Methodology & Evidence

How signals are produced — and the honest current state of their validation.

Live evidence status

Loaded from the public evidence feed. Shows forward-tracking, not a back-fitted backtest.

Loading evidence feed.

How a signal is produced

1. Data sources. End-of-day prices, fundamentals and macro series across ~6,700 global assets (25 markets).
2. Quality gates. Assets with insufficient history, stale data or failed integrity checks are excluded or flagged (no fabricated values).
3. Regime context. Market breadth / risk-on-off and macro state condition how signals are read.
4. Signal logic. A multi-condition Setup (e.g. trend, price>SMA200, SMA50>SMA200, volume) plus a Trigger (e.g. RSI breakout, price>SMA20, volume spike, MACD). Only assets passing both are flagged as research candidates.
5. Confidence & reliability. Each result carries a data-reliability and confidence state (e.g. provisional while history is still building).
6. Forward evidence. Every signal is recorded and its outcome tracked forward. Hit-rates with Wilson confidence intervals are published only once enough outcomes have closed.

Modules

Six research modules feed the system, each with its own contract and evidence tracking: Breakout, Forecast, Historical probabilities, Historical research, QuantLab, Scientific setups. A result may use a clearly-labelled fallback (e.g. a price-history proxy) when a module's primary data is unavailable.

Score scale

Most RubikVault scores use a 0–100 display scale for ranking and screening. They are not calibrated probabilities unless a panel explicitly says probability and shows its validation state. A score of 80 means the asset ranked strongly under that module's current rules; it does not mean an 80% chance of profit.

• Candidate score: relative rank within the current eligible universe and horizon.
• Risk score: risk level; higher means higher risk, not higher quality.
• Proxy score: clearly labelled fallback when primary inputs are unavailable; proxy rows are not promoted as primary fundamentals or macro truth.

Release and data gates

Public UI output is gated before release. Rows may be hidden, marked partial, or rendered unavailable when required fields are stale or missing.

• Freshness: public snapshots must carry a data-as-of date; stale feeds must not display as live.
• Completeness: missing market cap, RSI, volume, earnings, P/E, or price-series inputs are shown as —, not synthetic values.
• Browser proof: release candidates require page render proof and smoke checks before promotion.
• No fake green: unavailable modules render an unavailable state instead of borrowed rows from another module.

Evidence lifecycle

Signals pass through forward tracking before any performance claim is published.

1. Shadow: signals are logged and outcomes are still pending. Public copy must not claim a validated hit-rate.
2. Closed outcomes: matured outcomes are evaluated against the original signal contract.
3. Statistical release: hit-rates are published only with enough closed outcomes and Wilson confidence intervals.
4. Rollback: if a release fails UI proof or evidence gates, the previous known-good code/data release can be restored.

Status terms

• Registry: asset exists in the universe catalogue.
• Canonical: ticker/exchange identity is normalized.
• Processed: raw data was transformed by a pipeline module.
• Decision-grade: required inputs for that module and horizon passed quality gates.
• Visible: row is allowed to appear in the public UI for the current filters.

Known limits

• No validated public track record yet — outcomes are still maturing (Shadow phase).
• Data can be delayed, incomplete or wrong; providers can fail.
• Models carry uncertainty; past behaviour does not guarantee future results.
• Displayed research screens do not include brokerage fees, spreads, slippage, borrow costs, taxes or liquidity constraints.
• Signals are research candidates, not personalized recommendations.