01
What is real?
Live Research uses sourced, read-only Binance Spot observations and deterministic detection. Event dossiers and practice replays remain fictional until an attributable historical corpus is published.
Data & methodology
Evidence before
explanation.
MarketLens separates observed records, deterministic calculations, hypotheses, and educational prose so every conclusion can be inspected.
Analysis ledger / four boundaries
What the system knows—and what it does not
02
What is calculated?
Absolute return, abnormal return, anomalies, ranking, and confidence are deterministic. Every confidence score exposes its factors, normalized weights, contributions, formula version, and calibration state.
03
How conclusions are phrased
MarketLens ranks evidentiary support and competing explanations. Confidence is not proof of causation, probability that a claim is true, expected return, or a trade instruction.
04
Known limitations
Live crypto is venue-specific and may be stale or unavailable. Historical Replay has a validated boundary but no public production corpus yet. Analyst mutations remain disabled without durable reviewer identity and storage.
Operating modes
Real, historical, and fictional never blur.
Active now
Live research
Current sourced observations and deterministic calculations. Explanations remain hypotheses, not trading instructions.
Corpus pending
Historical replay
A frozen point-in-time case built from attributable archived records. Later evidence is excluded until its original availability time.
Practice only
Demo laboratory
Fictional scenarios for learning the workflow. Nothing in this mode is an actionable market observation.
Confidence composition · confidence-v1.0.0
Visible weights, contributions, and calibration state.
Source reliability18%
Timestamp alignment17%
Asset relevance16%
Movement magnitude14%
Volume10%
Novelty10%
Competing-event clarity8%
Cross-asset confirmation7%
round(sum(score × weight) / sum(weight))Current demonstration scores are not outcome-calibrated and report a calibration sample of zero. A score is an evidence index, not an empirical probability.
Confidence laboratory
Move the evidence.
Watch confidence respond.
A sensitivity exercise using the published weighted-mean method. It does not edit event records or replace analyst judgment.
81
of 100
Evidence component / input score
Public trust ledger
Track record, including the misses.
Outcomes are grouped by signal version and confidence band. Corrections and retractions remain visible; small samples are never presented as calibrated proof.
0
Reviewed briefs
0
Written corrections
0
Retractions
No reviewed production sample yet. The ledger will publish outcomes without hiding corrections once the minimum sample exists.
01
Replay frames
Reaction windows reveal seeded evidence progressively; they do not reconstruct real information arrival.
02
Calibration score
40 points direction, 30 magnitude proximity, and 30 confidence proximity against a fictional record.
03
Regime & strategy rules
Deterministic educational labels can qualify or suppress a simulation; neither output establishes causation.