Bitcoin · power law · the floor
This whole site is named after a floor, so it is worth being exact about what that floor is and where it comes from. The short answer surprises people. The floor is not in Giovanni Santostasi's founding power-law theory article. It is a single number layered on top of his trend line, added in his later charting work, and almost everyone who draws a floor is drawing the same line.
01 / THE STARTING POINT
Santostasi's founding power-law theory article defines one thing, the central fair-value line, a straight line through the middle of the data on a log-log chart.
That matters for everything below. A regression line through the centre tells you the trend. By itself it does not contain a lower bound. So a floor is never something the fair-value equation hands you. A floor is always an extra decision someone adds on top.
To be fair to Santostasi, he did add one later. His own BGeometrics charts draw a support floor, and the corridor framing was carried further by Harold Christopher Burger. The point here is narrow and precise. The floor lives in the later operational work, not in the founding equation. When this page says the floor is not in Giovanni's original math, it means exactly that, and nothing more. It is not a claim that he never drew a floor.
02 / THE REAL FLOOR
The floor that people actually mean when they say "the power-law floor" comes from BGeometrics, which is Santostasi's own charting implementation. Its rule is as plain as it gets.
The justification is empirical, not a distribution. In their own words the minimum price of Bitcoin has historically been about 58 percent below its fair price, so the floor is the fair line scaled by 0.42, which is 1 minus 0.58. It is pinned to the single worst drawdown ever recorded relative to trend. They also note this floor has been breached only once, on 13 March 2020, for a few hours during the pandemic crash.
So the honest description of the floor is a worst-case multiplier. It is not a 2.5 percent percentile, not a minus 2 sigma band, and not a separate regression through the cycle lows. It is one scalar, chosen because price has never stayed below it.
03 / THE CHECK
Rather than trust the description, the claim was checked directly against the two data series BGeometrics ships to its chart, the fair-value series and the floor series. Dividing the floor by the fair value at every point gives the multiplier they actually use.
| Date | Fair value | Floor | Floor / Fair |
|---|---|---|---|
| 2011-01-01 | $0.45 | $0.19 | 0.420000 |
| 2018-09-23 | $4,505 | $1,892 | 0.420000 |
| 2026-06-25 | $136,841 | $57,473 | 0.420000 |
Across all 5,645 daily points in the series, from 2011 to the forecast horizon, the ratio is exactly 0.420000. Minimum 0.42, maximum 0.42, mean 0.42. It is a flat scalar applied uniformly, confirmed at the data level rather than taken from a blog or a marketing line.
04 / THE PUNCHLINE
Once you convert each version of "the floor" into a single multiplier on fair value, they collapse onto each other. The differences are cosmetic.
| Source | Method | Multiplier |
|---|---|---|
| BGeometrics | worst-drawdown multiplier | 0.420 |
| This site's model | minus 2 sigma band (sigma 0.20) | 0.398 |
| Typical dashboards | varies, often labelled a percentile | ~0.37 to 0.42 |
This site's own floor, derived as a minus 2 sigma band, lands at 0.398, within five percent of the BGeometrics 0.42. They are the same line reached by two routes. So when a third-party dashboard quotes a floor near 0.4 times fair value, it is, at its core, the same Giovanni-based floor. The trend underneath is real power-law math, and the floor is the standard scale-it-down method on top.
05 / THE ONE THING TO WATCH
Calling the floor "fair value times 0.42, the worst historical drawdown" is honest and self-describing. It says exactly what it is and how it was chosen. You can agree or disagree with using the worst case, but you cannot be misled by it.
Some dashboards take the same kind of flat multiplier and present it as a measured 2.5 percent percentile, as if the bottom 2.5 percent of historical deviations had actually been counted. That implies a statistical pedigree the number does not have. A real 2.5 percent floor requires the distribution of price deviations to be measured, and those deviations are not normal, so a percentile read off an assumed bell curve will not match the real one. The floor can be perfectly reasonable while the label on it is wrong.
Honesty means turning the same test inward. This site's own floor is written as a minus 2 sigma band, and that is also a convenient placing, not a measured percentile. The sigma of 0.20 was chosen so the band sits on the historical drawdowns, which is exactly why it lands at 0.398, right next to the BGeometrics 0.42. So read this site's minus 2 sigma the way you should read anyone's, as a tidy way to position a multiplier near 0.4, not as a counted statistic. The number is sound. The sigma is shorthand.
The test is simple, and it applies to this page too. If a floor is presented as a percentile or a sigma, ask what distribution it was measured against. If the answer is that the fair line was multiplied by a fixed number, or that the sigma was chosen to fit the lows, then it is a multiplier in a statistician's coat, this site's floor included, and it should be read as a multiplier.
06 / THE UPDATE
After this page was written, Santostasi published the rigorous version of exactly the test asked for above. Rather than assume a shape for the spread, he fitted the distribution of deviations from the trend, log price minus the power law, across the full history (August 2010 to now, exponent 5.65, R squared 0.96). See the post.
The deviations are right-skewed, skewness +0.84. Price is not log-normal around the trend. A normal fit is rejected (KS 0.091) in favour of a skew-normal (KS 0.047, a large AIC gap), and the Q-Q plot confirms it. The normal bends away at both ends while the skew-normal hugs the line.
A symmetric normal overstates the floor, placing it too deep at minus 1.96 sigma. The real, skew-normal 2.5th percentile sits at minus 1.42 sigma, about 2.6 times below trend, near $50,000 at today's roughly $133,000 trend. That is why cycle lows cluster tightly rather than scattering. The upside is the opposite, the right tail is genuinely fatter, with blow-off tops reaching plus 3 to plus 4 sigma.
