Charged Lepton Flavour Violation: One Operator, One Number, and the Channels That Cannot See It

Introduction

The layer. The operator is a statement about the frame's gauge sector on a formal copy of the carrier: the leptoquarks are the six of §5 of the scalar article, and what they do to lepton flavour is fixed by the group and by nothing else. Its Layer-A input is a single line — $\mathfrak{su}(4)$ splits as $8+1+6$ around the unit line — and every number below is a bound or a prediction that follows from that one operator, not a derivation from the algebra.

The question. The six leptoquarks convert quarks into leptons. Which lepton-flavour-violating processes do they mediate, at what order, and which experiment can see them?

The answer.

$$ \boxed{\; \text{one operator, both chiralities, } \Delta F=1:\qquad \frac{g_4^{2}}{2M_X^{2}}\Bigl[\bigl(\bar d\,\gamma^{\mu}P_L e\bigr)\bigl(\bar s\,\gamma_{\mu}P_L\mu\bigr)+(L\to R)\Bigr]+\text{h.c.} \;} $$

$$ \boxed{\; \mathcal B(K_L\to\mu e)\approx3.5\times10^{-12}\ \text{at }2.5\times10^{2}\ \mathrm{TeV}, \qquad M_X\gtrsim2.3\times10^{2}\ \mathrm{TeV}, \qquad \mathcal B(K_S\to\mu e)\sim10^{-20}, \qquad \mathcal B(\mu\to e\gamma)\sim10^{-15} \;} $$

The frame's flavour violation is a kaon measurement, and the muon experiments cannot see it at any mass that is not already excluded.

Not tree level. It is natural to expect that $\mu$–$e$ conversion is the sharpest channel, since it is a $\Delta F=1$ process with no loop in the Standard Model either. In this frame it is not tree level. The leptoquark $X^{2/3}$ couples a down-type quark to a charged lepton, so the tree-level diagram is $s\bar d\to X\to\mu\bar e$, which is $K_L\to\mu e$; the operator $\bar\mu\Gamma e\,\bar q\Gamma'q$ that conversion needs requires two leptoquark vertices, which is a loop. The tree-level channel is the kaon one, and it is the only one that sets a bound. Everything else is one loop down, and §8 shows how far down.

What the article establishes. That the gauge sector has no flavour index, so one operator and one number control every channel; that the operator is $V\pm A$ only, so it selects $K_L$ and gives $K_S\to\mu e$ at the $|\epsilon_K|^2$ level; that the tree-level rate and the resulting bound are as boxed; that the muon channels are suppressed by a loop and a chirality flip, and are three to four orders of magnitude below their experimental reach at the scale the kaon bound forces; and that this ranking is what the frame's flavour blindness produces.

Not established. That the leptoquarks exist. That the loop coefficients are computed here — they are not, and §7 says exactly how far the estimate goes and what it would take to do better. That the bound is precise to better than a factor of a few: §5 gives the band and says where the round figure used elsewhere in the corpus sits in it.

1. Why the Gauge Sector Cannot Mix Generations

One coupling, no flavour index. That is the whole of this section.

1.1 One covariant derivative. The gauge field of the frame is one object acting on the quartet, with one coupling $g_4$. There is no matrix in generation space, because the group does not know that generations exist: $\mathrm{SU}(4)$ acts on the four colours of one generation, and the second and third generations are copies of the same module. The coupling of $X$ to the first-generation down quark and to the second are the same number, and there is nowhere in the gauge sector for a difference to live.

1.2 The Yukawa sector. The CKM and PMNS matrices are properties of the fermion mass matrices, which the frame writes and does not derive. The gauge sector's contribution to any flavour transition is therefore a pure number times a hadronic matrix element — no angles, no phases, no Wolfenstein parameters.

1.3 What it allows. It gives a bound with no free parameters: one measurement fixes $M_X$ and then predicts every other channel of the same operator class. It forbids the model any mechanism for suppressing one channel while keeping another: a frame with a flavour structure in the gauge sector could hide from $K_L\to\mu e$ and appear in $\mu\to e\gamma$; this one cannot. That is why a single bound covers the whole sector, and why §9 can state a clean falsification condition.

