How the Oracle score works

Read the score with its signal indicators

BUY means a score of 70–100, HOLD means 45–69, and SELL means 0–44, subject to the model’s eligibility gates. UNRATED means the available signal indicators cannot support a directional classification. A score of 81 does not mean an 81% chance of success.

Each factor earns points up to its stated maximum. Factor scores are bounded and rounded to one decimal place; their sum is rounded to the nearest whole number and bounded to 0–100. Different asset classes use different inputs and weights. This is a weighted scoring algorithm; it does not use an isometric distance transformation.

Digital assets · CR-2.4

Crypto uses six factors totaling 100 points. SMA means simple moving average; RSI is the relative strength index. The 200-day factor uses the asset’s percentage price return over that interval, not its distance from the 200-day moving average.

Example: 20 + 20 + 14 + 7.5 + 11 + 8 = 80.5, rounded to 81. An eligible asset receives BUY; a blocking quality gate can still make it UNRATED.

Crypto technical fallback · CRT-1.1

When the full crypto model lacks 200-day history or market-wide inputs, a separate technical screen can issue BUY, HOLD or SELL with LOW confidence. It requires 50 consecutive completed UTC daily candles, a verified positive USD quote no older than five minutes, and no identity, stale-price or cross-source disagreement failure. The latest candle may be yesterday or one UTC day earlier; any one-day delay is disclosed beside the signal. A price check requires either a qualified exchange or a fresh aggregate snapshot with positive capitalization, a valid rank and at least $1 million in global daily volume. When only aggregate pricing is available, the lack of an independent exchange check is disclosed. Unavailable aggregate figures remain unavailable.

The technical score uses price/SMA20/SMA50 structure (40 points), RSI14 (25), 30-day closing-price return (20) and candle bias (15). Structure, RSI and candle rules use the full model’s point proportions. The 30-day scale is −40% → 0, −15% → 4, 0% → 10, 15% → 16, 40% → 20 with linear interpolation. BUY/HOLD/SELL thresholds remain 70/45. Coverage refers only to these technical factors; full-model omissions are shown separately. This new fallback has not been validated as a profitable trading strategy. It is excluded from daily Top 5 / Top 10 rankings.

Operating companies · EQ-2.3

The operating-company model combines fundamental, analyst and technical indicators. It is not applied indiscriminately to every listed instrument.

Other asset-specific models

ETF candle points use a pattern ending in the latest completed trading session; older patterns remain historical context. Flat zero-range candles do not count as bullish patterns. Geared ETF participation pairs completed-session volume with that same session’s closing return. A 100-point score means the selected factors reached their caps; it is not a return forecast, safety rating or calibrated probability.

Coverage and quality gates

Missing inputs use explicit fallback points and reduce weighted coverage. Scores are not reweighted upward to hide missing signal indicators. When coverage is below 50%, the raw score is confined to 45–69 before any model-specific gate is applied.

Operating companies, ordinary ETFs, listed products and metals require at least 65% coverage, three supporting signal indicators and sufficient observed-risk history. Geared products require 70% coverage and at least 50 daily observations. Event equities require 80% coverage and complete event-price quality checks. Failing these requirements produces UNRATED.

Crypto size policy SIZE-1 accepts a verified market-cap rank from 1 through 100 as sufficient market size. The rank must come from the selected provider snapshot within five minutes; saved directory ranks do not qualify. Outside that exception, missing market capitalization prevents a full-model call. Missing global volume, current rank or 30-day momentum also prevents a full-model call; the separately labeled technical fallback may still qualify. The aggregate fallback can recover cap, rank and volume from a fresh independent provider snapshot. When that provider does not report 30-day momentum, the Oracle calculates it from completed daily closing prices, using the same history-return calculation as the 200-day input; the response identifies this basis. Provider timestamps remain intact. If an aggregate provider rounds a small coin price too coarsely for the 5% price check, the Oracle can use a precise USD exchange quote only when two independent qualified venues agree and that quote lies inside the aggregate price’s reported rounding interval. The original aggregate price and timestamp remain recorded. A top-100 exception never invents a dollar market cap or turnover ratio: unavailable liquidity keeps the existing fallback points and reduces coverage. Blocking quote, identity or cross-source failures still prevent a rating. Missing required history prevents the applicable model from issuing a rating. Stable-value assets remain HOLD with LOW confidence.

The crypto model requires 201 consecutive completed daily observations to calculate its 200-day trend. Daily publication also requires an eligible BUY, matching identity and model, at least 201 consecutive completed daily history observations and at least two qualified independent price venues. Stablecoins are excluded. Unsupported security structures and unknown or halted equity market states do not receive an ordinary directional call.

What confidence means

Confidence describes indicator coverage and distance from a classification boundary. LOW applies when coverage is below 75% or the score is fewer than 6 points from the relevant boundary. HIGH requires 100% coverage and at least 15 points of distance. Other eligible results are MEDIUM; model overrides can lower confidence.

Boundary distance is score − 70 for BUY, 45 − score for SELL, and the smaller of 70 − score and score − 45 for HOLD. These labels have not been calibrated as win probabilities. Market risk and signal confidence are separate concepts.

How the daily Top 5 and Top 10 are selected

Preparation starts at midnight New York time for an 08:00 release, adjusting for daylight saving, every day. Completed editions are held privately until the release boundary. Late completion retains its actual publication time. The daily page shows the last released edition, preparation progress and whether a recent scheduler heartbeat has been verified.

New runs freeze the available non-test Nasdaq Trader U.S. security directory and a crypto candidate set combining the top 250 by market cap, top 100 by volume, and up to 50 gainers and 50 decliners from a fresh 2,000-asset pool. Movers require at least $50 million in market cap and $1 million in 24-hour volume. Duplicates and identified meme or NFT assets are excluded. Stablecoins stay visible as neutral policy states; unavailable data stays in the denominator. Earlier full-catalog runs retain their original audit. This is not coverage of every global exchange or untracked token.

Verified USD spot quotes and completed daily candles are combined with market-wide capitalization, rank, volume and return context. These inputs are tracked separately; exchange volume never substitutes for global volume. The main screener and daily scan share fresh server-side snapshots, preserving source timestamps and a five-minute quote limit. A failed provider never authorizes an invented price, a stale directional call or a substitute lower-ranked asset.

All qualifying BUY candidates are rechecked. For each category, publishable candidates sort by score descending, coverage descending, then symbol and stable asset key ascending. Top 5 is the first five of the same Top 10. Fewer than ten qualifying assets means a shorter list. Failed rechecks cannot revive an earlier BUY.

A scan spans time, so rankings reflect the rolling scan window and available signal indicators, not a simultaneous all-market snapshot. Each result retains its data time, model and risk context. A high rank is not an entry order, exit rule or position size.

Version and performance transparency

Documentation reviewed October 5, 2026 (Arizona). CRT-1.1 adds the separate LOW-confidence technical fallback and explicitly disclosed one-day candle delay. ETF-2.4 and GETP-2.5 correct candle recency and completed-session alignment. Crypto size policy SIZE-1 adds the verified top-100 size exception. Scoring weights and BUY/HOLD/SELL thresholds are unchanged. Model versions appear above and accompany individual signals.

BUY, HOLD, SELL and UNRATED are educational screening classifications. A repeatable trading strategy also needs entry and exit rules, sizing, execution assumptions and costs. This methodology does not establish a verified daily return, and no score guarantees a profitable outcome.