This is field guide #29 in this desk’s Reading Room series. It exists because on Friday 21 August 2026, four reputable sources published four different closing prices for bitcoin on the same day, and the widest gap between them was $1,394 — 1.8% of the price. None of them was wrong. They were answering different questions, and none of them said which question.

The weekly close is the single most quoted number in bitcoin analysis. It anchors “best week since,” it triggers technical signals, it is the settlement reference for a great deal of published research, and this desk grades its own public markers on it. It is also one of the least standardised numbers in finance. This guide explains what a close actually is, why the answers differ, how to compute your own, and — with a worked example from this weekend — how a $572 move in price can change a headline from “best week since March 2023” to “best week since February 2024.”

1. The problem, demonstrated

Here is Friday 21 August 2026, one day, retrieved directly from each venue’s public API on 23 August 2026.

Two separate effects are visible here, and it is important not to confuse them.

The small effect is venue dispersion. Binance, Coinbase and Kraken all cut their day at 00:00 UTC, and they still disagree — by $12.49 between the highest and lowest close, and by $25.92 on the low. That is real. Bitcoin does not have one price; it has a price on each venue, and arbitrage keeps them close but never identical. Twelve dollars on seventy-eight thousand is 1.6 basis points. For almost every purpose, ignore it.

The large effect is the day boundary. CNBC’s figure is $1,394 lower than Binance’s. That is not a data-quality problem. It is that CNBC, like most US financial media, uses a 5:00pm Eastern reference rate — the convention inherited from equity and futures markets — while crypto exchanges cut the day at UTC midnight. Between 21:00 UTC Friday and 00:00 UTC Saturday, bitcoin rallied. CNBC’s day ended before that rally; Binance’s did not.

Everything else in this guide follows from that second point.

2. What a “close” actually is

In equities, a close is a genuine event: an auction, at a fixed time, on a specific exchange, producing an official price used for settlement, index calculation and mark-to-market. The 4:00pm ET closing auction on Nasdaq is a real thing that really happens.

Bitcoin never closes. It trades continuously, on hundreds of venues, forever. A bitcoin “close” is therefore a purely arbitrary convention: it is the last trade printed before an agreed cut-off. Change the cut-off and you change the close. There is no auction, no official print, no settlement obligation attached to it. That is why conventions proliferate, and why nobody has an incentive to reconcile them.

The practical consequence: a bitcoin close is only meaningful if you state the venue and the cut-off. “Bitcoin closed at $78,338” is an incomplete sentence. “Bitcoin’s Binance daily candle closed at $78,338.03 at 00:00 UTC on 22 August” is a complete one.

3. The conventions you will actually meet

A note on the last row, because it causes the most confusion. A rolling window is not a week. “Bitcoin’s best five-day stretch since March 2024” and “bitcoin’s best calendar week since March 2023” can both be true at the same moment about the same tape. A rolling window scans every possible five-session sequence in history; a calendar week only compares Monday-to-Sunday blocks. The rolling window has roughly five times as many candidates, so its records are broken far more often and its “since” dates are far more recent. Neither is dishonest. Blending them is.

4. Worked example: the weekend that moved a record

This is not hypothetical. It is happening as this guide publishes, and it is the cleanest illustration we have ever had.

Bitcoin’s Binance weekly candle opened Monday 17 August 2026 at $62,900.00. By Friday’s close it stood at $78,338.03, a gain of 24.5%. On Saturday morning, with price at $77,299, the week-to-date gain was +22.89%. This desk computed the ranking from its own klines: of 470 completed weekly candles going back to August 2017, only nine had gained more, and the most recent of the nine opened 13 March 2023 at +27.16%. Headline: best week since March 2023.

Then Saturday and Sunday both closed lower. By 06:20 UTC Sunday, price was $76,172.24 and the week-to-date gain was +21.10%. Rerun the same query: now fifteen weeks beat it, and the most recent is the week opening 26 February 2024, at +22.01%. Headline: best week since February 2024.

Nothing about the methodology changed. The price fell 1.5% and the record aged backwards by thirteen months.

The lesson is not that anyone lied. The lesson is that a superlative computed on a window that has not closed yet is a live quantity, and every outlet that prints one before the settlement is quoting a number with an unstated expiry. If you see “best week since,” ask three questions: which venue, which window, and has the window closed?

5. Compute it yourself in four lines

You do not need a data vendor. Every major exchange exposes weekly candles for free. The following returns the complete Binance weekly history and the current week’s standing.

import urllib.request, json, datetime
u = "https://api.binance.com/api/v3/klines?symbol=BTCUSDT&interval=1w&limit=1000"
w = json.loads(urllib.request.urlopen(u).read())
rows = [(datetime.datetime.utcfromtimestamp(k[0]/1000).strftime('%Y-%m-%d'), 100*(float(k[4])-float(k[1]))/float(k[1])) for k in w]
cur = rows[-1]
better = [r for r in rows[:-1] if r[1] > cur[1]]
print(cur, "beaten by", len(better), "weeks; most recent", max(r[0] for r in better))

Change interval=1w to 1d or 1M for daily and monthly candles. Binance returns a maximum of 1,000 candles per call, which covers the full weekly history from August 2017. Kline fields are [open time, open, high, low, close, volume, …].

