The price of a read
What deciding a condition costs in gates and in word
operations, and what the ring's own valuation does to evidence.
A rung of the tower is Z/N with N the product
of the first k primes, one channel per prime, and the
archimedean place deleted (Walls), so nothing
here is priced by size. A measurement program is a word — a
formula over the leaves, the input variables, in MUL/INV/NOT, where
INV is the meadow inverse (channel-wise inverse on the support, 0
elsewhere, so division is total) and dropping NOT leaves the
monomial alphabet, the aibj with
exponents in Z. Words are read by the gates
gatem(w) = [wm = 1],
which decide order divisibility channel by channel
(the quantifier ladder). What
a program is asked to decide is a shadow: an exact Boolean
condition on the ring. Three budgets can pay for one — the alphabet
the word is built from, the number of word operations, the number of
gates — and which of them binds is set by the alphabet level
(the measurement table). At
the monomial level the gate count is exact and provably so; open the
alphabet to the whole meadow and deciding any residue drops to a single
gate, at a word-op price with no ceiling.
Gate count and word ops
The orbit-cost
law theorem
At the monomial level, gate shadows are exactly the character
kernels of the discrete-log torus
(Z/n)L — n = p − 1
the order of the channel's unit group, one coordinate per
leaf — and a stable torsion orbit — the
generators of the cyclic group C = ⟨(a, b, …)⟩
of order d — costs exactly m(C) +
ω(d) monomial gates (ω the number of distinct
prime factors), m(C) the minimal number of monomial
equations cutting C exactly = the rank of the ambient
modulo C (at two leaves: 1 if d = n, else
2) — coset gates plus order read. Upper: the
canonical program — m(C) gates cutting C,
then for each prime q | d one gate on a frame word — a
word of full order d on the orbit.
Lower: forced separators — per prime q | d, its own
gate — plus covering rigidity (the trade-exclusion): any coset the
membership gates fail to kill must be covered by separator
slices — a separator gate's kernel cut down to that coset —
and a slice is pinned — a coset of order coprime to d
threads every slice through one point — so every rank the
membership gates drop comes
back, gate for gate, in separators — no covering system of
t membership gates and s separators, repeated or
composite moduli allowed, ever beats
t + s ≥ m(C) + ω(d).
The lower bound never uses the leaf structure, so the law holds in
any finite abelian ambient group. At one leaf the law is the
order-cost law (rule, proved field-free both directions):
[ord x = d] costs exactly ω(d) + 1,
dropping to ω(n) at d = n —
primitivity reads one gate cheaper exactly at the top — and
filters — [x = 1] and the gates
gatem(ab⁻¹) of
the equality
grading — cost exactly 1. On the orbits: the order-5 orbit of
the pentagon conic
reads at 3 = 2 + 1 gates against its cut cost of 5 collected
monomials — deciding a locus and cutting it are different
currencies, neither dominating; the Φ6 stable orbit of
the line a + b = 1 reads at 3 = 1 + 2 at
p = 7, where d = n, and 4 = 2 + 2 at
p = 13, so per-channel prices are not uniform across a
rung. Character kernels make a readout's atoms — the smallest sets
its gates can tell apart — obey Lagrange arithmetic: the
subgroup-cell bound (rule, proved), the ambient form the law
sharpens on cyclic targets.
Scope. Theorem proved both directions, any
finite abelian ambient. Censuses stand as verification: all
337,054,176 structured ≤ 4-gate candidates at the three-leaf
m(C) = 3 orbit, four fresh configurations sized so a
trade would fit under the law; the one-leaf law exhaustive over
lattice sweeps n = 6..360 plus boundary targets at
n = 2310.
verifiers:
explore_orbit_cost.py,
explore_trade_exclusion.py,
explore_gate_budget.py
The height
floor rule
The meadow level collapses gate count to 1 —
gate1 of a decision word — at a price moved wholly into
word ops: only cum(L) distinct words — the running total of
the formula-counting recurrence — exist at cost ≤ L, so the
dearest residue's decision-word cost grows without bound with
p. That cost is the constant's height
hp(c), the fewest word operations that
build c from one leaf holding any residue other than 0
and 1, and it carries the whole
price: [x = c] is decided by one gate and at most
hp(c) + 2 word ops. The deleted place
returns as the price of
omniscience: which residue you decide carries a height
tax that grows without bound.
