How the observatory works

Sentiment (how the community feels), bubble risk (how fragile speculative psychology looks) and price movement (what things actually sell for) are distinct. POKEPEAK.LIVE measures the first two only. It shows no sales prices and values no individual items.

Public, permitted pages found through web search: forums, Reddit, YouTube, news and social results, with an English-language / North American focus. Private groups and Discord are not covered. Each search set runs three queries with up to 8 results each, limited to roughly the last month of activity; duplicate URLs are removed before synthesis.

Each cycle deliberately searches for bullish and bearish arguments, removes duplicate URLs, and weights each source by estimated reliability.

Search results cover only what public engines return. POKEPEAK.LIVE does not independently verify every claim made inside a source, cannot see all of the community, and may miss relevant discussion. Missing coverage is reported as evidence gaps, not filled in.

Sources are synthesised by the Gemini 2.5 Flash AI model. Source reliability, sentiment, risk, confidence and scenarios are subjective model assessments of the found text — not observed statistics, measured sales data, predicted crash dates, or financial advice.

Every piece of evidence is tagged as discussion volume, buying-demand signal, supply, or price commentary. Loud discussion is not treated as proof of buying.

Modern singles, vintage singles, graded singles and sealed product are observed separately but combine into one overall index.

A metaphor: 0 risk ≈ 15 minutes to midnight, 100 risk ≈ 30 seconds. It never predicts a crash date and moves either way.

POKEPEAK.LIVE publishes one free comprehensive Daily Market Report plus five daily editions (Market Pulse, Doomsday Clock, Speculation Lab, Bull vs Bear, Catalyst Radar). Anyone can read them without signing in. Visiting or refreshing a page only reads already-published reports; it never starts an AI or web search.

Once per UTC day, a single scheduled cycle runs three web-search sets (sentiment, bubble risk, catalysts; three queries each) and shares that evidence across all six reports — about 9 searches and 6 AI write-ups per day. Each completed step is recorded so a retry never repeats finished work, and the cycle never runs twice at once. A forward-looking scenario lens is folded into the Daily Market Report rather than published separately.

Each report shows its UTC report date, when it was published (in your local time) and when its evidence was retrieved. If today's edition is not ready or generation fails, the previous published report stays up with a clear notice. Nothing is filled in or invented.

Every published daily report is kept and can be reopened by date. Older on-demand reports saved before the switch to free daily reports are preserved unchanged.

Paid on-demand reports, wallet charging and top-ups are switched off on the server. Existing wallet records are kept safely and untouched.

Charts plot only daily reports that were actually published. Nothing is backfilled or invented.

Spending is bounded by the single daily cycle: a fixed number of searches and AI write-ups, at most two tries per step and three attempts per day.