Layer 2 — No unpaid exact labels — rendered package. Rendered from layer-2-no-unpaid-labels.md; frozen technical content unchanged by rendering.

Layer 2 — No unpaid exact labels

If a theory uses an exact Standard-Model label, it must either explain where it came from or pay for writing it down. There is no free label.

This is the anti-smuggling rule — the second bridge in the stack that turns "is the 13D shape simplest?" from a vibe into an audit. Layer 1 fixed how we measure simplicity: by description length, the number of bits in a theory's full generator, not by counting spacetime dimensions. Layer 2 enforces a single discipline on top of that metric: every exact label a candidate uses to reproduce the observed Standard Model has to appear, with its cost, somewhere in that generator. You can generate a label from declared structure, you can charge it as a primitive input, or you can leave it openly unexplained — but you cannot get it for free. This is the cleanest, most uncontroversial result in the whole program, and it cuts in every direction, including against us.

The question this answers

When two theories both "reproduce the Standard Model," which one actually paid for that reproduction? A 4D effective field theory and a 13-dimensional geometry can both end up with the gauge algebra $su(3)\oplus su(2)\oplus u(1)$, three chiral generations, the hypercharge table, the masses and the mixings. If you only look at the output, they tie. Layer 2 asks the question that breaks the tie: where in each theory's description does each of those exact labels enter, and what does it cost there? A label that was simply typed in by hand is an input the theory must be charged for. A label that fell out of declared structure is a genuine prediction. The accounting is what separates the two, and the accounting is mandatory.

The theorem and its proof

Theorem L2 (no free exact label). Given the Layer-1 full-generator metric, every exact label a candidate uses to match the target must be one of:

  1. generated by the candidate's declared structure (geometry, bundle, quotient, index theorem, selection rule); or
  2. charged as a primitive input (written into the generator and counted in its bit budget); or
  3. left open (declared as not-yet-explained).

There is no fourth option. There is no free exact label.

The proof is short and it is the kind of argument that does not depend on which theory you favor:

Formally, write a theory's full generator as $B$ and its description-length cost as $I(B)$ in bits. Layer 1 established that $I(B)$ is the right currency. Layer 2 says: if the target is the observed spectrum and constant set $T$, then every exact label appearing in $T$ traces to a term in $I(B)$ — a generated term (cheap, because it rides on structure already paid for) or an injected term (a fitted real or a hand-set integer, each of which costs bits). The theorem does not say which is the case for any given theory. It says the books must balance, and the cost of an unexplained label is the cost of writing it down.

This is why Layer 2 is, on its own, neutral. It does not prove geometry is better than a 4D postulate. It forces both to declare their charges honestly. The verdict between them is a Layer-1 comparison of the resulting bit budgets, not a Layer-2 claim.

Discipline line: selection $\ne$ derivation. Choosing a structure that happens to carry the right labels is not the same as deriving why those labels are forced. Layer 2 is what makes the difference visible in the bit count.

Corollary 1 — a 4D effective theory is allowed, but it pays

A 4D effective field theory is a perfectly legitimate competitor. It is cheaper on dimension — four beats thirteen, and under a dimension-first ruler it would win outright. Layer 2 does not forbid it. It simply itemizes the bill.

The 4D theory may postulate the gauge algebra $su(3)\oplus su(2)\oplus u(1)$, the representation table, the family count, the hypercharges $Y$, the fermion masses, and the mixing angles. Every one of those is permitted — but each is a primitive input unless a mechanism inside the theory generates it. In description-length terms the honest 4D ledger runs to roughly $I(B_{\rm EFT})\approx 25\,b$ bits of injected structure — about twenty-five independently-measured reals, each costing $b=\log_2(1/\Delta_0)$ bits to pin to the operational resolution (the Layer-1 currency) — because the effective theory writes the whole spectrum-and-constant ledger in directly: it leaves the constants as free parameters and reads them from experiment. Its honest status is therefore precise and not pejorative:

Lower metric dimension; higher primitive-label cost. A complete, legitimate description with no geometric origin for the labels it postulates.

