c·t = CONST, THAT CLICKS BONUS ④ ／ dug after the main series closed

A 4.1-bit prediction beats a 15.7-bit discovery

"Theory first" meant
fixing the measure first Being near 137, degeneracies among constants, and the price of searching for relations.
All three collapsed onto one question.

What you need: Episode 19's scale, Episode 36's band, Episode 37's running, Episode 47's map, bonus ③'s typesThe sixth compression — three questions into one

"What would change if $1/\alpha$ were exactly 137?" — and "is changing this dimensionless quantity ever the same as changing that one?" These turned out to be one question: is there hidden structure among the constants? Digging led all the way to a 4.1-bit prediction beating a 15.7-bit discovery.

01First, 137 — a vast exclusion and almost no physical difference

CODATA 2022 $$\frac1\alpha=137.035999177(21)\qquad\Longrightarrow\qquad \frac{0.035999}{2.1\times10^{-8}}=\mathbf{1.7\times10^{6}\,\sigma}$$

yet as a value of $\alpha$ it differs by only 0.026 per cent

What	Shift	$\sigma$
Electron g−2	$3.05\times10^{-7}$	$2.3\times10^{6}$
Hydrogen 1S–2S	1.3 THz	$1.3\times10^{11}$
Neutron lifetime	$+1.53$ s	$3.1$
BBN helium $Y_p$	$7.4\times10^{-5}$	$0.025$ (invisible)
The Hoyle state	──	needs 4 per cent; this is 1/152 of that

The only thing that bites in the world is the neutron lifetime (the electromagnetic part of the $n$–$p$ mass difference, $-1.04$ MeV, shifts by 0.27 keV; $\tau\propto Q^{-5}$ gives 1.5 seconds) — and only at 3$\sigma$. The exclusion is vast because the measurements are precise, not because the physics is delicately balanced.

02So why does 137 nag? — because it is an integer

An integer would make $\alpha$ a count, not a ratioin bonus ③'s four types, type N rather than type R

1929

That is exactly what Eddington bet onfirst 136, then 137

◎

Measured by Episode 19's practice: 3.8 bitsuniform on $[136.5,137.5]$, landing within $\pm0.036$ of an integer has probability 0.072 — the bottom of the band of coincidences

Conclusion of §01–02

Eddington's mistake was treating a 3.8-bit coincidence as a 0-bit identity.
And there is a sharper objection — $\alpha$ runs (Episode 37; it is 128 at $M_Z$).
Without saying at which scale it is an integer, the claim is not yet a sentence.
── In his day the running was unknown, so it was a sentence then and is not one now.

03The other question — degeneracy comes in three kinds

Type	What it is	Really	Finding one tells you about
A	indistinguishable in principle	redundancy of notation	your notation
B	current instruments cannot separate them	limits of the tools	your instruments
C	thought independent, actually related	structure of nature	physics

04Type A — the question itself is real

The cleanest example $$\text{only the combination}\quad \bar\theta=\theta+\arg\det M_q\quad\text{is physical}$$

move $\theta$ and rotate the quark phases at the same time — and nothing happens

The CKM matrix is the same: of a 3×3 unitary matrix's nine parameters, rephasing removes five directions that are unobservable in principle ($26.8$ bits' worth of pure notation).

Conclusion of §04

But Episode 47's 19 parameters are the count after these are removed.
"No exact degeneracies remain" is the answer, and only because we constructed it that way.
── Finding a type A degeneracy tells you about your own notation, not about nature.

The oddest thing that fell out $\bar\theta$ is physical only because no quark mass is zero. Were $m_u=0$, $\bar\theta$ could be rotated away entirely and the strong CP problem would vanish (measured: $m_u=2.16\pm0.11$ MeV, excluding zero at $19.6\sigma$).
── The 19 parameters are not fully independent. Remove one Yukawa and $\bar\theta$ goes with it. The number of constants depends on the values of the constants.

05Type B — and one that was a mirage

◎

Real: the $m_q$ direction in spectroscopybonus ① — a condition number of 1196, still with us

✗

Mirage: $\alpha$ and $m_e$ in atomic physicsgross structure $\propto\alpha^2m_e$, fine $\propto\alpha^4m_e$ — but take the ratio and $m_e$ drops out

In fact $m_e$ never appears in a dimensionless observablethe ratio of two transitions in one atom always kills it — a mirage created by writing dimensionful quantities (Episode 3)

◇　◇　◇

06The core — searching for type C has a price

A type C relation, if found, is a discovery in physics. So may one simply search? The instinct is to call that numerology — but where exactly is the difference?

