Where your data bars come from

Overview

If you ask a retail trader where their equity price data comes from, they’ll usually tell you the name of the company they buy it from. They’re telling you who they pay, which is very different from understanding who really built it and how it was built.

The daily bar your backtesting platform reads is the end of a long chain of obligations, most of them regulatory. Exchanges are required to report trades and quotes. Nobody requires anyone to produce a “bar”. A vendor is entitled to redistribute what the exchanges report, and when it packages that into bars it decides what a bar means. Several decisions get made on your behalf somewhere along that chain, and your backtest inherits every one of them whether you know it or not.

On 6 December 2026 the chain changes again. US equities move to a 23-hour trading day, five days a week, and the trading date stops agreeing with the calendar date. Futures traders have lived with this for decades. For equities it’s new ground, and the interesting part is that the rules governing collection aren’t changing at all. What might change are the decisions your vendor has been making for you.

Why market data is harder than you think covered vendor conventions and archive management. This post goes underneath them: who actually produces US equity data, the regulation that compels it, what happens in December, how futures markets have coped with the same problem since 1992, why options are on a different clock entirely, and why providers like Norgate and Databento are in different businesses despite both selling you bars.

Nobody collects US equity data

US equity prices are reported, under compulsion, by the venues that execute the trades. Nobody goes out and gathers them.

Under Regulation NMS, every national securities exchange must report its quotes and trades to one of two SIPs (Securities Information Processors). Off-exchange executions (by alternative trading systems, wholesalers, and other broker-dealers) must be reported to a FINRA Trade Reporting Facility, which passes them on to the same place.¹ The two processors are governed by industry plans:

Together these are the consolidated tape. CTS carries trades, CQS carries quotes. From the quote stream the processors compute and publish the NBBO (National Best Bid and Offer): the best automated, round-lot bid and offer across all reporting venues. Rule 611 defines a trade-through against it, best execution is measured against it, and Rule 605 execution-quality statistics are computed from it.

Two properties of this arrangement matter for a daily bar more than anything else.

The first is that every trade carries sale conditions, and the sale conditions decide what a print is allowed to do. A trade executed outside regular hours carries the Form T condition and is not last-sale eligible. It can’t update the last sale price, so it can’t set an official high, low, or close.² It does count toward consolidated volume.² Hold on to that asymmetry, because it does most of the work in December.

The second is that the close is an auction. The official closing price comes from the primary listing exchange’s 4:00 pm closing auction. It isn’t the last print of the day across all venues, and it isn’t the last print in your data file either, unless your vendor has gone out of its way to make it so.

There’s also a second, parallel supply chain. Exchanges sell their own proprietary direct feeds, with more depth and lower latency than the consolidated tape, at a much higher cost. A vendor that captures those feeds (Databento is the accessible example) isn’t consuming the SIP at all. I’ll come back to that.

How the machinery got here

The current arrangement took five decades to assemble. Knowing roughly when each layer was added tells you where the discontinuities in your history probably are.

1975. The Securities Acts Amendments directed the SEC to facilitate a national market system. The consolidated tape and the consolidated quotation system followed, along with the Intermarket Trading System that linked the floors. Before this there was no single source of truth for a US equity price.

1997. The Order Handling Rules required market makers to display customer limit orders, and electronic communication networks took a real share of volume. Quoted spreads narrowed sharply.

1998. Regulation ATS created the category that Blue Ocean, Bruce, and MOON occupy today, and that most dark pools occupy.

2001. Decimalisation replaced fractional pricing. Any pre-2001 price series has a different tick structure, and the minimum spread fell by roughly an order of magnitude.

2005. Regulation NMS was adopted, with Rule 611 order protection and Rule 603 governing consolidated data. This is the direct cause of the fragmented market we have now. Order protection made it viable to launch a venue that couldn’t attract listings, because its quotes would be protected anyway. The market went from a few venues to 16 exchanges and dozens of ATSs.

2020. The Market Data Infrastructure Rule contemplated competing consolidators, variable round-lot sizes, and depth-of-book in the core data. Implementation has dragged. The round-lot changes matter most: odd lots don’t contribute to the NBBO, so a growing share of activity in high-priced names has been invisible to it.

