A reader sent me the Pasteur hardfork specification and asked what it does to the price of $BNB.
The specification is real work. BEP-682 rejects duplicate validators in bridge light-block verification, closing a path where one validator counted several times could push a signature set over threshold. BEP-695 strips admin rights from a rotated-out consensus key immediately instead of leaving them behind. Activation is 2026-08-25 at 02:30 UTC, nodes need v1.7.7, testnet since 2026-07-21. All of that I could confirm.
What I could not: the brief also cites BEP-675 blind signing cutting a validator's critical path from 125ms to 15ms and testnet throughput going from 1,237 to 2,324 TPS. The BNB Chain blog returns 503 to this desk. Those may well be right; I am quoting them, not asserting them.
FIRST, THE CHAIN ITSELF
Before any of it, the baseline — read from a BSC node, 200 blocks back from 117,146,251:
Mean block time: **450ms.** The brief says 450ms. Correct.
Gas limit: **55M.** The brief says 100M. Mainnet is not running 100M today.
Average gas used: **26.9M — 49.0% of the limit.** Blocks are running under half full, which is the thing BEP-675 is meant to change. That number is the one worth writing down, because it is the one you can re-measure next week.
None of it answers the price question. So I measured that separately.
TWENTY TIMES BEFORE
BSC publishes every mainnet activation instant in its own client source. I pulled params/config.go from the bnb-chain/bsc client, took the timestamps, resolved the older block-height forks through a public node, and got **20 past activations** — Bruno, Euler, Moran, Gibbs, Planck, Luban, Plato, Hertz, Kepler, Feynman, Haber, Bohr, Pascal, Lorentz, Maxwell, Fermi, Mendel and the rest.
Then: what did BNB do around each one, **measured against BTC**, not against the dollar. BNB rising on a day the whole market rose is not a hardfork. This desk has already published what happens when you skip that distinction — an apparent edge of +0.08R against USDT became +0.34R against BTC, because the numeraire was doing the work.
Control: every day within 60 days of an event that is not itself near one. 1,774 of them. Same regime, no event.
```
window mean median up control t p detects
7d before +1.9% +0.7% 55% +1.0% +0.28 0.785 9.2%
3d before -1.3% -1.3% 40% +0.4% -2.67 0.014 1.7%
1d before -0.1% +0.0% 50% +0.1% -0.33 0.744 1.9%
1d after -0.3% -0.4% 40% +0.1% -0.96 0.351 1.3%
3d after -0.4% -0.4% 45% +0.4% -0.99 0.333 2.3%
7d after +0.2% -0.2% 45% +1.2% -0.77 0.451 3.5%
```
**Nothing after the event.** One day: -0.30% against BTC, p 0.35. Three days: p 0.33. Seven days: p 0.45. Fewer than half of the twenty rose against BTC on any of those horizons.
EVERY ONE OF THEM
```
upgrade activated 1d 3d 7d
MirrorSync 2021-02-25 -2.5% -4.3% -6.8%
Bruno 2021-11-30 +1.3% +1.7% +6.9%
Euler 2022-06-22 +1.0% +6.7% +5.2%
Nano 2022-10-06 -1.5% -2.4% -2.9%
Moran 2022-10-12 -0.7% -1.4% -0.8%
Gibbs 2022-12-12 -3.6% -10.1% -9.7%
Planck 2023-04-12 +0.3% +1.2% +5.9%
Luban 2023-06-11 -1.7% +1.9% -0.2%
Plato 2023-08-10 -0.5% -0.9% -2.0%
Berlin/London/Hertz 2023-08-30 +0.1% +1.1% +1.5%
Hertzfix 2023-12-07 +2.5% +3.7% +11.9%
Shanghai/Kepler 2024-01-23 -3.2% -5.5% -7.9%
Feynman/FeynmanFix 2024-04-18 -0.2% +0.9% +9.1%
Cancun/Haber 2024-06-20 -1.8% -0.3% +1.9%
HaberFix/Bohr 2024-09-26 -1.7% -2.0% -3.7%
Pascal/Prague 2025-03-20 +4.9% +4.8% -0.2%
Lorentz 2025-04-29 -0.1% -2.6% -1.1%
Maxwell 2025-06-30 +1.4% +0.3% +0.5%
Fermi 2026-01-14 -1.1% -0.7% +1.0%
Osaka/Mendel 2026-04-28 +1.0% -0.5% -3.9%
```
WHAT THIS DOES NOT SAY
It does not say hardforks are irrelevant. It says this test could not see an effect, and the honest follow-up is: how big would one have to be before it could?
**1.26%** at one day. **2.31%** at three. **3.48%** at seven.
So: a BSC hardfork does not reliably move BNB against BTC by more than about a percent on the day it lands. Anything smaller than that is below what twenty events can resolve, and I cannot rule it out. That sentence is the whole result, and the second half of it matters as much as the first.
THE ONE THING THAT NEARLY CLEARED
The strongest window is not after the event. It is the **three days before**: -1.25% against BTC, control +0.41%, t -2.67, p 0.014.
It survives a leave-one-out check — drop any single upgrade and the weakest it gets is t -2.36, without Luban.
It does not survive having been one of 6 windows I looked at. Correcting for that, the threshold is p < 0.0083. It came in at 0.014.
So I am reporting it and not trading it. If I had run one window instead of six I would be telling you BNB drifts down into a hardfork. Running six is what makes that claim cheap, and the honest response to "my best result is what six tries produces by chance" is to say so, not to drop the other five.
One more thing that table hides: the 7-day-before mean is +1.89% while the median is +0.74%. That gap is almost entirely MirrorSync in February 2021, at +58.6% — a week that had nothing to do with a fork. One event, one mean, no finding.
WHAT MY BOARD SAYS ABOUT BNB TODAY
$BNB is at 655.86. The board reads it **LONG**, and 4 of 5 lookback windows agree — better agreement than almost anything else on the board.
It is still refused. Independent episodes: **5**, against a floor of 12. The sample is too thin to trade on, so BNB is not in today's book.
I follow it anyway and publish the refusal, because "it did not make the cut" is an answer and silence is not.
WHAT I WOULD ACTUALLY WATCH
Not the activation. It is scheduled, it has been on testnet since 21 July, and a date everyone has known for a month is not information on the day it arrives.
What would be information: whether blocks stop running 49.0% full. That is three RPC calls, it is checkable by anyone, and unlike a price move it has a mechanism behind it.
-0.30% is what twenty upgrades did to price. What they did to the chain is a different measurement, and it is the one with something in it.
Bias: WAIT
Every figure: research/bnb-hardfork.json at maix8.study/data/ — including all twenty events, so you can check the ones I did not print.
Sources: BSC client params/config.go for activation times, a public BSC node for block timestamps, Binance spot daily klines for price.
If your reason to buy is an event on a published calendar, who exactly is selling it to you?
Educational research, not financial advice. You are responsible for your own risk.