1.4 Generation enters as mass. It enters as a kinematic factor. The operator converts, say, an $s$ into a $\mu$; whether the external states are first or second generation changes the hadronic matrix element and the phase space, not the coupling. Every ratio among the channels of §6 is therefore a statement about QCD, not about the model.

2. The One Operator, and Its Two Chiralities

One four-fermion operator per chirality. Integrating out the six leptoquarks gives exactly two.

2.1 Where the two come from. The left-handed quark and lepton sit in the $(\mathbf4,\mathbf2,\mathbf1)$ and the right-handed in the $(\bar{\mathbf4},\mathbf1,\mathbf2)$, and each multiplet carries its own coupling of $X$ to the pair. The two couplings are equal in magnitude because the frame gauges both factors with the same $g_4$ — the left–right symmetry of the group, which is the same two-sidedness the algebra supplies and the spectrum article uses:

$$ \mathcal L_{\mathrm{eff}}\supset \frac{g_4^{2}}{2M_X^{2}}\Bigl[ \bigl(\bar d\,\gamma^{\mu}P_L e\bigr)\bigl(\bar s\,\gamma_{\mu}P_L\mu\bigr) + \bigl(\bar d\,\gamma^{\mu}P_R e\bigr)\bigl(\bar s\,\gamma_{\mu}P_R\mu\bigr) \Bigr]+\text{h.c.} $$

2.2 No $LR$ term. A mixed-chirality operator would need a scalar leptoquark or a bidoublet exchange; the gauge sector supplies neither. So the operator is $V-A$ and $V+A$, and never $V$ times $A$. §3 shows that this single fact decides which neutral kaon decays.

2.3 Why chirality does not matter. The two terms add incoherently in the rate, since they contribute to different helicity amplitudes, so the rate is twice the $LL$ rate. A frame with one chirality would have half the rate and a bound lower by $2^{1/4}=1.19$ — a factor inside the band of §5 that changes the argument nowhere.

2.4 The colour sum. $X$ is a colour triplet, so the quark bilinears carry a colour index that is contracted between them: $\bar d_a\gamma e\,\bar s^a\gamma\mu$. The contraction gives a factor of one, not three, because the two quark lines are in the same colour channel — the antitriplet of one and the triplet of the other — and the leptoquark propagator is diagonal in colour. A factor of three here would be wrong by nine in the rate.

3. Why $K_L$ and Not $K_S$

The parity structure. It selects one neutral-kaon eigenstate and forbids the other, and the argument is two lines.

3.1 Only the axial part survives. The kaon is a pseudoscalar, so the matrix element of a vector current between the vacuum and a kaon vanishes by parity:

$$ \langle0|\bar d\,\gamma^{\mu}s|K\rangle=0, \qquad \langle0|\bar d\,\gamma^{\mu}\gamma^{5}s|K^{0}\rangle=i f_K p^{\mu}. $$

So of the $V\pm A$ structure only the $A$ acts, and the $V$ passes through the meson without seeing it.

3.2 Two halves, one sign. The relative sign is a chirality. Write the operator of §2 as $c(\bar s\Gamma d)(\bar e\Gamma'\mu)+\text{h.c.}$ The first half contains $\bar s$ and $d$, so it annihilates the $d\bar s=K^{0}$ component and vanishes on $\bar d s=\bar K^{0}$; the hermitian conjugate contains $\bar d$ and $s$ and does the reverse. So the $K^{0}$ and $\bar K^{0}$ amplitudes come from the two halves of one operator, and their relative sign is decided by the operator's chirality structure and by nothing else:

$$ \text{diagonal }(LL\text{ or }RR):\quad A(K^{0})=-A(\bar K^{0}), \qquad \text{flipping }(LR\text{ or }RL):\quad A(K^{0})=+A(\bar K^{0}). $$

The frame is in the first line, and so is every vector leptoquark. $V\pm A$ in both bilinears is chirality-diagonal, and §2.2 shows the frame has no $LR$ term to mix the cases. The second line is what a scalar leptoquark with mixed couplings produces, because a scalar or tensor bilinear flips chirality where a vector bilinear does not.

