What stocksonar measures, how it reaches a classification, how sure it is, and what it cannot tell you.
stocksonar reads public conversation about listed stocks and answers two questions about each ticker: how far is today’s attention above this ticker’s own normal, and does that attention behave like organic discussion or like coordinated posting?
It describes the conversation around a security. It does not assess the security, the company or its prospects.
One platform read properly is a stronger claim than five named loosely, so this is exactly what each platform contributes today.
| Platform | What is collected | Activity multiple | Behaviour checks |
|---|---|---|---|
| Post and comment text, from a fixed set of equity communities | Yes | Yes | |
| X | Hourly post counts only. No post text is read | Yes | No |
| Bluesky | Post text | Shown as context | Yes, judged on its own |
| YouTube | Titles and descriptions, for the most-discussed tickers each day | Shown as context | No |
| StockTwits | Not collected | No | No |
| TikTok | Not collected | No | No |
StockTwits allows automated access only through an approved API, which this project does not have. TikTok search does not reliably tell a ticker from other uses of the same letters. Reddit is read over a fixed set of equity communities, so discussion elsewhere on Reddit is outside what was read.
A platform that is not read is shown on every record as absent, never as a zero.
Every ticker is compared with itself. The activity multiple compares current activity with the ticker’s own normal over its previous 30 days: 1.0× is a normal level, 2.0× is twice it. It is never a comparison with another stock.
Reddit and X are each read against their own 30-day history and weighted equally, so a platform with far more posts cannot drown out the other. A platform joins the multiple only after 30 complete days of its own; until both have that, the multiple is the one platform’s own figure and the record says it is single-platform. Bluesky and YouTube are shown beside it as context and do not move it.
Records chart activity against that normal over 24 hours, 72 hours, 7 days and 30 days.
Buzzing Now ranks tickers by this multiple, and by nothing else.
Posts are grouped by the claim, event or story they repeat, and each group gets a plain-language label, with representative posts kept so you can read the language yourself. The label describes what people are saying. It does not judge whether it is true.
Each check compares the conversation with the same ticker’s own history.
| Check | What is examined |
|---|---|
| Text similarity | How often posts share identical or near-identical wording. |
| Timing clusters | Whether posts arrive in tight bursts rather than spreading out over time. |
| Source concentration | Whether a small number of accounts produce a large share of the posts. |
| Amplification | Repost and reply patterns around the posts. |
| Account age | How much of the discussion comes from recently created accounts. Not active in the current version: it needs each author’s account creation date, which is not yet available. Records say it was not measured rather than leaving it out. |
Exact thresholds are not published, so the checks cannot be gamed. No single check decides a classification: what matters is how many sit outside the ticker’s normal range, how far, and whether they point at the same narrative.
Since Transducer v2.2, Reddit and Bluesky are each judged on their own posts against their own 30 days, with the same checks and the same thresholds. Nothing is pooled, and the record keeps both findings.
Every classification carries a confidence level, high, medium, low or insufficient data, and states what limits it. Confidence reflects:
A record states the exact version of the method that produced it, and that version is a frozen set of parameters with a content hash. The same inputs and the same version always produce the same record.
When the method changes, a new version ships and the old one stays. A record published months ago keeps its version and can still be replayed against the rules it was actually built under, rather than against today’s.
A language model may rephrase one paragraph of a record, Why this classification, and nothing else. The classification, the confidence and every number are settled before it is asked. It never sees a post, and its paragraph is checked: one that adds a number the evidence does not contain is thrown away and the plain template is published instead. A record says when its paragraph was written by a model.
A classification is an evidence-backed description of public conversation. It is not advice about any security, and it is not an accusation or a legal determination about any person or company. See Disclosures and Corrections.