INDEPENDENT ADVERSARIAL AUDIT — SELECTED HAAR REFILL LAW
25 October2026. Worker PinaColada. CPU only; no API writes/GPU/SYMBIOSE/children.

VERDICT
The proposed exact one-interface distribution, full covariance kernel and
all-boundary functional CLT are mathematically coherent under normalized Haar
measure on odd2adic integers. I independently derived them; the derivations
below did not infer the answer from the parent's formulas or numerical output.
These supply a more focused paper contribution than aggregating the previous
elementary consumer ledgers. The conditional measure and clock must remain
explicit. No theorem here bounds an individual positive-integer trajectory.

1. THE ACTUAL IID INPUT AND ITS NULL SET
For canonical blocktype T=(r,s), its word(1^(r-1),s+1) has binary costS=r+s.
The inverse affine block maps all odd2adic endpoints bijectively to one exact
odd source cylinder of relative Haar mass2^(-r-s). Thus successive types are
IID with p_(r,s)=2^(-r-s); r ands are independent geometric(1/2), support1,2,... .
This is standard parity coding/inducing, not new novelty. Each conditional
endpoint remains odd Haar and independent of all preceding types.

Remove the measure-zero set on which a block never ends, U has an infinite
valuation, or a reserve hits its exact rational pin. Countably many pins and
finite inverse branches give a countable null exceptional set. Do not assign
a fake finite valuation to0. In particular countable fixed positive integers
have Haar measure0, so almost-sure statements do not imply their genericity.

Let X_(i+1) be endpoint ofblocki and R_i=V_(T_i)(X_(i+1)). Conditional on the
entire type history through T_i, R_i is geometric(1/2): P(R_i=a)=2^-a,a>=1.
Hence E R_i=2,VarR_i=2. This does NOT say R_i is independent of future types.
It is strongly dependent on the upcoming same-type run and its first change.

2. INDEPENDENT PGF DERIVATION
At an interface oldtype(r,s),newtype(u,v), let t be the canonical pin separation.
Same type meansJ0. Otherwise J=u+v+R-t with a fresh marginal reserveR~Geom1/2
independent of the old/new types. Here fresh means its conditional endpoint law,
not independence of J across successive interfaces.

Write a=z/(2-z). The PGF of the base u+v-t, restricted to different types,
has four disjoint contributions:
 r<u :a^2/3;
 r>u :a/3;
 r=u,s<v :a/9;
 r=u,s>v :1/9.
For example r<u withu=r+d gives weights4^-r 2^-d 2^-v, andbase d+v.
Summing gives H(z)=4(1+z)/[9(2-z)^2]. The independent marginal reserve has
PGFz/(2-z), whileP(sametype)=sum p_t^2=1/9. Therefore
 G_J(z)=1/9+4z(1+z)/[9(2-z)^3], |z|<2.
Extracting coefficients:
 P(J0)=1/9,
 P(J=h)=h(3h-1)/(18*2^h),h>=1,
 P(J>=h)=(3h^2+5h+8)/(9*2^h).
Exactmoments:EJ4,EJ^2=212/9,VarJ68/9,EJ^3=172,EJ^4=13508/9.

IMPORTANT CLOCK: these include zero-change boundaries. Conditioning on a
GENUINE type change gives probability8/9, mean9/2 and marginal variance25/4.
Do not label mean4 as the mean per actual type change.

3. COB0UNDARY, LIMITS AND EXPLICIT FINITE CONTROL
The canonical identity holds at EVERY boundary, including identical types:
 J_i=S_(i+1)+R_(i+1)-R_i.
Thus Q_m=sum_(i0..m-1)J_i=Z_m+R_m-R0,
whereZ_m=sum_(j1..m)S_j is a sum of IID costs, ES4,VarS4. In fact
 P(Z_m=z)=binom(z-1,2m-1)*2^-z,z>=2m.
The boundary residues are dependent onZ_m, so do NOT call this an independent
sum. Independence is unnecessary for the following conclusions.

