How the conviction score is derived

After a KOL publishes on X, the system fetches the post and routes each ticker view to an AI model for evaluation. The model assigns a single integer between 1 and 10. This value represents the model's read of language strength, not the KOL's stated position sizing or private conviction. Consequently, a KOL's actual position weight on a given ticker may diverge from what their public language, as scored by the model, would suggest.

1–3: passing reference

The post carries no defined view. Typical examples include incidental phrases such as "AAPL looks stretched here," or tickers that appear only while the post argues an adjacent thesis.

4–7: articulated thesis

A defined bull or bear case is laid out in measured language. The majority of routine KOL research posts fall in this range.

8–10: core position / high conviction

The post uses explicit strength language ("highest conviction," "core position," "all in," "largest position"), or the KOL explicitly identifies the ticker as one of the largest holdings in the book.

— (null)

The post contains no readable strength signal — typically a retweet, an incidental mention, or a statement too vague to interpret. A null indicates "no signal available from this post," and is not equivalent to a medium 5.

Strength vocabulary

The most direct lift. Phrases such as "highest conviction," "core," "all in," or "largest position" push scores into the 8–10 range.

Catalysts and quantitative detail

Views paired with specific figures, time windows, or price targets (e.g. "Q3 ASP +30%, target $120") carry materially more weight than generalized directional statements (e.g. "long-term bullish on AI").

Position in the value chain

In supply-chain frameworks (X → Y → Z), the KOL's actual bet is typically on the upstream anchor X. Downstream Y and Z generally appear as context and are scored low, or flagged as a mention rather than a distinct view.

Reiteration does not compound

When a KOL restates an existing view in a subsequent post, the score remains at its prior level. Repetition alone is not treated as an independent signal of increased conviction.

Null is not equivalent to 5

A dash indicates the model could not read the KOL's stance strength from the post. It should be interpreted as "no view expressed," not as "neutral / medium conviction."

Mid-range values drift slightly

The 1–3 and 8–10 boundaries are stable across runs. Scores in the 4–7 mid-range may drift by roughly one point between evaluations; the tier is a more reliable reference than the exact integer.

First, track a single ticker's score trajectory over time — a gradual move from 5 to 8 across a few months typically indicates building conviction, and the reverse indicates fading conviction. Second, compare scores across tickers to reveal preference ordering. Trends and relative ranks carry more usable signal than absolute values.