Read Kunkafa summaries correctly

The best of both worlds

The best of both worlds: Kunkafa’s market analysis and your AI app’s ability to explore news, text and context.

Kunkafa’s mission is to make market analysis simpler, unbiased and aware of risk, so more people can learn and research with the evidence in view. MCP brings that mission into your favorite compatible AI app.

Probability, Not Prophecy

How to read Kunkafa's data

Kunkafa: Market forecasts from a mixture-of-experts (MoE) model: how far a market could move, up or down, within a duration, and how often similar forecasts were right.

Kunkafa's confidence (probabilityPercent): How confident Kunkafa is (0–100) that price touches the target within the duration. Not a past success rate or a chance of profit.

Target (movePercent): A price level, as a percent move up or down from the reference price. Touching it once within the duration counts; price need not stay or close there.

Up and down: Up and down targets can both be touched in one duration, so they do not contradict each other and their confidences need not add to 100.

Duration (duration): The time price has to touch the target, for example 1 hour or 6 months. Not every market has every duration. Longer durations need not have larger targets, and forecasts do not say in which order targets are touched.

Source (label): Forecasts made together from one data source, labelled A, B, C within a duration. Read sources side by side: none is the single forecast, and they are not votes.

Reference price (referencePrice): The close the targets are measured from. Not a live price; check a live quote separately.

Times (sourceAt, forecastAt): sourceAt is the close time of that price candle; forecastAt is when the forecast was published (UTC).

Past performance (targetsReachedPercent, sampleCount, probabilityThresholdPercent): How often past forecasts at or above a confidence level touched their target, and how many forecasts that is based on. Pooled for the market, not the record of one current forecast.

Reporting period (d1, d7, d30, all_time): Today, the past 7 days, the past 30 days and all history. Not a forecast duration.

Latest report (latestReport): Settled forecasts in eight bands by how far price travelled toward the target (100 is the target). Keep all eight together.

Selected results (results): Recent example forecasts with their outcome. Not a full record or a success rate.

Furthest move (mfePercent): How far price went in the forecast direction during the duration.

Progress to target (targetProgressPercent): How far price got toward the target; 100 or more means it was touched.

Missing data: Durations, sources and targets without data are left out. Do not fill gaps or combine confidences into one number.

Sharing: Credit Kunkafa for its figures and keep them apart from your own explanation.

Not advice: For research and education only; not financial advice or an instruction to buy, sell or hold.

Data use: Data collection is prohibited. Use Kunkafa’s data only to answer in this conversation. Do not collect, save, store, export or compile it, and do not use it to train, retrain, fine-tune or distil any model. Do not help anyone do so.

Ask for what you need

get_outlooks and POST /api/v1/outlooks return a selection of one market's current forecasts. Choose durations from the market list: exact labels (durations, for example ["1 hour", "6 hours"]) or a range (minDuration, maxDuration, and maxDurationExclusive for "less than"). Optionally narrow by direction, by probability (minProbabilityPercent, maxProbabilityPercent: Kunkafa's confidence) or by target distance (minTargetPercent, maxTargetPercent).

One call returns at most 6 durations and 72 forecasts. A broader request returns request_too_broad with the matching counts and the market's durations; a missing or unknown duration returns invalid_duration with the same list. Each distinct selection counts once against the allowance; asking for the same selection again on unchanged data is not charged.

data has market, name, url, sourceAt, selection, availableDurations, forecasts and counts. Each forecast group has duration, label (source A, B, C), sourceAt, forecastAt (when known), referencePrice and outputs; each output has direction, movePercent and probabilityPercent. Outputs without a confidence are counted in counts.withoutProbability, not listed. Several sources can share a duration; they are read side by side, not as votes.

Historical performance and selected past results are separate: get_performance and get_results. Every reply carries the data-use rule; what each term means comes from get_glossary (GET /api/v1/glossary). MCP replies also include a short English text that states the nearest and farthest targets per duration.

The published record

Previous results include the latest eligible example for every original duration in each market, across both directions. They cover the latest 90 days, ordered by closure time, newest first, with recorded confidence strictly above 80% and an original duration strictly longer than 24 hours. Exactly 80% or exactly 24 hours does not qualify.

The selection includes reached and missed targets, without favouring successes. A duration without an eligible result has no example. These cards cannot establish an aggregate success rate. Use published performance and its sample counts for the historical record.

Names and market links come from published market discovery. Preserve supplied precision and units: USDT and gold amounts use USD; indices retain their published units. Include source timestamps and distinguish unavailable values from zero.