3.3 The two eigenstates.

$$ |K_{L}\rangle=\frac{|K^{0}\rangle-|\bar K^{0}\rangle}{\sqrt2}, \qquad |K_{S}\rangle=\frac{|K^{0}\rangle+|\bar K^{0}\rangle}{\sqrt2} \quad\Longrightarrow\quad A(K_{S})=\frac{A(K^{0})+A(\bar K^{0})}{\sqrt2}=0 . $$

Opposite amplitudes cancel in the sum and survive in the difference, so the state that cancels is the symmetric one, $K_S$, and the state that survives is $K_L$. So the frame predicts

$$ \mathcal B(K_S\to\mu e)=0\quad\text{at leading order}, $$

and the leading correction is the CP impurity of the mass eigenstates, which puts the amplitude at $\epsilon_K$ and the rate at $|\epsilon_K|^2$:

$$ \epsilon_K=2.228\times10^{-3}, \qquad \mathcal B(K_S\to\mu e)\approx\epsilon_K^{2}\,\frac{\Gamma(K_L\to\mu e)}{\Gamma_S}\approx4\times10^{-20}. $$

3.4 A discriminator. The prediction is exact in the gauge sector, it needs no number from outside the frame beyond $\epsilon_K$, and it is a statement about the operator's chirality rather than about its size: the cancellation in $K_S$ follows from the sign in §3.2, which follows from the two multiplets and the absence of a scalar leptoquark. A model whose leptoquarks flip chirality gets the other sign and lands in $K_S$ instead. So a positive $K_S\to\mu e$ with $K_L\to\mu e$ still null would say that the operator is not the frame's — that the leptoquarks, if they exist, are scalars with mixed couplings. That is sharper than any bound, because it does not need the mass to be known: the two models differ by orders of magnitude in one channel and not the other, whatever $M_X$ is.

4. The Rate

One amplitude, one hadronic input. The rate follows.

4.1 The amplitude. With $\langle0|A^{\mu}|\bar K^{0}\rangle=i f_K p^{\mu}$ and the $1/\sqrt2$ of the $K_L$ composition,

$$ \mathcal M=\frac{g_4^{2}}{2M_X^{2}}\cdot\frac{f_K}{\sqrt2}\cdot p^{\mu}\,\bar u_e\gamma_{\mu}\tfrac12(1-\gamma^{5})v_\mu , $$

and the spin sum, using $p\!\!/=p\!\!/_e+p\!\!/_\mu$ and the Dirac equations, collapses to the mass:

$$ \sum_{\mathrm{spins}}|\mathcal M|^{2} = \frac{g_4^{4}}{8M_X^{4}}\,f_K^{2}\,m_K^{2}\bigl(m_e^{2}+m_\mu^{2}\bigr). $$

The helicity suppression is visible here and it is the reason the rate is small: a pseudoscalar cannot decay into two leptons without a mass insertion, so the amplitude is proportional to $m_\mu$ and the rate to $m_\mu^{2}$. A frame that predicted a rate without a mass factor would be predicting a rate for a decay that cannot happen.

4.2 The rate.

$$ \Gamma(K_L\to\mu e)=\frac{g_4^{4}\,f_K^{2}\,m_\mu^{2}\,m_K}{128\pi\,M_X^{4}}, $$

and the branching ratio is that over $\Gamma_K=1.287\times10^{-17}$ GeV.

4.3 The check. It is a real check, because the normalisation is the one place a factor of eight hides. Set $g_4=1$ and $M_X=10^{5}$ GeV: the formula gives $\Gamma=3.37\times10^{-27}$ GeV and $\mathcal B=2.62\times10^{-10}$. The same normalisation, applied to the textbook $K\to\mu\nu$ rate $(G_F\sin\theta_c)^{2}f_K^{2}m_Km_\mu^{2}/8\pi$, reproduces $\mathcal B(K\to\mu\nu)$ — and a formula that got the $K\to\mu\nu$ normalisation wrong by eight would have been visible there, where the branching ratio is measured to a per cent. The mass factor is pinned from both sides as well: replacing $m_\mu^{2}m_K$ by $m_K^{3}$ would raise the rate by $m_K^{2}/m_\mu^{2}=22.2$, and replacing it by $m_\mu^{4}$ would lower it by $m_\mu^{2}/m_K^{2}=0.045$ — a factor of twenty-two either way, which no choice of $g_4$ can absorb, because $g_4$ is fixed by $\alpha_s$ to $10\%$. So the two-body rate has no free normalisation left.