Look-elsewhere $$\text{having tried }M\text{ candidate relations, the threshold for meaning is }\log_2 M\text{ bits}$$

Reading	Search space $M$	Threshold	Koide (15.7 bits)
Narrow (only Koide's form)	$126$	$7.0$ bit	passes
Wide (subsets of 12 masses, etc.)	$1.4\times10^{8}$	$27.1$ bit	11.4 bits short

The main point of this episode

b–τ unification was predicted in advance by SU(5), so there was one candidate — a threshold of zero bits.
In raw surprise Koide (15.7) beats b–τ (4.1) by 11.6 bits, but
subtract the price of the search and it reverses.
── The difference between "theory first" and "numbers first" comes out as exactly $\log_2M$ bits.
Not a social convention — the result of a subtraction.

Figure: subtract the price of the search from the raw surprise and the ranking flips. Move the slider for how widely you searched — raising $M$ sinks Koide alone, while b–τ, predicted in advance, does not move.

breadth of the search, $\log_2 M$ 7

Koide's relation (found by searching) b–τ unification (predicted first) the line of meaning (0 bits)

07And this was the same disease as naturalness

Setting	The "space" required	Since
Ep. 19: surprise	the quantity's prior range	flagged in Ep. 19 §01
Ep. 48: naturalness	the parameter's prior	became a theorem in bonus ③
Here: numerical coincidence	the search space	the verdict flips with the declaration

Conclusion of §07 — the sixth compression

All three were one question: is there a canonical measure?
A probability needs a measure, and without one there is no probability.
── A search space is finite if you declare a bound. Declaring it first is the only way to make the question well-posed.
"Theory first" was never a social convention — it was fixing the measure in advance.

08A bonus — Episode 36's "band" now has a threshold

Episode 36 observed that interesting coincidences cluster at 4–7.5 bits and called it a selection effect. Now the threshold can be computed (7–27 bits) — the band sits far below it, which is why none of them mean anything. Koide alone escaped the band, and is still 11.4 bits short on the wide reading.

Conclusion of §08

An observation became a subtraction.

The honest line

(1) The weakest point is §06's estimate of the search space. The narrow and wide readings differ by 20 bits, and there is no way to decide between them — which is precisely §07's claim.

(2) Koide's 15.7 bits is inherited from Episodes 19 and 36 and depends on the prior range, as does b–τ's 4.1.

(3) b–τ's "zero search space" is an idealisation. There are other grand unified candidates, and counting which models were tried would add a few bits here too — "small", not "zero".

(4) §01's neutron-lifetime estimate is rough. The electromagnetic part of the $n$–$p$ mass difference ($\approx-1.04$ MeV) varies between lattice calculations, and $\tau\propto Q^{-5}$ is a phase-space approximation — do not read it more precisely than "of order 3$\sigma$".

(5) §07's unification is this series' reading, not the standard formulation in statistics — multiple-comparison corrections and Bayes factors are finer existing machinery.

Exercises

If $1/\alpha$ were exactly 137, what in the world would change?
Show the answer
The neutron lifetime alone ($+1.53$ s, 3.1$\sigma$). In the laboratory g−2 sees it at $2.3\times10^6\sigma$ and 1S–2S at $1.3\times10^{11}\sigma$, but BBN sees 0.025$\sigma$ and the Hoyle state needs 152 times more. The exclusion is vast because the measurements are precise.
What was Eddington's mistake?
Show the answer
Treating a 3.8-bit coincidence as a 0-bit identity. And today there is more: $\alpha$ runs (128 at $M_Z$), so without naming the scale the claim is not a sentence — it was one in his day and is not one now.
What do the three types of degeneracy each tell you?
Show the answer
A (exact) — about your notation; B (observational) — about your instruments; C (hidden relation) — about physics. The test is "could observation break it?" A cannot; C can (if b–τ failed, SU(5) would die).
Why does a 4.1-bit prediction beat a 15.7-bit discovery?
Show the answer
Because you subtract $\log_2M$, the price of the search. b–τ was predicted first, so one candidate and a threshold of zero; Koide was found by searching, with a threshold of 7–27 bits — on the wide reading it falls 11.4 bits short. The "theory first" difference comes out as exactly $\log_2M$ bits.
(Harder) What do surprise, naturalness and numerical coincidence share?
Show the answer
All three require a "space" — a prior range, a prior, a search space. All three were the one question "is there a canonical measure?" (the sixth compression). And "theory first" means fixing the measure in advance.

