Amgis

Reading public posts on X and pricing every call against Kalshi.

Coming soonThe new token is not launched and no contract address exists. The engine already scores real posts.

Price–
Market cap–
24h volume–

$AMGIS

Every call, scored against the odds.

The water is the market price at the minute each call was posted. Every account’s settled calls stack into one landform: up for a call that beat its price, down for one that did not.

Docs App App

Follow one call to its score.

Every post read gets a verdict. Only committed calls on one exact Kalshi market are scored, and the price is fixed by the post id, not by us.

  1. Read.Original posts and quote posts from the last 90 days, at most the latest 1,000 per account. Replies, reposts, images and video are not read. In a quote post only the quoting account’s own words count.
  2. Decide.A keyword filter keeps posts that look like a forecast on something Kalshi lists. An AI reader decides call or not a call, and its model id and prompt version are stored with every verdict.
  3. Match.The matcher finds exactly one Kalshi market and side: same subject, same date, same number. Anything looser is “no exact market” and is not scored.
  4. Price.The call time comes from the post id itself. From Kalshi’s one-minute data, A is the hour before, B the last five counted minutes, C the highest value in the hour after. The largest is used, because a larger number is always worse for the caller.
  5. Score.Edge is 100 if the called side happened, else 0, minus that price. A 15% call that hits scores +85. An 80% call that misses scores −80.
  6. Fix.When each UTC hour ends, every new or changed verdict becomes a leaf and the hour’s Merkle root goes to our Solana program. The program is not deployed yet, so roots are built and not posted.

Drag the price

Starts at a real scored call.

Count every post, not only the hits.

Nobody can say calls were picked out when every post read carries a verdict.

    Pull the row sideways.

    Every account with a settled call, sorted by total edge. Each card is that account’s landform from the first screen, with the waterline at zero.

    Know what counts as a call.

    Only committed wording is scored. Confidence is not a call, however strong it sounds.

    WordingVerdictWhy
    “Denver will beat Kansas City tonight”Call“will” commits to an outcome
    “My pick: UConn +3.5”Call“my pick” is the account’s own
    An outlet’s “best bet”Callthe outlet stands behind it
    “Denver is likely to win”Not a callprobability language
    A race rated “Lean” or “Favored”Not a calla rating, not a commitment
    “If he plays, they win”Not a callconditional
    “Who wins tonight?”Not a calla question
    “Vegas has Denver”Not a callrelays someone else’s forecast

    Example wording written for this table, not real posts. Real posts appear only through X’s official embed.

    Ask before you trust a score.

    Short answers. The full rules, with versions, are in the docs.

    Why is the largest of A, B and C used?

    A larger number is always worse for the caller. X allows edits for an hour and the data has no edit history, so C reads the hour after the post too. Using the largest means a call can never be scored at a price kinder than the market offered.

    Not scored: fewer than 10 counted minutes in the hour before, an average spread wider than 10 points, volume under 1 contract in that hour, open interest under 1,000 contracts, or a void market.

    CheckOpen a real call and recompute its time, its price and its score in your browser.