That is the correct thing to say about a 4D EFT — not "the EFT loses, full stop," which would be a Layer-3 cross-grammar overclaim we explicitly do not make.

Corollary 2 — the 13D shape pays too, in a different currency

The geometric program is not exempt. It pays through the cost of the geometry, the bundle and quotient data, the measured anchors, and the rulebook that maps structure to observables — and then it attempts to generate the labels that the 4D theory simply posited. The honest charged cost of the 13D branch is on the order of $I(B_{13})\approx(4+9\text{–}10)\,b\approx 13\,b\text{–}14\,b$ bits: roughly four measured anchors $\{M_{\rm Pl},\alpha_i,y_t,|V_{us}|\}$ plus nine or ten further injected reals (the fitted normalizations, the mixing-phase data, the Higgs-sector angle, the threshold triple, and so on), each real again charged at $b$ bits. This bit ledger is the strict whole-construction count, and it answers its own question — "what does the whole machinery run on?" It sits beside, and does not replace, the calibration count published on the overview page: from the two flavor anchors $\{y_t, |V_{us}|\}$ alone the geometry returns 19+ flavor observables, with 6–8 more following from $\{M_{\rm Pl}, \alpha_i\}$ — a handful in, 20+ out. Two counts, honestly separated: the calibration count is where the predictions live; the bit ledger above is the audit that keeps it honest.

What the geometry buys for those bits is generation rather than postulation: given the frozen shape and the observed matter content, it reads off the gauge algebra, three chiral generations, and charge quantization from the structure, instead of typing them in. Whether that trade — more dimensions and a richer generator, in exchange for generating labels the 4D theory leaves free — comes out ahead is exactly the Layer-1 comparison of $I(B_{13})$ against $I(B_{\rm EFT})$. The current ledger has the geometry ahead on the full target ($\sim 13\,b\text{–}14\,b$ bits vs $\sim 25\,b$ — close to half the bill), because the EFT injects the entire constant set directly while the geometry generates much of it. This is a conditional comparison — conditional on the Layer-1 metric, not a proof of absolute minimality — and it lives at Layer 1, not here.

Discipline line: frozen/reproducible $\ne$ proven-unique. The 13D generator is specified, hash-frozen, and reproducible blind. That makes its bit ledger auditable. It does not make it the shortest possible generator — that is the open question Layers 3 and 4 attack.

Corollary 3 — anomaly cancellation is a filter, not a selector

This is the corollary with the most teeth, because it is exactly the place where a well-known true fact is routinely inflated into a false claim. The true fact: the Standard Model spectrum is anomaly-free. The false claim that gets built on top of it: "anomaly cancellation selects the Standard Model." Layer 2 says the second statement does not follow from the first, and the distinction is the difference between a filter and a selector.

Anomaly cancellation is a constraint. Write the anomaly functional as $A$. The observed spectrum $E_{\rm SM}$ satisfies $A(E_{\rm SM})=0$ — it lies in the kernel $\ker A$. That is a genuine, non-trivial consistency check that the spectrum passes. But passing a filter is not being selected by it. To upgrade "filter" to "selector" you would have to prove that the spectrum is the only admissible thing in the kernel:

$$\ker A \cap \mathcal C_{\rm admissible} = \{E_{\rm SM}\},$$

a singleton theorem. That theorem is not proven, and there is strong evidence it is false: any vector-like pair $R\oplus\bar R$ cancels automatically, so the kernel contains infinitely many anomaly-free spectra. The constraint forbids a great deal, but it does not single out the Standard Model. So:

$$E_{\rm frozen}\in\ker A,\qquad\text{but}\qquad \ker A\ne\{E_{\rm SM}\}.$$

The honest disposition is that the claim "anomaly cancellation selects the SM" is dissolved — it was never a derivation; it was a filter mistaken for a determiner. The observed matter content $E$ therefore remains a primitive of the program, charged as input, not generated by the anomaly condition.