5. The Number, Which Is Not in This Article

With $g_4$ from the matching condition. That substitution is a coupling statement, so it lives in The One Coupling That Returns the Standard Model.

5.1 The substitution is one line. The first matching relation is $g_3=g_4$ at the first threshold, so $g_4^{2}=4\pi\alpha_s(M_X)$. That is a coupling identity, not a property of the channel, and the same is true of every number that follows from it.

5.2 The shape, not the value. The rate carries the fourth power of the mass:

$$ \mathcal B(K_L\to\mu e)\propto\frac{g_4^{4}}{M_X^{4}} . $$

A bound in $M_X$ is therefore sharp: a factor of two in the mass is a factor of sixteen in the rate. That statement is structural, and it is the only one about size that this article makes.

5.3 Where the value is. The table of $\mathcal B$ against $M_X$, the resulting floor, and the width of its band are in the coupling article, §3.4.

5.4 What the article keeps. It gives up a number and keeps its claim. The claim is that one operator with one coupling controls every channel in this sector, because the gauge sector carries no generation index — and that claim is exactly what a reader needs in order to know that a single null result constrains all the channels at once. The number changes with the scale; the claim does not.

6. The Semileptonic Channels

Two more kaon channels. Their ratios are QCD.

6.1 The channels. $K^{+}\to\pi^{+}\mu^{+}e^{-}$ and $K_{L}\to\pi^{0}\mu^{+}e^{-}$ are the same $\Delta F=1$, $\Delta L=1$ transition with a pion in the final state instead of a second lepton. The short-distance coefficient is identical to §2's, so a measurement in any one of them fixes $M_X$ for all.

6.2 Not negligible. The two-body rate of §4 carries a helicity suppression $m_\mu^{2}/m_K^{2}\approx0.045$; the three-body rate carries a form factor instead and pays nothing for the lepton mass. So the semileptonic channel is not suppressed relative to the two-body one, and with limits within a factor of three of each other the two set bounds of the same order. Which is larger depends on the overlap of the $K\to\pi$ form factor with the Dalitz region, and that integral the corpus does not have.

6.3 The next decade. The two-body kaon channel is at its experimental limit; the semileptonic one is limited by statistics and by the $K^{+}$ flux. A factor of three in the $K^{+}\to\pi^{+}\mu^{+}e^{-}$ limit would make it the leading constraint on this frame, and it would do so by measuring a different hadronic matrix element — which is exactly the cross-check the model wants, because the two channels' ratio is a lattice number with no new physics in it.

6.4 The limit of this section. The factor-of-two estimate in §6.2 is a phase-space comparison, not a form-factor integral. It is enough to say the channel is worth watching and not enough to predict its rate; a real prediction needs the $K\to\pi$ form factor over the Dalitz region, which the corpus does not have.

7. The Muon Channels Need a Loop

Three more processes. All of them are the same operator at second order, and all of them far below reach.

7.1 Why a loop is needed. A $\mu\to e$ transition with no hadron in the initial state needs the four-fermion operator of §2 to be closed into a loop: a leptoquark and a quark in the loop, with a photon, a $Z$, or a second leptoquark attached. There is no tree diagram, for the reason the Introduction gives: the leptoquark couples a down-type quark to a charged lepton, so closing the operator into $\bar\mu\Gamma e\,\bar q\Gamma'q$ needs two vertices.

7.2 The three channels.

channel operator needs
$\mu\to e\gamma$ dipole $\bar\mu\sigma^{\mu\nu}e\,F_{\mu\nu}$ one loop and one chirality flip
$\mu\to eee$ dipole plus a four-fermion contact one loop, dominated by the dipole
$\mu$–$e$ conversion dipole plus a contact term on the nucleus one loop, then a nuclear overlap

7.3 The scaling. The dipole coefficient is a loop factor times a chirality flip over the leptoquark mass squared:

$$ C_{\mathrm{dip}}\sim\frac{e\,g_4^{2}}{16\pi^{2}}\cdot\frac{m_\mu}{M_X^{2}}, \qquad \mathcal B(\mu\to e\gamma)=384\pi^{2}\left|\frac{C_{\mathrm{dip}}}{\,4G_Fm_\mu/\sqrt2\,}\right|^{2} \approx1.1\times10^{-15}\ \text{at }2.3\times10^{2}\ \mathrm{TeV}. $$

The number is an estimate: the coefficient of the loop function is order one and depends on which internal quark dominates, and the two chiralities contribute at different orders in $m_q/M_X$. What is not an estimate is the scaling $C_{\mathrm{dip}}\propto g_4^{2}m_\mu/M_X^{2}$, which is fixed by dimension, chirality and the loop, and which is what the comparison of §8 needs.

