Polymarket Live Odds

• The displayed odds are the market's mid, expressed as probability up to the cent, refreshed on the feed cadence.
• Behind the line sits the ladder: bid/ask levels with sizes at 0.5–1.0 cent steps, which is what actually moves when odds change.
• The depth tells you whether the line is supported: a one-sided book near a strike is an odds line with no executable exit.
• The archive's live-capture mirrors this view at 250ms for every market; screens are refreshed, and history is recorded.

• The live feed is the current state; the archive is the record — they agree at the same timestamp by construction.
• The three numbers (mid, last, floor) diverge during repricing: a 99.6-cent ladder jump in one second means the served mid is stale the moment the book moves.
• Final-minute books are the extreme: with ask presence in 3 of 18 recorded frames, live odds near resolution are often a no-ask line.

Track the line and the book together: a screenshot of the odds without the ladder is like a quote without a market — the page teaches the pairing.

For products, the live feed is the heartbeat and the archive is the audit; pairing them keeps any odds-based UI honest and reconstructable.

• Stream the live odds for one market for an hour while the archive records the same window; at the end, join the two by timestamp.
• Mark the frames where mid and executable floor diverged; count the seconds of divergence — that is the honest "quote gap" of the hour.
• Find the odds line around a final-minute window and count the ask-absent frames; the measured 3-of-18 pattern is the reference for how extreme this gets.
• Build a table (ts, line, mid, floor, oneSided) and render the line plus floor as two step-lines; the gap between them is the live-odds story.

Live odds are the display layer of a probability market, and the display is only as honest as the book beneath it — which is why the archive pairs with the feed rather than replacing it.

A 99.6-cent ladder jump in one second means a served odds line can be stale before the next refresh; the divergence from the floor is the information, not the bug.

The same-second recording of all three numbers (line, mid, floor) is precisely what makes "live" auditable instead of vibes.

Products display the line; studies read the floor; both read the archive for confirmation — the division of labor that keeps every screen honest.

Final-minute markets are the edge case to handle in code first: an ask-less book near settlement produces odds with no executable exit, and panels should render that state explicitly.

Live odds are the action, the floor is the executable reality, and the timestamped record is the proof.

Divergence events — the repricing second, the final-minute ask absence — are where live odds teach the most.

The free feeds cover live, and the archive covers the proof; together they are the complete odds system.

An odds panel that cannot answer "what was the floor at the timestamp of that line?" is incomplete; wiring the archive lookup into the same view upgrades the panel from display to evidence in one feature.

Track the divergence count per hour (frames where mid and floor separated) as a single rolling number; when it spikes, a repricing or a final-minute book is the cause, and the number tells the product team before a user complains.

The endpoint of the page is practical: live and recorded odds are the same three numbers, captured at the same grid, used for different jobs — and the free tier covers the live half.

• odds explained
• current prices
• cross market data