The formula, in the open.

What it measures

Conviction weight —

the summed weight of every mention in the window, scaled against the highest-scoring ticker in that same window.

Momentum —

weight over the last 72 hours divided by weight over the 72 hours before that; a ratio of 1.0 contributes nothing, 2.0 or higher contributes the full 30 points.

Coverage —

how many distinct videos and channels mentioned it, scaled against the widest-covered ticker in the window.

Hype share —

the proportion of the ticker's weight from mentions classified as retail hype.

Reading it

A high BubbleIndex means the ticker is close to the window's most-discussed ticker on weight and coverage, and/or that its weight accelerated sharply over the last 72 hours. A low BubbleIndex means the opposite: little weight relative to the field, flat or falling momentum, narrow coverage. Because hype share adds to the score rather than subtracting from it, a high value does not distinguish sustained analytical coverage from a short burst of hype — read it alongside the hype flag and the signal-type breakdown lower on the page.

Why every mention is weighted

The panel is not 54 equal voices, and no scoring that treated it as one would be honest. Volume is concentrated: a handful of channels publish far more than the rest, so an unweighted average of the room would substantially be one channel speaking. The figures below are what the weighting above exists to correct.

14.3%

held by the single loudest channel

39.3%

held by the five loudest

62

mentions from the median channel in 30 days

12.6%

held by the quieter half — 27 channels

Measured over the same rolling 30 days as the figures at the top of this page. Concentration is a property of the conversation, not a judgement about any channel: the quieter half is where most tracked names first appear, which is why the panel is read whole rather than sampled from the top.

Hype detection

A ticker is flagged ⚠ when its share of hype-classified mentions runs clear of the panel's own rolling hype baseline, and only once it has enough mentions in the window to be statistically meaningful — a single hype-flagged video is never enough to trip the flag on its own.

The Hype Watch rail ranks flagged tickers by hype share, highest first — and only tickers that cleared the flag appear on it at all.

The data floor

A ticker clears the data floor when at least two channels mentioned it, or at least three videos did. Most do not: roughly two-thirds of tracked tickers in any window are named once, by one channel. Those are kept — they are where a name usually appears first — but they are marked "Low data" in every listing, ranked below the tickers that clear the floor, and left off the bubble map, where a dot built on one sentence would sit beside dots built on hundreds and look like the same kind of thing.

A ticker's own page goes further. It carries a caveat whenever it is built from at most two videos, whoever made them — including the cases that clear the floor on two channels. The floor decides what can be ranked against the rest of the panel; a page has to say what it is built from before it says anything else.

What this covers

Coverage is a measurement here, not a claim. Every ticker the panel named in the last 30 days is classified by what it is and, where it is a security, by where it trades — then weighted by the same attention score the rankings use. A name mentioned once counts for less than one the whole panel argued about, so these are shares of attention rather than shares of a list.

75.2%

listed on a US exchange

6.5%

market indices

4.0%

commodities and currencies

8.6%

crypto

1.5%

private companies

0.6%

listed outside the US

3.7%

not yet classified

Counted inside the US-listed share: 4.2% is US listings of companies headquartered elsewhere, such as Taiwan Semiconductor, Alibaba and ASML. That is US market exposure to foreign businesses, which is why the trading venue and the issuer's home are recorded separately instead of collapsed into one country.

"Not yet classified" is an honest residual, not a rounding error. The listing registries are US-only, so a symbol missing from them may be listed abroad, may be a crypto asset not yet curated, or may be a symbol a transcript mangled. Reporting it as its own share beats distributing it across the others.

Listings read from the SEC and Nasdaq Trader registries on 2026-09-04. A venue changes rarely, and the date says how old the answer is.