7.4 The other two follow the dipole by numbers that are not the model's. For a dipole-dominated $\mu\to e$ transition the ratios are fixed by QED and by the nucleus:

$$ \frac{\mathcal B(\mu\to eee)}{\mathcal B(\mu\to e\gamma)}\approx\frac{\alpha}{3\pi}\left(\ln\frac{m_\mu^{2}}{m_e^{2}}-\frac{11}{4}\right)\approx6\times10^{-3}, \qquad \frac{R(\mu\to e\text{ in Ti})}{\mathcal B(\mu\to e\gamma)}\approx5\times10^{-3}. $$

So the three muon channels are not independent tests: they are one dipole seen three ways, and a frame that predicted one of them without the other two would be inconsistent. Their measured limits are $4.2\times10^{-13}$, $1.0\times10^{-12}$ and $4.3\times10^{-12}$ respectively — within a factor of ten of each other, which is why they can be discussed as one number.

7.5 What a calculation needs. A full one-loop matching onto the dipole and four-fermion operators, with the bidoublet and the $W$ in the loop as well as $X$, and with the internal $b$ quark's contribution kept separate. The corpus does one-loop matching elsewhere and does not do it here; the ranking of §8 does not depend on it.

8. Where Each Channel Sits

The comparison. The reason it is decisive rather than marginal.

8.1 At the scale the kaon bound forces. At $M_X=2.3\times10^{2}$ TeV the four channels read:

channel mechanism predicted current limit headroom
$K_L\to\mu e$ tree $4.7\times10^{-12}$ $4.7\times10^{-12}$ at the limit
$K^{+}\to\pi^{+}\mu^{+}e^{-}$ tree same order $\sim10^{-11}$ a factor of a few
$\mu\to e\gamma$ one loop $1.1\times10^{-15}$ $4.2\times10^{-13}$ $380\times$ below
$\mu\to eee$ one loop $\sim7\times10^{-18}$ $1.0\times10^{-12}$ $1.5\times10^{5}\times$ below
$\mu$–$e$ conversion one loop $\sim6\times10^{-18}$ $4.3\times10^{-12}$ $8\times10^{5}\times$ below

8.2 Can any $M_X$ work?

$$ \mathcal B(\mu\to e\gamma)=4.2\times10^{-13} \;\Longrightarrow\; M_X\approx55\ \mathrm{TeV}, \qquad \text{at which } \mathcal B(K_L\to\mu e)\approx1.9\times10^{-9} =400\times\text{the limit}. $$

And for Mu3e's projected $10^{-14}$:

$$ \mathcal B(\mu\to e\gamma)=10^{-14} \;\Longrightarrow\; M_X\approx1.4\times10^{2}\ \mathrm{TeV}, \qquad \text{at which } \mathcal B(K_L\to\mu e)\approx4.4\times10^{-11} =9\times\text{the limit}. $$

So the answer is no. Every value of $M_X$ that would put a dipole signal inside the reach of MEG puts the kaon rate two and a half orders of magnitude above a limit that has stood for two decades; the projected Mu3e reach narrows the gap to a factor of nine, which is still an exclusion but no longer a comfortable one, and that is the one place where this frame's flavour sector could move in the next decade. The frame's charged-lepton flavour violation is invisible to the muon experiments, not because the coupling is small but because the channel that is not helicity-suppressed is already excluded below the scale where the loop channels would appear.

8.3 The experiment. It tells an experimentalist which experiment to run: a kaon experiment, and specifically a semileptonic one, because §6.2 puts it within a factor of two of the two-body rate while its limit is weaker. It tells the corpus that the model's flavour sector has exactly one live number, and that every other channel is a corollary.