It is worth seeing how specific the cancellation actually is, because "specific" is real content even though "selecting" is not. On the single-generation spectrum, the six independent local anomaly ledgers all vanish exactly by rational arithmetic:

$$ \begin{aligned} [U(1)_Y]^3:\ \textstyle\sum Y^3 &= 0 &\quad ([\text{grav}]^2U(1)_Y:\ \textstyle\sum Y = 0)\\ [SU(2)]^2 U(1)_Y:\ 3\cdot\tfrac16 - \tfrac12 &= 0 &\quad [SU(3)]^2 U(1)_Y:\ 2\cdot\tfrac16 - \tfrac23 + \tfrac13 = 0\\ [SU(3)]^3 &= 0\ (\text{vector-like color}) &\quad [SU(2)]\ \text{(Witten, mod }2):\ \#\text{doublets}=4,\ \text{even}. \end{aligned} $$

And the cancellation is not an automatic consequence of a trivial sum — the diagnostic $\sum Y^2 = 10/3 \ne 0$ shows the charges are genuinely non-trivial and the zeros are real cancellations among real terms. The hypercharge lattice itself, $Y\in\tfrac16\mathbb Z$, follows from center-locking on the observed multiplets (the spacing $\tfrac16 = \tfrac{1}{\mathrm{lcm}(3,2)}$ comes from the color and weak center denominators), and the electric charge relation $Q=T_3+Y$ then comes out with the neutrino neutral and the down-quark at $-\tfrac13$, with no per-multiplet fitting. All of that is genuine derived-given-E content: given the spectrum $E$, the consistency is forced and tight. None of it derives $E$.

Discipline line: given-E $\ne$ derivation-of-E. Every result in this section is conditional on the observed spectrum being fed in. The framework shows that $E$ is beautifully self-consistent. It does not show that $E$ had to be this and nothing else — that is the given-E wall, attacked at Layer 4.

Status — what is reached, what is the open target

Reached. Layer 2, given the Layer-1 metric, is a clean accounting theorem. It is proven and it is not seriously contestable: a label used to match the target must be generated, charged, or open, and a non-unique filter does not generate. The anomaly corollary is settled in the honest direction — cancellation is a filter, the singleton theorem is unproven (and likely false), and "anomaly cancellation selects the SM" is dissolved. The six-ledger cancellation is verified exact, with $\sum Y^2 = 10/3$ confirming it is specific rather than trivial.

What Layer 2 deliberately does not do. It does not force the geometry. It is not an argument that 13 dimensions are better than 4 — that comparison is a Layer-1 bit count, and across different search grammars it needs the Layer-3 bridge. Layer 2's job is narrower and stronger for being narrow: it forces honest charging, so that no theory in the comparison can smuggle a label in for free. After Layer 2, the matter content $E$ stays primitive on every side of the table — including ours.

The named open target. The one theorem that would change this picture is the singleton / selector theorem — a proof that

$$\ker A \cap \mathcal C_{\rm admissible} = \{E_{\rm SM}\}$$

under some declared admissibility class $\mathcal C_{\rm admissible}$ that does not build the observed answer into its own definition. Proving it would upgrade the anomaly condition from filter to selector and would partly remove "given-E." It is not proven. The current evidence (the vector-like loophole, the infinitude of the kernel) points against it in full generality. So the honest endpoint here is sharp and unapologetic: anomaly admissibility is a genuine, specific filter that the Standard Model passes; it is not a selector; $E$ is charged as input until a singleton theorem says otherwise.

That is the methodology working as intended. We say exactly which game the labels win — the consistency game, decisively — and exactly which game they do not yet win — the selection game. The honesty is not a hedge; it is the audit that makes every number on the page trustworthy.

Cross-links