2021 onward. Overnight trading emerged outside the exchanges entirely. Blue Ocean ATS launched in 2021 and by late 2025 handled roughly 83% of overnight volume, trading 8:00 pm to 4:00 am ET, Sunday to Thursday.³ Exchanges could have matched orders overnight for years. What was missing was a way to report the trades: no overnight NBBO, so no reference price and nothing to measure compliance against.³

2024. Settlement moved to T+1.

2026. DTCC’s National Securities Clearing Corporation went live with 24x5 clearing on 29 June, removing the last structural dependency.⁴

Session hours and reporting rules change more often than you’d think, and the changes get documented in exchange notices and plan amendments, rarely in vendor release notes. If your history spans any of the dates above, then the meaning of a bar changes somewhere inside it.

What changes on 6 December 2026

The SIPs extend to a 23-hour day on 6 December 2026. The CTA and UTP plan amendments were filed in January 2026, approved on 26 June, and the launch calendar went public on 7 July.⁵ A pre-launch period runs from 8 November to 5 December 2026 with test symbols only in production.⁶

Here is the nightly schedule, from the CTA’s own FAQ:⁶

Listing exchanges must deliver their daily XML file (corporate actions, adjusted closing prices, and the overnight band values) by 8:00 pm.⁶ The whole corporate-action and reference-data pipeline that used to have an overnight batch window now has 55 minutes, with a hard cutoff and a documented degraded mode. If an exchange misses the deadline, then the previous session’s symbol configuration carries over, no overnight bands are generated for its symbols, and the exchange decides whether to halt them.⁶

For a data consumer, the big change is that the trading date decouples from the calendar date. The entire 9:00 pm to 8:00 pm window is one trading day, and trades and quotes reported between 9:00 pm and midnight count under the next calendar date for all statistical purposes, including volume.⁶ Monday’s session begins at 9:00 pm on Sunday. This is the futures convention arriving in equities, and I’ll go through the futures market’s experience of it in the next section.

Overnight trades still can’t set a price. Form T applies across the 9:00 pm to 9:30 am stretch and after the 4:00 pm close, exactly as it does for today’s extended hours.⁶ The regular session’s open and close remain the primary benchmarks and the official closing price remains the 4:00 pm auction.⁵ But overnight trades do count toward volume,⁶ so price series and volume series will behave differently across the cutover.

Overnight risk controls are static. Subject to SEC approval, each listing exchange precomputes fixed LULD bands from a percentage of the 4:00 pm close and the 7:45 pm consolidated round-lot sale, publishes them once at Start of Day, and leaves them unchanged until 4:00 am.⁶ They don’t respond to overnight trading. Between 4:00 am and 9:30 am there are no bands at all, and dynamic bands resume at the regular open.⁶ So there are now three band regimes in a day.

FINRA’s TRFs move to the same schedule. Unreported, cancelled, or corrected trades from the prior session can be submitted as Prior Day messages from 9:00 pm in the following session,⁶ which means corrections to Monday arrive while Tuesday is already trading.

Participation is optional. Every listing exchange must support the session, and none is required to trade in it.⁵ The overnight NBBO will be assembled from whichever venues opt in. I’d expect a thin, wide consolidated quote that arrives in the same message fields as the regular-session NBBO and behaves nothing like it.

There’s a backstop for the date. On 7 August 2026 the SEC granted 24X a conditional exemption to run an overnight session without consolidated tape coverage, effective 24 January 2027 and only if the SIPs miss 6 December.⁵ Overnight equities trading arrives either way.

What doesn’t change: the 4:00 pm auction, regular-hours dynamic LULD, market-wide circuit breaker publication times, and the absence of bands between 4:00 pm and 8:00 pm.⁶

Index providers haven’t committed to overnight calculation.⁵ So for a while the constituents will trade for 23 hours and the index won’t.

Futures got here first

Everything the equity tape is about to go through, the futures market went through between 1992 and 2013, and its data conventions carry the scars. If you already trade futures, then most of December will look familiar. If you don’t, then the conventions the SIPs are borrowing are the futures conventions, and it helps to know how they behave in practice.

The futures session has started the night before for decades. CME Globex launched on 25 June 1992 with four contracts, trading after the pits had closed.⁷ The E-mini S&P 500 arrived on 9 September 1997 and traded nearly round the clock from the start.⁸ Today most CME Group products trade from 6:00 pm ET Sunday to 5:00 pm ET Friday, with a 60-minute maintenance halt each weekday afternoon, so a “day” is 23 hours long and Monday’s trade date opens on Sunday evening.⁹ CME writes its notices this way as a matter of course. A change takes effect “with the opening of the CME Globex session on Sunday, April 7, 2013 (trade date Monday, April 8th)”.¹⁰ Futures traders stopped equating the trade date with the calendar date a long time ago, and futures data vendors carry the exchange trade date as the primary key because nothing else works.

The Globex shape, 23 hours of trading and an hour of maintenance, is exactly the shape the equity SIPs adopt in December, and there are nearly three decades of futures data recorded under it. Thin overnight quotes, gaps that open at the maintenance break rather than at a close, bursts of activity at hours when the cash market is shut: all of it has a long record in ES, CL, and ZN to calibrate against. The one thing futures never had is the two-class print. An overnight futures trade is as good as a mid-morning one for the session’s high and low. Equities are borrowing the futures trade-date convention for volume while keeping Form T for price. That hybrid is one a futures trader won’t expect and an equity trader won’t notice.

Futures traders have also always distinguished the full electronic session from the “day” or “regular trading hours” session: the old pit hours, which outlived the pits and still carry most of the volume and produce the settlement. The window differs by product, because each pit kept its own hours:⁹

The “day session” column isn’t an exchange term. It’s a convention that brokers and vendors maintain, each in their own table, and it’s one of the settings you choose when you request a daily bar. Interactive Brokers’ historical data call takes a regular-hours-only flag. CSI and Barchart sell day-session and full-session series as separate products. TradeStation’s session templates do the same job. Two “daily” ES series from two vendors can differ in open, high, low, close, and volume, and both be correct. If you’ve never tried it, pull the same ES date from IBKR with the regular-hours flag on and then off and compare the two rows. It’s a cheap way to see how much of a bar is convention.

And the definitions move. In May 2012, CBOT grains traded 6:00 pm to 7:15 am and 9:30 am to 1:15 pm CT. ICE launched competing grain contracts on a 22-hour day, CME announced 22 hours in reply, then settled on 21 hours, 5:00 pm to 2:00 pm CT, effective 20 May 2012.¹¹ Eleven months later, for the 8 April 2013 trade date, the hours were cut back to 7:00 pm to 7:45 am, a 45-minute pause, and 8:30 am to 1:15 pm CT.¹⁰ A grain “day” changed shape twice inside a year, and whether the market was open when USDA released its crop reports changed with it.¹¹

The close moved as well. CME’s settlement procedure filed with the CFTC in January 2018 settles the E-mini S&P 500 off the volume-weighted average of trades between 3:14:30 and 3:15:00 pm CT, with a second tier for the bid-ask midpoint and a third for a carry calculation from the cash index when no two-sided market exists.¹² In September 2020 CME moved that determination period to 3:00 pm CT to synchronise it with the cash equity close.¹³ So an ES “close” from 2017 and one from 2021 are struck 15 minutes apart, and the trading between 4:00 and 4:15 pm ET that once fed the settlement now happens after it. Single-stock options walked into the same situation with their Curb session in August 2026: a contract that trades after the price that margins it has already been set.

Which brings up the deeper point about futures closes. A futures daily bar’s close is an administrative number, computed by the exchange under a tiered procedure with staff discretion at the bottom tier.¹² It may or may not coincide with the last trade. Vendors differ on whether the “close” field holds the settlement or the last print, and the difference matters most on thin contract months where the two diverge. Databento’s guidance for its CME feed is that official settlement prices and open interest come from the statistics schema, not from the OHLCV bars.¹⁴ The equity closing auction is the same idea under another name.

Then, in May 2026, CME removed the weekend. From 4:00 pm CT on Friday 29 May 2026, its cryptocurrency futures and options trade continuously.¹⁵ The weekday maintenance halt for those products shrank to two minutes, 4:00 to 4:02 pm CT, and the only scheduled break is a two-hour maintenance window early on Saturday morning.¹⁶ More than 7,200 contracts traded over the first weekend.¹⁷ The trade-date rule for that trading is the futures convention pushed to its limit: “All holiday or weekend trading from Friday evening through Sunday evening will have a trade date of the following business day, with clearing, settlement and regulatory reporting processed the following business day.”¹⁵

Read that as a data consumer. Monday’s trade date for Bitcoin futures now runs from Friday afternoon to Monday afternoon, roughly 70 hours, while Tuesday’s runs 24. By construction, Monday’s daily bar has three days of range and volume in it. Any daily-bar statistic on those contracts now carries a day-of-week artefact that the exchange put there, and a comparison of Monday bars before and after 29 May compares two different objects. There’s no weekend settlement, so Friday’s settlement stands through the weekend while the price moves. A vendor that builds bars by calendar date will emit Saturday and Sunday bars the exchange says don’t exist. One that follows the trade date will emit a fat Monday. Both are defensible, they don’t reconcile, and a continuous-contract roll that lands on a weekend will be handled differently again by each.

I’d have filed this under crypto curiosities a year ago. CME has since announced 24/7 trading for its 1-ounce gold futures from 26 July 2026 and a new 10-barrel WTI contract from 30 August 2026, pending regulatory review.¹⁸ The seven-day trade date is coming to commodities that have five-day cash markets, and the equity tape, having just caught up to the 1997 futures schedule, is still two steps behind.

What equities can borrow from thirty years of futures practice is a short list. Key everything by exchange trade date, never by calendar date. Keep the session table as versioned data, per product, with the notice that changed it. Store the designated close in its own field and never let it be confused with the last print. And sell, or buy, regular-session and full-session bars as two different products, because they are.

And then there are options

The December change is about the equity tape, and futures reached the same schedule on their own years ago. Listed options are the third case. They run under a separate plan through a separate processor, and that processor isn’t moving to 23 hours on 6 December.

Options quotes and trades are consolidated by the Options Price Reporting Authority, which is to listed options what the CTA and UTP plans are to equities. Nasdaq has confirmed that its options exchanges will continue on their current schedules when its equity market goes to 23 hours.¹⁹ So from December there will be roughly seven hours a night, 9:00 pm to 4:00 am, during which the underlying trades on a consolidated tape and the options on it don’t trade at all. Between 4:00 am and 7:30 am there is an equity pre-market with no listed options market of any kind.

Options hours have been moving too, on their own track. Cboe extended global trading hours for SPX and VIX options to nearly 24 hours a day in November 2021. Single-stock equity options followed on 17 August 2026, when Cboe’s C1 exchange added a morning Global Trading Hours session from 7:30 to 9:25 am ET and a Curb session from 4:00 to 4:15 pm ET for a designated list of liquid classes. The SEC approved it in Release 34-105569 on 28 May 2026, and the original 13 July launch target slipped.²⁰

The structure Cboe adopted deserves a close look, because it’s the same pattern the equity SIPs use and it produces the same kind of silent data change:

Look at the fourth column.²¹ Equity GTH and Curb activity is disseminated over the existing OPRA RTH channels, on existing lines, with no additional fees.²¹ Cboe marks those trades with the Extended Hours “v” sale condition and says outright that they aren’t last-trade eligible and don’t count toward the daily high or low.²¹

For a vendor this means the extended session needs no integration work. It just appears in the feed. Massive told its customers exactly that: the data lands alongside the regular session with no changes required at your end.²² Which is convenient right up until you realise what it implies. If your options pipeline doesn’t inspect the sale condition, then your regular-session file now contains 7:30 am prints, and nobody will have told you.

A few other details matter if you model options.

Settlement marks stay anchored to 4:00 pm. OCC marks closing and settlement prices from the 4:00 pm ET NBBO, and determines in- or out-of-the-money status from the 4:00 pm closing price of the underlying.²¹ Expiring single-stock options trade until 4:00 pm in RTH and 4:15 pm in Curb, because American-style physical settlement means holders want the chance to close rather than deliver.²¹²³ So there’s a 15-minute window in which an expiring contract trades after the price that determines its exercise has been struck.

There’s a separate marking artefact as well. Cboe publishes “smoothed” NBBOs built from 4:00 pm and 4:15 pm snapshots for equity options it trades after 4:00 pm, as CSV files on its website and disseminated to OPRA.²¹ If your prime broker’s marks disagree with a close you derived from the tape, this is a likely explanation. It’s a distinct object from both the tape and the OCC settlement price, and most retail feeds don’t carry it.

Order types and protections differ by session. Only limit orders are accepted in GTH and Curb; market, stop, and stop-limit orders are rejected at entry.²¹ Complex instruments containing a stock leg trade in RTH and Curb but not GTH.²¹ Wide Market Protection is enabled for eligible equity classes in GTH and Curb only, and not during regular hours.²¹ A microstructure assumption calibrated on regular-hours behaviour is calibrated on a different rulebook.

Eligibility is a list that changes, and failures are silent. Cboe designates eligible classes on its own criteria and updates the list semi-annually. If you submit an extended-session order for a class that isn’t on the list, then it isn’t rejected. It’s accepted, treated as RTH-only, and rests until 9:30 am.²¹ Cboe’s own guidance is that firms should maintain their own eligibility logic.²¹ A backtest that assumes a 7:30 am fill in a name that was never eligible will still produce results, just fictional ones.

Implied volatility gets harder. Any IV surface needs an underlying price synchronised with the option quote. After December the underlying quote exists for 23 hours and the option quote for a handful, on a partial class list, on one exchange. Choosing which underlying observation to pair with a 7:45 am option quote becomes a modelling decision, and the equity NBBO at that hour is a thin thing assembled from whichever venues opted in.

So “extended hours” is at least three cutovers on separate plans with separate rules: index options in 2021, single-stock options in August 2026, the equity tape in December 2026. Each arrives through existing channels, distinguished only by a sale condition.

Collectors and curators

Two very different kinds of vendor will sell you data, and December is where the difference might start to cost you money. I’ll use Databento and Norgate as the examples, because they sit at opposite ends of the chain and both document what they do.

Databento is a collector. Its equity data is sourced directly from each exchange across 18 datasets, normalised, with up to four timestamps per event at sub-microsecond accuracy.²⁴ Its consolidated end-of-day product is built to maximise CTA and UTP coverage across exchanges and ATSs.²⁵ You’re buying a capture of the wire. What arrives is what was on the feed, with the vendor’s normalisation layer between you and it. The failure modes are capture gaps and normalisation bugs, the sort of thing that shows up in a published issue tracker.

Norgate is a curator. Its FAQ refers only to “our primary upstream suppliers” and never names them.²⁶ It sells no live, delayed, intraday, or tick data at all.²⁷ What it sells is a reference layer: delisted securities back to 1950 for NYSE listings, daily historical index constituents showing exact membership on each trading day rather than a month-end approximation, a stable assetid that survives symbol changes and exchange transitions, OTC continuation past downlisting, and four selectable corporate-action adjustment methods.²⁶

That work is expensive and labour-intensive, and Databento doesn’t compete on it. The two are answering different questions. But where each sits on the chain has consequences.

Norgate has already committed to the December changes, by delegation, and the commitment is written down. Its stated rule is that only regular last-sale-eligible trades during regular trading hours contribute to OHLC, while those trades plus pre-market, post-market, block trades, and other omplex order types contribute to Volume and Turnover, with the exact rules determined by the Consolidated Tape Association and the UTP Plan.²⁶

Read that against the December schedule. Overnight prints carry Form T, so they can’t touch Norgate’s OHLC. They’re trades on the consolidated tape, so they’ll land in Volume and Turnover, dated to the session under the new convention. I’d expect a volume discontinuity with no matching price discontinuity, arriving without an announcement, because under Norgate’s own rules nothing changed. (That’s my inference. Norgate hasn’t said anything either way.)

The other structural difference is point-in-time integrity. Norgate is open about the fact that corrections flow through automatically from upstream, with no notification and no versioning, applied continuously to your local database while your subscription is active.²⁶ The history you backtested against last quarter may not be the history you hold today, and you can’t reconstruct what you had. Databento publishes data-integrity issues and exposes dataset conditions and modification dates through its API, so you can at least identify affected dates and decide what needs rerunning.

A curator serving daily-bar backtesters reasonably optimises for a clean current database over an auditable one. But if your research standard requires that a result be reproducible against its original inputs, then a silently mutating archive is a problem you have to solve yourself.

Most serious retail operations I know of end up running both: a curator for universe construction and the daily research loop, and a collector for when a result has to survive contact with realistic execution assumptions.

Pitfalls to plan for

Volume changes meaning in December and price doesn’t. Any signal using dollar volume, volume-based liquidity screens, relative volume, or volume-weighted anything will see a level shift on 6 December. The shift is in the definition of the field. The market didn’t do anything. Anything comparing across the boundary needs to know the boundary exists.

“Daily bar” becomes ambiguous. Does the bar span the regular session, the old 4:00 am to 8:00 pm extended window, or the new 23-hour session? All three are defensible and vendors will diverge.

Availability timing moves. A complete full-session bar labelled Monday isn’t finished until 8:00 pm Monday. If your backtester hands it to a strategy at the regular close, then you’ve supplied four hours of future information. The same trap exists today with post-market bars. It’s about to get much larger.

Date labelling becomes a data-model decision. Monday’s session spans two New York calendar dates and two UTC dates. Grouping by local calendar date splits sessions; grouping by UTC date puts Monday’s final hour into Tuesday. Keep the execution timestamp for sequencing and the exchange trading date for session membership, and never let the date label change when a trade happened.

History mustn’t be restated under the new rules. Version the session definition with an effective-from date. Applying December’s 23-hour session to 2015 produces a bar structure that never existed.

The overnight NBBO is a different animal. It arrives on the same message fields, assembled from a subset of venues that chose to participate, in thin conditions. Any execution model that treats a consolidated quote as fillable size will be badly wrong overnight, and any gap-risk model calibrated on dynamic regular-hours LULD bands doesn’t transfer to the static overnight ones.

Options and the underlying now run on different clocks. A hedge that assumes you can adjust an options position against an overnight move in the underlying has no market to trade in for most of the night. And options data acquires its own extended-hours prints, on the same OPRA channels as the regular session, distinguished only by the Extended Hours “v” condition.²¹ A daily options bar built without a sale-condition filter will start picking up 7:30 am and 4:10 pm prints in its high and low, contrary to how the exchange itself defines them.

Futures bars carry the same ambiguity, and since May a weekend. A futures daily bar is already a choice between day session and full session, and between settlement and last print. For CME cryptocurrency futures, and from July and August for 1-ounce gold and 10-barrel WTI, Monday’s trade date now absorbs the whole weekend.¹⁵ Anything that treats one daily bar as one day’s trading, from range estimators to volume averages, has a built-in day-of-week distortion.

Your provider may say nothing. At the time of writing, vendor communications are thin. Massive has published a clear explainer but committed only to preparing support with details to follow.⁵ Tiingo treats the overnight session as a separate product sourced from Blue Ocean rather than something folded into its EOD service, and its own vendor survey found documented extended-hours coverage stopping at 8:00 pm across most retail APIs.³ Norgate, CSI, and similar shops appear to have published nothing at all. I wouldn’t take that silence to mean nothing is changing.

Questions to ask your provider

If you rely on a reseller’s decisions, then these are the questions I’d put to them before 6 December:

• Session definition. What window will the daily OHLC bar cover after 6 December, and what window does it cover today? Under what timezone and trading-date convention?
• Volume. Which sale conditions contribute to volume and turnover, and will overnight prints be included? Dated to which session?
• History. Will pre-December history be restated under the new convention, or preserved under the old one? How will I tell which I have?
• Availability. At what time is a completed daily bar published, and does that time change?
• Corrections. Are Prior Day messages and late corrections applied retrospectively? Is there any notification or version record?
• Options. Does the options feed expose OPRA sale conditions, and does the daily options bar exclude GTH and Curb prints from the high and low? Are the Cboe smoothed-NBBO marking files carried at all?
• Overnight coverage. Which venues contribute? Is the overnight NBBO carried at all, and is it distinguishable in the data from the regular-session NBBO?
• Futures. Is the daily close the exchange settlement or the last print? Which session does the bar span, and is the day-session series a separate product? For 24/7 contracts, do weekend trades follow the exchange trade date or the calendar date?

The answers tell you more about a provider than the number of exchanges on its marketing page.

Taking data management seriously

The meaning of a price series is a moving target, set by a regulatory and commercial apparatus that has no obligation to tell you when it moves. Today’s bar is the product of 50 years of accreted rules, and a plan amendment filed in January and approved in June will change it in December.

Treat the data layer as infrastructure with an owner. Most of what that involves has already come up: a canonical instrument identity separate from any vendor’s symbology, session and adjustment definitions kept as dated facts with the notice that established them, a replayable backtest with frozen inputs, and a second source obtained early enough that switching is still possible.

The alternative is the default: build years of research on one provider’s conventions, never learn what they are, and find out the expensive way that something moved. These changes don’t announce themselves. The bars will still be well-formed, and the volume will look plausible because it is a whole session’s volume. It’s just a session that started more than 12 hours earlier than you expected.

If all of this seems overly complicated, it’s worth remembering the wisdom of H.L. Mencken from 1917: “for every complex problem there is an answer that is clear, simple, and wrong.”