Summary: "theory first" meant fixing the measure first

$1/\alpha=137$ exactly is excluded at $1.7\times10^6\sigma$, yet the physical difference is only 0.026 per cent — the sole effect in the world is a 3$\sigma$ shift in the neutron lifetime. It nags because 137 is an integer, and the surprise of that nearness is 3.8 bits, the bottom of the band. Eddington's mistake was treating a 3.8-bit coincidence as a 0-bit identity — and since $\alpha$ runs, the claim today is not even a sentence.

"Changing this is the same as changing that" came in three kinds. Type A is real — $\bar\theta=\theta+\arg\det M_q$, and the CKM's five directions. But the 19 parameters are the count after those are removed, so finding a type A degeneracy tells you about your own notation. Type B survives as bonus ①'s $m_q$ direction, while atomic physics' $\alpha$–$m_e$ was a mirage made by writing dimensionful quantities.

And searching for type C has a price. Having tried $M$ candidates, the threshold is $\log_2M$ — Koide's 15.7 bits passes the narrow reading (7.0) and falls 11.4 short of the wide one (27.1). But b–τ unification was predicted first: one candidate, zero threshold. A 4.1-bit prediction beats a 15.7-bit discovery.

Finally — surprise (prior range), naturalness (prior) and numerical coincidence (search space) were one question: is there a canonical measure? (the sixth compression). "Theory first" was never a social convention; it was fixing the measure in advance. And Episode 36's band now has a threshold — an observation became a subtraction.

On to bonus ⑤ §07 collapsed everything onto "is there a canonical measure?". So — is there? Bonus ③ answered "compact gives Haar, non-compact gives nothing", but that was too coarse. Next we follow the thread that the renormalisation group hands out the measure — and score all 20 Standard Model parameters by the shape of their beta functions, hitting the three great fine-tuning problems with zero false positives.

This document is bonus episode ④ of "c·t = const, That Clicks", written after the main 50 episodes closed, for physics-minded high-school and university readers. The numbers are computed in kenshou/calc60.py, calc61.py and calc62.py. Eddington's numerology, CKM rephasing, the $\bar\theta$ combination, b–τ unification and the look-elsewhere effect are all standard material. The weakest point is §06's estimate of the search space — the narrow and wide readings differ by 20 bits and there is no way to decide between them (which is precisely §07's claim). Koide's 15.7 bits and b–τ's 4.1 both depend on the prior range, and b–τ's "zero search space" is an idealisation (counting other grand unified candidates would add a few bits). §01's neutron-lifetime estimate is rough: the electromagnetic part of the $n$–$p$ mass difference varies between lattice calculations and $\tau\propto Q^{-5}$ is a phase-space approximation — do not read it more precisely than "of order 3$\sigma$". §07's unification is this series' reading rather than the standard formulation in statistics, where multiple-comparison corrections and Bayes factors are finer existing machinery. ── To make a PDF, use your browser's Print dialogue (sliders freeze and answers are hidden in the print version).

Print / PDF: ⌘+P (Ctrl+P on Windows). On screen, widen the search space and watch the ranking flip. "Show the answer" opens each solution.

What	Shift	\(\sigma\)
Electron g−2	\(3.05\times10^{-7}\)	\(2.3\times10^{6}\)
Hydrogen 1S–2S	1.3 THz	\(1.3\times10^{11}\)
Neutron lifetime	\(+1.53\) s	\(3.1\)
BBN helium \(Y_p\)	\(7.4\times10^{-5}\)	\(0.025\) (invisible)
The Hoyle state	──	needs 4 per cent; this is 1/152 of that

Reading	Search space \(M\)	Threshold	Koide (15.7 bits)
Narrow (only Koide's form)	\(126\)	\(7.0\) bit	passes
Wide (subsets of 12 masses, etc.)	\(1.4\times10^{8}\)	\(27.1\) bit	11.4 bits short

"Theory first" meantfixing the measure first Being near 137, degeneracies among constants, and the price of searching for relations.All three collapsed onto one question.

01First, 137 — a vast exclusion and almost no physical difference

02So why does 137 nag? — because it is an integer

03The other question — degeneracy comes in three kinds

04Type A — the question itself is real

05Type B — and one that was a mirage

06The core — searching for type C has a price

07And this was the same disease as naturalness

08A bonus — Episode 36's "band" now has a threshold

Summary: "theory first" meant fixing the measure first

"Theory first" meant
fixing the measure first Being near 137, degeneracies among constants, and the price of searching for relations.
All three collapsed onto one question.