Three practical notes. Binance’s weekly candle opens Monday at 00:00 UTC, so rows[-1] is the current, incomplete week — that is a feature, but do not compare it to completed weeks without saying so. Binance’s history begins in August 2017, so any “of all time” claim built on it silently excludes 2009–2017. And the same call against Coinbase or Kraken will give you slightly different numbers, which is the point of section 1.

6. Where the weekly close genuinely matters, and where it does not

It matters for comparison. If you want to say anything about a period longer than a day, you need a consistent block, and the calendar week is the least arbitrary one available. Ranking, seasonality, drawdown measurement and volatility estimation all need a fixed grid.

It matters for marker discipline. This desk publishes falsifiable markers with stated bars and stated deadlines, and several of them settle on a weekly close. That only works if the venue and the cut-off are named in advance. Our current Z1 marker requires a Binance weekly close at or above $73,025.15. Note the trap in our own construction: $73,025.15 is Binance’s 20 August daily close. Kraken’s close that day was $73,001.10 and Coinbase’s was $73,011.87. Written against a different venue’s tape, the same marker would have a bar $24 lower. Two decimal places imply a precision that the underlying number does not have. We name the venue for exactly this reason.

It matters much less than technical folklore claims. A weekly close above or below a level is a convention with no mechanical force. Nothing settles against it. No fund is compelled to rebalance to it. Its influence, to the extent it has any, is reflexive: enough participants watch it that some behaviour clusters around it. That is a real effect and a weak one, and it is not the same thing as a structural break. Be suspicious of any analysis whose entire thesis rests on a close being four dollars on one side of a line rather than the other.

7. Monthly and quarterly closes: the same rules, higher stakes

The monthly candle is the weekly candle’s problems with a longer lever. August 2026’s Binance monthly candle opened at $62,887.88 — note that this differs from the weekly open of $62,900.00, because 1 August fell mid-week — and stands at +21.13% with eight days to run.

That number carries a robust superlative, unlike this week’s. Binance’s nine prior Augusts read: 2017 +10.87%, 2018 −9.37%, 2019 −4.89%, 2020 +2.77%, 2021 +13.60%, 2022 −13.93%, 2023 −11.26%, 2024 −8.75%, 2025 −6.49%. August 2026 is already the best August in the record by a margin of more than seven percentage points, and the first positive August since 2021, ending a four-year losing streak. For that claim to break, bitcoin would have to close August below $71,440.63 — a further 6.2% fall with eight days left. That is a superlative worth printing, because it has a real cushion. “Best week since” had $572 of cushion. Know which kind you are quoting.

8. Seven traps, in the order you will hit them

1. Comparing an open week to closed weeks. The current candle has not finished. Rankings involving it are provisional by construction. Say so.
2. Blending a rolling window with a calendar window. “Five-day” and “week” are different populations with different record books.
3. Mixing venues inside one calculation. If your open is Binance, your close must be Binance. Cross-venue arithmetic manufactures moves that did not happen.
4. Treating exchange history as all history. Binance data starts in August 2017. Bitcoin does not.
5. Forgetting daylight saving. A 17:00 ET cut-off is 21:00 UTC in summer and 22:00 UTC in winter. Year-on-year comparisons across the boundary are misaligned by an hour.
6. Grading on a wick. An intraday print through a level is not a close through it. A close is a settlement; a wick is an argument.
7. False precision. Quoting a close to two decimals is fine. Believing those two decimals are meaningful across venues is not — the cross-venue spread on a normal day is 100 to 1,000 times larger.

Frequently asked questions

When exactly does the bitcoin weekly candle close? On the major crypto exchanges, at 23:59:59 UTC on Sunday, immediately rolling into a new candle at 00:00 UTC Monday. TradingView lets you change the week start, so a chart you see on social media may be cut differently. CME’s futures week runs Sunday 17:00 CT to Friday 16:00 CT, which is a different thing again.

Why does CNBC show a different bitcoin price from my exchange? Almost always the day boundary, not the data. US financial media typically use a 17:00 ET reference rate; exchanges use 00:00 UTC. On a day when bitcoin moves in the intervening three hours, the two figures diverge by exactly that move. On 21 August 2026 the gap was $1,394.

Which venue should I use? Whichever one you will use consistently. Binance has the deepest BTCUSDT book and the longest continuous kline history among the large venues, which is why this desk uses it and names it. Coinbase is the natural choice if your reference currency and jurisdiction are US dollars and the United States. The wrong answer is “whichever one supports the point I am making today.”

Does the weekly close predict anything? Not reliably, and be careful with anyone who tells you otherwise. Its usefulness is descriptive — it gives you a consistent grid for comparison and a fixed reference for settling a claim. Any predictive power is second-order and reflexive, and the published evidence for it is much weaker than the confidence with which it is usually asserted.

Why do “best week since” headlines disagree so much? Because three variables are almost never stated: the venue, the window definition, and whether the window has closed. Fix all three and the disagreements mostly vanish. This weekend produced three simultaneous, defensible answers — March 2023, February 2024 and March 2024 — from one tape.

The one-line version

A bitcoin close is a convention, not an event. Name the venue, name the cut-off, wait for the window to shut, and never compare a rolling window to a calendar one. Everything else in this guide is a consequence of those four rules.