The wall-breaker
word is optimal — hp(−1) = 6 exactly at
p = 5, 7, the cheapest nontrivial constant (observation;
heights are not monotone: h(4) < h(3) at
p = 7) — and one ring gate reads all k channels in
parallel, so every per-channel price above is already the
ring's.
Scope. The counting bound and the
one-gate-plus-hp(c)+2 decision are proved;
the optimality of the wall-breaker word is an observation at
p = 5, 7 and the non-monotonicity of heights one at
p = 7.
verifier:
explore_gate_budget.py
Measuring evidence
Every set S of channels has its own idempotent
eS — residue 1 on S, 0 off it — and those
2k elements are the layer on which the ring's logic
is exactly Boolean
(the graded logic). That layer
carries a valuation (observation): disjoint join is ring sum
(eSeT = 0 implies
eS∪T = eS +
eT), so additive strengths are exactly
channel-weight functionals, while the canonical counting valuation
μ(eS) = 1/∏(excluded p) is
multiplicative — μ(S∧T)μ(S∨T) =
μ(S)μ(T) — additive in the log domain:
the strength of a truth value is the log-mass of what it
excludes, content rather than chance. That valuation is
where statistics attaches.
One test, every
semiring rule
For product-form f the ring-total factorizes per channel
in any commutative semiring (the generalized distributive law +
CRT), and f equals the normalized product of its own
channel marginals iff f is product-form — a
semiring-indexed rank-1 test, exact in each of the three semirings
tested. On relations the verdicts are semiring-stable (a 0/1
product form can be taken with 0/1 factors — the lemma needs only
that the semiring has no zero divisors): equality and divisibility
pass, the order relation x ≤ y fails, at all three
points tested. The walls do not
depend on the temperature, the semiring the total is taken
in: Bayesian sum, tropical max, and Boolean feasibility agree on
what cannot factor.
Scope. Exhaustive at Z/30
(sum/max/Boolean) and Z/210 (Boolean + sum); the
classical family is the generalized distributive law
(Aji–McEliece) and valuation algebras (Shenoy–Shafer) — ours is
the tower-side composition.
verifier:
explore_evidence_calculus.py
Evidence is
one-directional rule
An e-variable is a nonnegative statistic of mean at most
1 under the null — the capital of a bet against it. The support
cylinder AS = {x :
x·eS = x} has exact probability
μ(eS), so
ES = 1[AS]/μ(eS)
is an exact e-variable, and the product of two is an e-variable
iff the two events are CRT-independent
(S ∪ T = all channels) — the multiplicative
valuation law is the bookkeeping. The classicality e-variable
reads the defect δ(x) = x² − x;
composing it with the collapse
x → xλ — λ the ring's
universal period, so the power map sends every unit to 1 — returns
∏p∈S p/2 for every
x (mean 105/8 at Z/210 with S = all
channels): the measurement fabricates
classicality, and the e-calculus is exactly the frame that rejects
it — bets precede measurement, e-processes on the ring adapted to
the pre-collapse filtration. Merging shows the same arrow: min and
harmonic merges are e-valid on all pairs, the arithmetic merge is
exactly tight (the Vovk–Wang canonical merge, exact on the ring),
and past it merges die — quadratic with an exact witness of mean
> 1.345, max-merge at worst mean exactly 2 − 1/N. The
max-tropical end fabricates evidence, the min-tropical end only
wastes it: the same one-directionality as
the pair's destructive
interference under OR — truth can be lost, never fabricated — now
enforced by a formalism. The e-calculus is not a third semiring
point but a second-order axis (bets over masses) anchored at the
summing end, whose log-shadow is the tropical valuation.
Scope. The product rule exhaustive at
Z/30 and Z/210 (all idempotent pairs;
Z/510510 all 128² by count formula, spot-enumerated); the
merge verdicts are the script's record. The e-value frame is
Shafer–Vovk's, the merges Vovk–Wang's — the ring composition is
the claim.
verifier:
explore_evidence_calculus.py
Degree quantifiers come
free with the ring, channel quantifiers cost the deleted place, and complete knowledge of a residue costs a
height that grows without bound — measurement without size is not
measurement without price, and the price list is exact.