Data updated 2026-10-23 10:08 UTC+8 MCP · 7 tools 中文
Sandbox demo Change anything: this copy of the demo data is yours alone, invisible to others and cleared when idle. Alert delivery and user management are off; words not in the data can be searched on the web live (3 per visitor).
· Self-host it for your brand →
Estimated Sentiment here is the lexicon baseline estimate: the dashboard never fans out one LLM call per mention per refresh. LLM-scored sentiment lives in briefs and MCP, labeled as such.
Demo tour

First time here? Start like this
  1. Search a word — try Claude OpenAI MCP or use the search box at the top. It first looks through collected conversation; for a word that is not there, the results page offers Search the web now (3 per visitor).
  2. See whether it matters — the big verdict on this page (calm / watch / rising / alert) answers "do we need to act now?"; below it are the stories driving it and the rules that fired.
  3. Triage mentions — on the Mentions page, mark each one to do / done / ignored, like an inbox.
  4. Add your own subscription — on the Subscriptions page, enter a brand name and its usual spellings; verdicts and mentions then filter by it. For a new name, run Search the web now once on the mentions page to get data; a self-hosted install collects on a schedule.

Your changes are visible only to you; press Reset in the yellow bar at the top to start over at any time.

How to read this page: verdict first → then the stories driving it → then the caliber (Zone 7, every threshold printed)

01Verdict · Claude · 7d

Calm

Nothing needs action right now.

Evidence quality · high75/100

Evidence quality = how hard the evidence is (sample, sources, baseline, freshness), not how severe things are. The verdict and the evidence score are two different things.

  • google-news, hn, lemmy, bluesky, reddit, gdelt-news are newly connected (341 mentions) with no history, so they are left out of the volume ratio.
  • Volume 0.0× baseline, 15% negative: both under the watch line.
Triage inbox

02Key numbers

341 Mentions 341 from newly connected sources: not comparable with the previous window Measured · n=341 15% Negative share 50 / 341 Estimated · lexicon 6 Active stories multi-mention story clusters Estimated · cluster 6/ 6 Sources sources with data Measured

03Volume & sentiment

341 mentions in 7d, no baseline yet

Measured · Estimated

Daily mentions

Daily mentions010-10010-11010-12010-13010-14010-15010-16010-17010-18010-19010-209010-2118510-226610-23

Sentiment split

Positive23%78 Neutral62%213 Negative15%50

Daily mean sentiment (−1 to +1; days without mentions break the line, never interpolated)

Daily mean sentiment-10+110-1010-23

04Stories driving it

Story clusters ranked by mentions

Estimated
  1. 1 Trend
  2. 2
    Anthropic posts $80bn loss in first half of 2026: WSJ 4 mentions · Negative 100% · first seen lemmy · 3 Source
    Trend
  3. 3
    Anthropic unveils Claude Duet AI agent app 4 mentions · Negative 0% · first seen google-news · 2 Source
    Trend
  4. 4
    Could Anthropic's Compute Spend Outpace Its Revenue Growth? - The Motley Fool 4 mentions · Negative 0% · first seen google-news · 2 Source
    Trend
  5. 5 Trend
  6. 6
    RIP Cowork: Claude Code for everything? 3 mentions · Negative 33% · first seen hn · 2 Source
    Trend

05Source × sentiment

Source

SourcenShareNegativeMean
google-news 13740% 14+0.04
hn 6820% 8+0.05
lemmy 6018% 20-0.04
bluesky 4112% 3+0.06
reddit 226% 5+0.09
gdelt-news 134% 0+0.15

06Data health

Data sources

SourceStatusLast storedIn windowTotal
google-news Fresh just now137314
hn Fresh just now68134
bluesky Fresh just now41130
lemmy Fresh just now60114
gdelt-news Fresh just now1389
reddit Fresh just now2243

07Caliber & assumptions

Verdict rules · Evidence-quality breakdown
ConditionResult
n < 5Insufficient data
Mentions ≥ 3.0× Baseline & Negative share ≥ 30%Alert
≥ 2.0× Baseline | Negative share ≥ 35%Elevated
≥ 1.5× Baseline | Negative share ≥ 20%Watch
—Calm
Baseline = mean daily mentions over the 28 days before the window × window days. Sources with no history before the window are excluded from the ratio (counted in every other number).
PartScore
Sample size (50+ mentions = full)30 / 30
Source spread (3+ sources = full)20 / 20
Comparable baseline (share of mentions from sources with history)0 / 25
Sentiment backend (lexicon baseline)5 / 5
Freshness (newest mention ≤6h = full)20 / 20
Evidence quality75 / 100