2026, n=108 resolved reads with a published range.
05 · Why a range and not a number
A single price target is either right by luck or wrong by design. A range that is scored every day says something checkable: how much room the market needed, and whether we named that room in advance. Over the series the published width has run from about 4.5% to 28% of the baseline, wide on volatile days and narrow on quiet ones, and coverage has been broadly flat across those widths. That is the property we are watching: the band should encode uncertainty, not buy safety.
06 · Versions and changes
The engine evolves. Material changes to how the read is produced are versioned and dated, and the record is never recomputed retroactively under new rules: a read is scored under the method that was in force when it was published. Prices are taken from CoinGecko at the moment of measurement; we do not run a proprietary price feed and do not claim one.
07 · What we do not publish
The internal weights, thresholds and parameter values of the engine are not public: what we look at is public, how we compute is not. Nothing in that boundary affects the record above, which is defined entirely by published reads and public market prices.