Skip to content
Proof Watch

How we calculate

What this screen shows

We collect public Korean-language posts (news, public Telegram channels, and communities such as Naver blogs and cafés) plus international media that mention the coins and exchanges we track, and tally their tone. Everything is calculated as of 09:00 KST each morning.

This is not a trading signal. It says nothing about price or outlook; it only shows the tone of the posts we observed. It also does not claim to represent the whole Korean market.

Sentiment index and verdict

Sentiment index = 50 + 50 × (positive − negative) ÷ judged posts

  • It runs from 0 to 100. 50 means positive and negative posts are equal.
  • With fewer than 30 judged posts we show no index or comparison (too few posts, dashed cell).
  • The verdict comes from a 95% estimate range. Above 50 is mostly positive, a range that includes 50 is too close to call, below 50 is mostly negative. The range only covers sampling variation, assuming posts are independent; it does not include automatic tone errors or collection bias.
  • “vs yesterday” is the difference between the 7-day window ending at the cut-off time and the 7-day window ending 24 hours earlier.

Which posts count

  • Only posts with a finished tone judgement (positive, neutral, negative). AI judges first; where a reviewer has checked a post, their judgement wins.
  • We drop posts unrelated to the subject, posts that mention it only as part of market context, and briefings or newsletters bundling several events.
  • A post covering several coins counts once per coin, so “Analysed mentions” at the top counts mentions per tracked coin or exchange, not unique posts. “Coverage” at the bottom counts the posts collected in the 7 days up to when the page was computed, so the unit and the end of the window differ (the index is fixed at 09:00 KST). We do not collect from private groups.
  • TokenPost belongs to the same group as the operator. It counts exactly like any other outlet. Whether someone is a paying customer is not an input, so it cannot change the index, the order or what is shown.

Today in one line and score contribution

Story contribution = 50 × [(pos − neg)story, today’s window ÷ judgedtoday’s window − (pos − neg)story, yesterday’s window ÷ judgedyesterday’s window]

  • Contributions add up to the “vs yesterday” change. The screen shows the three largest stories plus “Other posts” for the rest, so the total reconciles. They are arithmetic contributions, not causes.
  • The one-line summary is built only from story names and the sign of their contribution shown on screen. It never says anything the numbers don’t.
  • A story is a group of posts about the same subject, clustered by title and keywords. Titles are shown in their original language.

Stories on the move and spread path

  • We show the three stories whose post count over the last 24 hours rose the most against the 24 hours before (falling stories are not pushed up). Stories with no posts in the last 24 hours are left out. If the same article is grouped under several subjects, it appears once.
  • Grades: Watch = the story’s 7-day verdict is mostly negative; Spreading = at least 5 posts in the last 24 hours, at least twice the previous 24 hours (zero counts as one), seen in two or more source types in the last 24 hours; otherwise Ongoing. Spreading measures activity, so it does not use the positive green.
  • “First observed” is when each source type (news, Telegram, community) first carried the story. Values such as “+1d” are the time since the very first observation, not since the previous source type. It does not show who quoted whom.
  • Story cards on the dashboard use the last 7 days; story detail uses every post from the first observation up to the cut-off time (at most 37 days).

Coin hive and source gaps

  • Each coin gets an equal-size cell coloured by its 7-day verdict. More posts do not make a cell bigger.
  • Coins are ordered by how much their index moved since yesterday, up or down. The list under the hive repeats that order with names, verdicts and changes. It is not a good-to-bad ranking. We never rank individuals (YouTubers, channels) by name.
  • ▲ Attention surge: posts in the last 7 days are at least 1.5× the 7 days before, and 30 or more.
  • Source gap = the highest minus the lowest per-source index for the same coin. Shown only when two or more source types have 30+ judged posts.

Coin page and story detail

  • The coin page uses the same rules as the dashboard for one coin: verdict, 30-day chart by source, stories and posts.
  • Story detail plots every post of one story by source type and time, coloured by tone, with per-day counts.

Collection status and corrections

  • “Collection on time” at the top means a collection job succeeded within the last 6 hours. Otherwise it reads “Collection delayed” and no share card is published that day.
  • Event lines on the chart are entered by the operator with a date and a fact only (no rating or forecast).
  • If a judgement is wrong or something is factually off, use Request a correction. We keep a record of how it was handled.

Proof Watch is operated by TokenPost Inc.