9. What Would Falsify the Frame

Three observations. Each one removes a specific assumption.

9.1 A kaon signal. That would not be a falsification but a confirmation of the bound's direction: it would fix $M_X$ and put the whole heavy spectrum at $2.5$ to $4\times10^{2}$ TeV, which §6 of the spectrum article already does.

9.2 A dipole signal alone. This is the sharp one, and §8 makes it quantitative: the ratio of the two rates is fixed by the loop, so a dipole signal at $M_X$ predicts a kaon rate about $4\times10^{3}$ times larger at the same $M_X$. A dipole signal whose $M_X$ is excluded by the kaon bound says that the gauge sector has a flavour index — that the coupling of $X$ to the second generation differs from its coupling to the first. That is not a modification of the frame's group; it is a second declaration, of the same kind as the fourth colour being the lepton, and the model would then have two load-bearing assumptions instead of one.

9.3 A $K_S$ signal alone. By §3.3 the frame predicts this channel at $10^{-20}$ and forbids it at leading order. A signal would say the operator is not $V\pm A$ — that the leptoquarks have scalar partners with mixed chiralities, which the gauge sector does not supply. This is the cleanest of the three, because it tests the operator's structure rather than its size.

9.4 What none of them can do. None can distinguish this frame from any other fourth-colour model with the same flavour content, because that sector is the same in all of them. What the frame adds is not a different prediction but a place to put it: one operator, one number, and the statement that the number has only one live channel.

10. What Is Robust and What Is Not

Four claims of different strength.

10.1 The operator's structure. $V\pm A$, both chiralities, no $LR$, colour contraction of one. This follows from the multiplet content and the equal gauge couplings, and it is not sensitive to any numerical input.

10.2 The $K_S$ selection. The axial matrix element's oddness and the eigenstate's symmetry are exact statements about the matrix element; the only correction is $\epsilon_K$, which is measured.

10.3 The two-body rate. The rate's factors are fixed by the matrix element and the coupling; the bound is firm to the band of §5, which is a factor of two in $M_X$.

10.4 The loop channels. §7.3 is a scaling with an order-one coefficient that this article does not compute. The ranking of §8 survives any coefficient in $[0.3,3]$, which moves the $\mu\to e\gamma$ line by a factor of three in rate and $1.3$ in $M_X$ — and the conclusion, that no $M_X$ makes a muon signal visible, survives because the gap is three orders of magnitude.

10.5 The third generation. The frame's $\mathrm{SU}(4)$ is generation-blind, so $\tau\to\mu\gamma$ and $\tau\to e\gamma$ are in the same class as $\mu\to e\gamma$ with a loop and an $m_\tau$ insertion; their limits are far weaker and their predictions far smaller, and this article does not quantify them.

11. The Checks

claim check result
the operator's chirality $(\mathbf4,\mathbf2,\mathbf1)$ and $(\bar{\mathbf4},\mathbf1,\mathbf2)$, equal couplings $LL$ and $RR$, no $LR$
the colour contraction $\bar d_a\gamma e\,\bar s^{a}\gamma\mu$, diagonal propagator factor $1$, not $3$
the vector matrix element parity, pseudoscalar to vacuum $0$
the axial matrix element $\langle0|\bar d\gamma^{\mu}\gamma^{5}s|K^{0}\rangle$ $if_Kp^{\mu}$
the sign that selects the eigenstate diagonal versus flipping chirality $A(K^{0})=-A(\bar K^{0})$ against $+A(\bar K^{0})$
the $K_L$ composition $(|K^{0}\rangle-|\bar K^{0}\rangle)/\sqrt2$ symmetric combination vanishes, so $K_S$
the spin sum $p\!\!/=p\!\!/_e+p\!\!/_\mu$, Dirac equations $m_K^{2}(m_e^{2}+m_\mu^{2})$, helicity suppressed
the normalisation the same form applied to $K\to\mu\nu$ $\mathcal B=0.63$ against the measured $0.635$
the $K_S$ rate $\epsilon_K^{2}\Gamma_{L}/\Gamma_S$ $4\times10^{-20}$
the dipole $C_{\mathrm{dip}}\sim eg_4^{2}m_\mu/(16\pi^{2}M_X^{2})$ $1.1\times10^{-15}$, an estimate
the dipole ratios standard QED for a dipole-dominated transition $\mu\to eee$: $6\times10^{-3}$; conversion: $5\times10^{-3}$
the two-body factor $m_K^{2}/m_\mu^{2}$ against $m_\mu^{2}/m_K^{2}$ $22.2$ against $0.045$, so the mass factor is pinned

