Polymarket 15-Minute Data

• Series-arc studies: a 21-game tournament at 15m is ~2,02 rows per market — a full arc in one table.
• Dashboard digests: 96 rows/day keeps dashboards light while staying dense enough for visible intraday patterns.
• Scheduled reports: email a 15m table; nobody reads a 345,600-row day in a digest.
• Storage economy: 96 rows vs 345,600 is a 3,600-to-1 reduction of the same truth.

A 15m row is a window aggregate over the 250ms record: price at the window's open, close at its end, high/low over it, volume summed.

The archive gives you the fine record and the 15m view from the same schema — so every 15m number is verifiable down to the frame.

Window alignment is a choice: calendar (00:00 aligned) versus market-relative (game clock aligned) produces different 15m rows for the same truth; state which.

• A 99.6-cent ladder move inside one second is invisible in 15m rows.
• Final-minute one-sidedness (ask in 3 of 18 frames) lives only in the seconds scale.
• Co-moving families (the 26-second four-coin collapse) require 250ms, not 15m, to measure.

• Pull one sports series at 15m for a two-week span: 96 rows/day means 1,344 rows per market — a full arc in one table.
• Recompute one 15m row (say 12:00–12:15) from the 250ms frames and check open/close/high/vol agreement; the derivation must round-trip exactly.
• Build a series-arc chart from the 15m table and overlay volume; the digest-level picture of a tournament emerges without 345,600-row pages.
• Save the market-clock alignment choice (calendar vs game clock) into a metadata field; state it when you publish.

The 15m resolution is the readability dial: it keeps every row meaningful for series-arc work while remaining dense enough to show intraday structure in a table.

Its one honest weakness is stated plainly on the page: the sub-minute behaviors — the repricing second, the final-minute ask absence — simply do not exist at 15m.

Use it as the all-purpose digest resolution and always anchor its claims to the frame record when precision matters; that pairing is the documented workflow.

In the stack, 15m is the study-and-report layer: derive it from the 250ms raw, cache it as a table, and let digests read it.

A dashboard that shows 15m bars and a deep-dive button that opens the same market at 250ms is the whole design, done.

The 15m view is the readability dial that keeps long tournaments and series arcs in one table: 96 rows a day means a two-week market is still a scan, not a scroll.

Use it for the breadth work — series comparisons, volume-at-range over days, digest reporting — and respect its boundary: repricing seconds, final-minute one-sidedness, and cross-market collapse live only in the frames.

Anchor every 15m claim with the frame round-trip: recompute one row from the 250ms record and the open-close-volume agreement is your trust contract for the rest of the table.

Store the calendar-vs-market-clock alignment explicitly; a 15m table whose window origin is undocumented is a number you will eventually misquote in a meeting.

The decision for 15m is really about breadth: it is the resolution for comparing every market in a series on the same page, and the 96-row day is what makes that possible without a scroll marathon.

The price of breadth is precision, and the page says it plainly: repricing seconds, final-minute one-sidedness, and family collapses do not exist at 15m. Any study of those must borrow the frames.

Anchor the table with a metadata field for the window origin (calendar or market-clock), and verify one row per run against the frame record so the table earns trust once and keeps it.

For digests and series-arc reports, 15m is the workhorse; for the claims those reports make, the frames are the notary.

• ohlc data
• intraday data
• candlestick charts explained