This is the measured 2.5 percent floor this page asked for, and it confirms the central point rather than denting it. The deviations are not normal, so a floor read off a plain bell curve is misplaced. The trend fit itself differs a little between implementations. Giovanni's post rounds to about $133,000, the BGeometrics data gives about $136,800, and this site's own model gives about $141,100, all within roughly six percent. The table below puts the floors on Giovanni's fit and on this site's fit side by side.
| Floor method | Depth | On Giovanni ~$133K | On this site ~$141K |
|---|---|---|---|
| Naive normal 2.5% (minus 1.96 sigma) | divide by 3.85 | $34,500 | $36,700 |
| Giovanni skew-normal 2.5% (minus 1.42 sigma) | divide by 2.66 | $50,000 | $53,100 |
| This site, minus 2 sigma | divide by 2.51 | $53,000 | $56,200 |
| BGeometrics, times 0.42 | divide by 2.38 | $55,900 | $59,300 |
The three real floors cluster tightly, between $50,000 and $56,000 on Giovanni's trend, or between $53,000 and $59,000 on this site's slightly higher trend. Only the naive normal is the outlier, far too deep at around $35,000, which is the exact trap described in the section above. Notice too how close this site's own minus 2 sigma shorthand sits to Giovanni's freshly measured skew-normal floor, within about six percent on either fit. The number was sound before it was measured. Now it has a measured backing.
This is testable, and the test is brutal. By construction a normal distribution sits below its minus 2 sigma line about 2.3 percent of the time. Across Bitcoin's roughly 15.9 years of price history that is about 132 days the price should have spent below a minus 2 sigma floor. A skew-normal predicts about 4 days. Bitcoin has actually spent close to zero, a single brief breach during the 13 March 2020 crash. The normal is wrong by a factor of roughly thirty. The skew-normal matches reality. That gap, 132 expected days against almost none observed, is the skew written in a single number.
One quiet lesson falls out of the two columns. The choice of trend fit moves the floor by about six percent, which is roughly the same size as the gap between one floor method and the next. The line you fit through the middle matters as much as the rule you use to drop below it.
For where this sits today, in Giovanni's own reading the price near $59,000 is at minus 1.17 sigma below his $133,000 trend, left of centre but well inside the body of the distribution. On this site's higher $141,000 trend the same price reads a little deeper and sits close to the cruder 0.42 floor, while staying above the more rigorous skew-normal floor near $53,000. Either way it is routine deep-cycle territory, not a breach.
07 / THE VERDICT
The floor is the fair-value power law scaled down by a flat factor of about 0.42, chosen because Bitcoin's price has never durably traded below it. It sits on genuine Santostasi math, the trend line, but the multiplier itself is a downstream modelling choice, added in the later charting work rather than the founding theory article.
That is why the floors you see around the web all agree on roughly where the floor is. They are all multiplying the same trend by roughly the same number. They differ only in the story they tell about it. The plainest story is the BGeometrics multiplier, fair value times 0.42. The most rigorous, Giovanni's measured skew-normal 2.5th percentile, lands a touch deeper at about 2.6 times below trend. Both point to the same neighbourhood, near half of fair value.
Which is how this site treats it. Anchor to the floor because, breached once in fifteen years, it is the most reliable level we have, while remembering it is a regularity, not a promise, and that the math behind the trend it rests on survives attack but was never proof.
This is an explanation of how a chart level is constructed, not financial advice. The fair-value and floor figures are read from the BGeometrics power-law data series and checked arithmetically. Numbers move as the model updates. Read Santostasi's original theory and the BGeometrics chart and judge for yourself.
Write a page explaining what the Bitcoin power-law "floor" actually is. Key points to establish, in order: 1. Giovanni Santostasi's original power-law theory defines only the central fair-value line (1.0117e-17 * days^5.82). It contains NO floor, no support line, no sigma band, no percentile. A floor is always an addition on top. 2. The operational floor everyone means comes from BGeometrics (Santostasi's own charting site): floor = fair value * 0.42, justified empirically as Bitcoin's worst-ever drawdown (about 58% below fair). Breached only once, 13 March 2020, intraday, during the COVID crash. 3. Verify the 0.42 against the raw data, not the prose: divide the shipped floor series by the fair-value series. It is exactly 0.420000 across all 5,645 daily points (min = max = mean = 0.42). 4. Every common floor is basically the same line once expressed as a multiplier: BGeometrics 0.42, this site's -2 sigma model 0.398, typical dashboards ~0.37 to 0.42. They agree on where the floor is, differing only in the story. 5. The one caveat: a fixed multiplier is honest and self-describing. Relabelling the same scalar as a "2.5% percentile" or "-2 sigma" implies a statistical pedigree it does not have, because the deviation distribution is not normal. If a floor is called a percentile, ask what distribution it was measured against. Verdict: the floor is a fixed multiplier (~0.42) on Giovanni's genuine trend line, anchored to the worst historical drawdown. It rests on real Santostasi math but the multiplier itself is a downstream modelling choice, not in the original theory. Style: match support_math_disproved.php (Fraunces / Hanken Grotesk / JetBrains Mono, dark theme, .eq / .claim / .verdict-box / .caveat blocks). Follow the site rule: no em dashes, and no semicolons inside sentences. Wrap data tables in a .table-scroll div for mobile. Source it to Santostasi (Medium) and the BGeometrics chart only, no third-party blogs.