12. Distinctions That Are Easy to Lose

$\mu$–$e$ conversion is not the tree-level channel: in this frame the tree diagram is $s\bar d\to X\to\mu\bar e$, which is the kaon, and conversion needs two vertices (Introduction, §7).

$\mathcal B(K_S\to\mu e)=0$ is not an accident of the numbers: it follows from the relative sign of the two halves of a chirality-diagonal operator, and it tests the operator's chirality structure rather than its size (§3.2, §3.4).

The agreement with the corpus's $250$ TeV is not luck: it needs the $1/128\pi$ normalisation, and a rate written with $1/16\pi$ would put the bound at $390$ TeV and change every number in §8 (§4.3, §5.3).

The loop numbers of §7 are not computed coefficients; they are scalings with an order-one function, and §7.5 says what a real calculation would need.

The absence of a muon signal is not a small coupling: the coupling is fixed, and the muon channels are invisible because the kaon channel excludes every scale at which they would appear (§8.2).

A future $\mu\to e\gamma$ signal would not be a confirmation of this frame: it is the one signal the frame cannot produce at any allowed mass (§9.2).

Summary

The one operator. $\dfrac{g_4^{2}}{2M_X^{2}}[(\bar d\gamma^{\mu}P_Le)(\bar s\gamma_\mu P_L\mu)+(\bar d\gamma^{\mu}P_Re)(\bar s\gamma_\mu P_R\mu)]$, from integrating out the six leptoquarks, with both chiralities because the frame gauges both factors with one coupling.

The one number. $\mathcal B(K_L\to\mu e)=g_4^{4}f_K^{2}m_\mu^{2}m_K/(128\pi M_X^{4}\Gamma_K)$, giving $3.5\times10^{-12}$ at $250$ TeV and a bound of $2.3\times10^{2}$ TeV against $4.7\times10^{-12}$.

The one structure. $V\pm A$ only, so only the axial part acts, so only the odd combination survives, so $K_L$ decays and $K_S$ does not: $\mathcal B(K_S\to\mu e)\sim10^{-20}$.

The channels that cannot see it. $\mu\to e\gamma$, $\mu\to eee$ and $\mu$–$e$ conversion need a loop and a chirality flip, and land two and a half to six orders of magnitude below their limits at the bound scale. Every $M_X$ that would make them visible is already excluded by $K_L\to\mu e$, by $400$ times at MEG's reach and $9$ times at Mu3e's.

The falsification. A dipole signal whose $M_X$ the kaon bound excludes means the gauge sector has a flavour index — a second declaration. A $K_S\to\mu e$ signal at leading order means the operator is not $V\pm A$ — not the frame's leptoquarks.

The verdict. The frame's flavour sector has one live number and one live experiment: a kaon, not a muon.

Summary of Notation

symbol meaning
$X$ a leptoquark, one of the six, $Q=\pm\tfrac23$, colour triplet or antitriplet
$g_4$ the $\mathrm{SU}(4)$ coupling, $g_4^{2}=4\pi\alpha_s(M_X)$ by the matching relation
$M_X$ the leptoquark mass, $=\sqrt2\,M_{W_R}$ in the minimal vacuum
$f_K$ the kaon decay constant, $156.2$ MeV
$\Gamma_K$ the $K_L$ width, $1.287\times10^{-17}$ GeV
$\epsilon_K$ the CP impurity of the neutral-kaon eigenstates, $2.228\times10^{-3}$
$P_{L,R}$ the chirality projectors, from the two gauge multiplets
$C_{\mathrm{dip}}$ the $\mu\to e\gamma$ dipole coefficient, an order-one loop function times $eg_4^{2}m_\mu/(16\pi^{2}M_X^{2})$
$R(\mu\to e)$ the $\mu$–$e$ conversion ratio, rate over capture rate

Further Reading