How to read prediction scores and scenario ranges

Bharat Markets AI attaches a prediction to every stock and fund with enough history: a 0–100 score, a label from Very Weak to Very Strong, and three scenario percentages for each horizon. Used well, they speed up research. Used badly — as a buy button — they mislead. Here is what each piece actually is.

The score is a ranked opinion, not a probability

A score of 72 does not mean "72% chance of rising". It means the model’s rules — momentum, trend, fundamentals, valuation and news sentiment — net out more positively than for an asset scoring 55. The useful reading is comparative and explanatory: open the reasons list and see which signals dominate. If you disagree with the top reason, you have just turned a black box into a research question, which is exactly the point.

Horizons change everything

The same asset can carry a Weak 1-month score and a Strong 5-year score, and both can be reasonable. Short horizons are dominated by momentum and news — noise-heavy signals — so their model weights are deliberately lower and their outcomes are the least reliable. Long horizons lean on fundamentals and trend quality. Always compare scores within one horizon; cross-horizon comparisons are meaningless.

Bear / base / bull is a spread, not a target

The three percentages estimate outcomes if the asset follows its own history: the base case annualises its typical return with some mean-reversion toward market norms; bear and bull widen the band by the asset’s realised volatility. A staid large-cap might show −8% / +6% / +18% for a year, while a jumpy small-cap shows −30% / +5% / +45%. The width is information: a wide range is the model telling you it genuinely does not know, and position size should respect that.

The four most common misreadings

• Treating the base case as an expected return. It is the middle of a wide distribution, not a promise.
• Anchoring on the score after reading the reasons — if the reasons look stale (an old news spike, a lagged fundamental), discount the score too.
• Ignoring the "data as of" timestamp. Predictions built on last night’s candles cannot know today’s 4% gap.
• Skipping thin-history assets. Fewer than ~60 sessions of data means no prediction or a low-confidence one — that restraint is a feature.

A sane workflow

• Screen by score within one horizon to build a shortlist.
• Read the reasons; reject anything whose logic you cannot explain to yourself.
• Check fundamentals manually (see the fundamentals guide) and the news flow.
• Let the scenario range — especially the bear case — size the position, not the base case tempt it.
• Re-check after earnings and major news; scores refresh daily but your thesis should refresh too.

Read next: how our predictions are computed. More guides: