On Friday 28 August 2026, bitcoin fell 3.00 per cent and hundreds of headlines said the Federal Reserve Chairman’s Jackson Hole speech was why. That is probably true. It is also, as written, unfalsifiable — and the difference between “probably true” and “checked” is about twenty minutes of work that almost nobody does.
This is field guide #35. It is a procedure for answering one question: did this event actually move this market, or did it merely happen nearby? The procedure is deliberately mechanical. You do not need a view, a model or a data subscription. You need a clock, an hourly price series, the primary source document, and the discipline to write down what would prove you wrong before you look.
Why “X caused Y” is the most common bad claim in financial media
Markets move every day. Events happen every day. Any given day therefore offers a menu of true statements of the form “the market fell and this thing occurred,” and the writer picks whichever pairing makes the cleanest sentence. The result reads like an explanation and functions like a coincidence.
It is not a conspiracy or even laziness. It is a deadline problem. The daily close is published, the event is on the calendar, and the causal sentence writes itself in four seconds. Checking it requires a different data granularity than the one the story is written at — and that is the whole trick. A causal claim about an event that lasted thirty minutes cannot be tested with a number that summarises twenty-four hours.
Step 1 — establish the clock, in one time zone, before anything else
Write down the event time in UTC. Convert everything else to UTC too. Most attribution errors we have caught on this site were, at bottom, time-zone errors dressed as analysis.
For the Warsh example: the Kansas City Fed’s press release specified the keynote for Friday 28 August at 10:00 a.m. EDT / 8:00 a.m. MDT. That is 14:00 UTC. The published agenda put the next paper at 8:30 a.m. MDT, so the entire keynote falls inside a single 14:00 UTC hourly candle. That constraint — the event fits in one hour and we know which hour — is what makes the rest of the procedure possible.
Do this before you look at any price. If you establish the clock after seeing the chart, you will unconsciously choose the boundary that flatters the story.
Step 2 — pull the hourly bars, not the daily close
The daily close tells you the day was down. It cannot tell you when, and “when” is the entire question. Pull hourly open, high, low, close and volume for the whole session, from several hours before the event to several hours after.
| Hour (UTC) | ET | Open | High | Low | Close | Volume (BTC) |
|---|---|---|---|---|---|---|
| 08:00 | 4am | 79,660 | 79,840 | 79,480 | 79,517 | 526.2 |
| 13:00 | 9am | 79,421 | 79,635 | 79,035 | 79,331 | 843.2 |
| 14:00 | 10am — event | 79,330 | 79,707 | 78,423 | 79,542 | 2,930.1 |
| 15:00 | 11am | 79,542 | 79,839 | 78,313 | 78,331 | 1,255.0 |
| 16:00 | 12pm | 78,331 | 78,404 | 76,888 | 77,788 | 2,955.6 |
| 17:00 | 1pm | 77,788 | 78,132 | 77,643 | 77,904 | 722.8 |
Binance BTCUSDT 1h klines, own pull. Day’s average hourly volume: 823.2 BTC.
Now read the event hour on its own terms. It opened at 79,330 and closed at 79,542 — up 0.27 per cent. It traded down to 78,423 during the remarks and that low was bought back before the hour ended. The following hour made a higher high than the event hour. The day’s low came in the hour beginning 16:00 UTC — noon Eastern, two hours after the speaker began.
That does not mean the speech was irrelevant. It means the specific sentence “bitcoin fell as the Chairman spoke” is false at hourly resolution, and any explanation you build has to accommodate a two-hour gap between the event and the break.
Step 3 — check volume, because volume is where conviction shows
Price tells you where; volume tells you how much anyone cared. Compare each hour against the day’s average hourly volume, which is simply the day’s total divided by 24.
In the example, the two largest hours were the event hour (2,930.1) and the noon hour (2,955.6) — 3.6× and 3.6× the daily average, separated by less than 1 per cent from each other. Both hours mattered. Only one of them closed lower than it opened. A useful heuristic: an event hour with high volume and a round-trip close is a market arguing; an hour with high volume and a directional close is a market deciding.
Step 4 — eliminate the competing explanations, by clock
Every big session has at least two candidate causes. List them, then time-stamp each one. Most will eliminate themselves.
On 28 August the rival story was a large options expiry: roughly $6.44bn of bitcoin options on Deribit, about 81,700 contracts, with max pain reported between $68,000 and $70,000, far below spot. Plenty of coverage blended it with the speech into a single vague pressure. But the expiry settled at 08:00 UTC — six hours before the keynote, eight hours before the low. And the 08:00 hour traded a $360 range on 526.2 BTC, 36 per cent below the day’s average hourly volume. The expiry cleared quietly and was over before New York woke up. One clock check, one eliminated explanation.
This is the highest-yield step in the whole procedure and it costs nothing. An explanation that has already happened cannot cause something later.
Step 5 — read the primary source, and count the words
If the event is a document — a speech, a filing, a release, a court opinion — get the original, not the summary of it. Then do something slightly unusual: count the words that the coverage says the document is about.
This is not a gimmick. A word count is a falsifiable, zero-cost measurement of what a document actually contains, and it routinely contradicts the reporting. Applied to the Warsh keynote, against the Board of Governors transcript: 3,584 words of delivered text; “raise” appears zero times; “hike” appears three times and all three are about walking on trails; “payment” appears zero times at a symposium titled Financial Innovation: Implications for Payments and Policy; and “stablecoin,” “bitcoin” and “digital” appear zero times in the body.
The correct conclusion from that count is not “therefore the speech was dovish.” It plainly was not; “we have work to do” is a hawkish sentence and it is in there. The correct conclusion is more precise and more useful: the market priced a commitment the document does not contain, which means the market was pricing an interpretation, and interpretations can be revised. That is a tradeable distinction and it comes from counting.
https://www.youtube.com/watch?v=DfyE1WNAHhICNBC Television, 28 August 2026 — “Warsh: A quieter Fed is better able to meet its objectives.” Useful here because the clip is the speaker describing his own communication strategy, which is the thing the market was interpreting.
Step 6 — look for the second market, and check it agrees
A real macro event shows up in more than one place, and the pattern across assets is more informative than the move in any one of them. If your explanation is true in bitcoin but false in bonds, your explanation is probably wrong.
On 28 August the front end of the Treasury curve did what a hawkish repricing implies: the two-year rose about 6 basis points to 4.298 per cent, with an intraday print near 4.32 per cent. The 30-year fell about 2 basis points, to roughly 5.168 per cent. Brent crude closed at $88.29, down 0.26 per cent. That combination — short rates up, long rates down, energy soft — is a bear flattener, and it describes a market that believes inflation will be contained, not one that fears it.
Now test the crypto story against that. “Bitcoin fell because inflation fears rose” is incompatible with a falling 30-year and softer oil. “Bitcoin fell because the discount rate applied to long-duration risk assets rose” is compatible with all three. The cross-asset check does not prove the second story. It does eliminate the first, which is most of the value.
Step 7 — state what would have proved you wrong
This is field guide #34’s rule applied to attribution. Before publishing, write the sentence that would have falsified your claim, and check whether it is checkable.
For the claim “the sustained break did not occur during the speech”, the falsifier is explicit: the event hour would have had to close below its open, and no later hour would have had to make a higher high. Both are checkable in one table by anyone with a free exchange API. That is a good claim. For the claim “bitcoin fell because of Warsh”, there is no falsifier at all — which is why it is everywhere.
The seven steps, as a checklist
| # | Step | What it eliminates |
|---|---|---|
| 1 | Fix the event time in UTC before looking at price | Time-zone errors; boundary-shopping |
| 2 | Pull hourly bars across the whole session | Daily-close reasoning about a 30-minute event |
| 3 | Compare each hour’s volume to the daily average | Mistaking a thin drift for a decision |
| 4 | Time-stamp every competing explanation | Causes that had already finished |
| 5 | Read and word-count the primary source | Reporting on a summary of a summary |
| 6 | Check a second, unrelated market | Stories true in one asset and false everywhere else |
| 7 | Write the falsifier before you publish | Claims that cannot be graded, including your own |
Four failure modes to know about
Retiming. The single most common error. Coverage says a market fell “after” an event when the fall began before it, or hours later. Always check whether the move started before the event; if it did, the event is at most an accelerant.
Window shopping. Choosing the measurement interval after seeing the data. If you measure from the pre-event high to the post-event low you can make almost any event look decisive. Fix your window in step 1 and do not move it. We have made this error — twice in one week, on a funding series and an open-interest series — and both times the fix was to pull the endpoint’s maximum depth before writing the sentence.
Single-session extrapolation. Treating one day’s extreme reading as a change of regime. Our own recent example: we described a 3.001 per cent one-day rise in open interest as “the leverage arrived,” and it round-tripped within 24 hours to a net two-day change of +15.55 bitcoin. A single-session extreme is a position, not a regime, until it survives the event it was placed for.
Liquidation circularity. Forced liquidations are reported as a cause of a move when they are largely a consequence of it. CoinGlass recorded $487.68m of crypto liquidations in the 24 hours around Friday’s speech, more than $360m of them longs, across 97,691 accounts. Those numbers describe the mechanism by which a fall became a faster fall. They do not explain why it started.
https://www.youtube.com/watch?v=xCks1GVgXmMCNBC Television, 28 August 2026 — “Warsh on inflation: The Fed’s predominant focus should be on prices.” The clip and the transcript agree; the headline interpretation is the part that needs testing.
Frequently asked questions
Where do I get free hourly bars? Most major exchanges publish candlestick endpoints without an API key. The Binance klines endpoint returns open, high, low, close and volume per interval. Whichever venue you use, use one venue consistently: exchange prices differ by tens of dollars and mixing them mid-analysis creates differences you will then try to explain.
Isn’t hourly still too coarse for a speech? Sometimes, yes. Minute bars are better if the event is a data release with a precise timestamp, like an 8:30 a.m. CPI print. For a thirty-minute speech, hourly is adequate and less noisy. Match the granularity to the duration of the event, not to how dramatic you want the chart to look.
Does any of this tell me what to do? No, and that is deliberate. Establishing that a market priced an interpretation rather than a commitment tells you where the uncertainty lives. It does not tell you which way it resolves, or when, or whether the position sizing makes sense for you. This site publishes journalism and analysis, not advice.
What is the single highest-value step if I only do one? Step 4. Time-stamping the competing explanations takes about five minutes and eliminates more bad stories than everything else combined. Friday’s $6.44bn options expiry was being blended into the speech narrative all day by outlets that never checked it had settled at four in the morning.
Applying it, one last time
Run the checklist on Friday and you end up with a claim narrower than the headlines and much harder to dislodge: a hawkish keynote repriced the September meeting from roughly 35 per cent to roughly 59 per cent; bitcoin’s sustained break came around midday Eastern rather than during the remarks; the options expiry had settled six hours earlier and cannot be responsible; the long end of the Treasury curve fell rather than rose; and the text that did all this never uses the word ‘raise’.
Every clause there is checkable by a reader with a browser and no subscriptions. That is the standard. If your explanation of a market move contains no clause a stranger could disprove, you have written a mood, not an analysis.
Disclaimer. This article is journalism and market analysis, not investment advice. Bitcoin and other digital assets are volatile and you can lose the entire amount you put in. Nothing here is a recommendation to buy, sell or hold anything. Figures are as of the timestamps stated and may be stale by the time you read them. Do your own research and, if you need advice, speak to a regulated professional.