For integera>=1,
 P(|R_m-R0|>=a)<=2^(1-a),
because one of the two reserves is then at leasta+1. Thus
 P(|Q_m-4m|>=t+a)
 <=P(|Z_m-4m|>=t)+2^(1-a).
Uniformly over partial sums throughm, the boundary difference is bounded by
max_(i<=m)R_i-1. A union bound on dyadic m and BorelCantelli show that
max_(i<=m)R_i=O(logm)almost surely. Consequently Q_m-Z_m=O(logm)a.s., the
LLN has limit4, and the functional CLT has Brownian variance4 by Donsker and
uniformly negligible coboundary. The one-time marginal variance68/9 is NOT
the long-run variance. This is a useful precise reason an IID-refill model fails.

4. INDEPENDENT COVARIANCE DERIVATION
Define K_h=Cov(R0,R_h),A_h=Cov(R0,S_h),h>=1.

Reservoircovariance: if a typechange occurs by h, R0 is measurable from the
past through h, while E(R_h-2|thatpast)=0. If T0=...=Th=i, then
R0=h*S_i+R_h. The event has probabilityp_i^(h+1), and the conditional
variance ofR_h is2. Therefore
 K_h=2*sum_i p_i^(h+1)=2/(2^(h+1)-1)^2, includingK0=2.

Costcovariance: let
 c_(r,s)=E[R0*(S1-4)|T0=(r,s)].
This is NOT E[R0-2|T1=(r,s)]; confusing these meanings makes negative values
appear impossible. Tail indicators provide an independent derivation:
 for1<=ell<=r, eventR0>=ell means nextu>=ell, with probability2^(1-ell)
 and conditional nextS meanell+3, so contribution(ell-1)*2^(1-ell);
 forr<ell<=r+s, eventR0>=ell means nextu=r,nextv>=ell-r, with probability
 2^(1-ell) and conditional nextS meanell+1, contribution(ell-3)*2^(1-ell);
 beyondr+s, nexttype mustequal(r,s), giving total tail contribution
 2*p_(r,s)*(r+s-4).
Summing the three terms gives
 c_(r,s)=2-4*2^-r-6*2^(-r-s).

In A_h, all first changes before h have zero covariance withS_h. If no change
occurs inT0,...,T_(h-1), its conditional contribution is c_i, with probability
p_i^h. Hence A_h=sum_i p_i^h c_i. Withz=2^h this equals
 2/(z-1)^2-4/[(2z-1)(z-1)]-6/(2z-1)^2
 =-2(z-2)(3z-2)/[(z-1)^2(2z-1)^2].
In particularA1=0, despite dependence of reserves on future types.

Now gamma_h=Cov(J0,J_h),h>=1. Independence ofS1 from later costs and from
all later conditional reserves gives
 gamma_h=A_h-A_(h+1)+2K_h-K_(h-1)-K_(h+1).
Exact algebra reduces this to
 -4z(36z^2-34z+7)/[(z-1)(2z-1)^2(4z-1)^2],z=2^h.
Every denominator ispositive and36z^2-34z+7=36z(z-1)+2z+7>0, so EVERY
nonzero-lag covariance isnegative. gamma1=-664/441. The kernel isO(4^-h),
and summation telescopes:
 sum_(h>=1)gamma_h=A1+K1-K0=-16/9.
Thus VarJ+2sumgamma=68/9-32/9=4, agreeing with the independent coboundary CLT.
This is a complete independent algebraic audit, not a deduction from the CLT.

5. OPTIONAL GENUINE-CHANGE CLOCK (DO NOT INCLUDE WITHOUT SECOND AUDIT)
Let I_i=1[T_i!=T_(i+1)], p=EI8/9. The indicators are1-dependent:
 q=P(sametype)=1/9; P(threetypesequal)=1/49.
Their long-run variance is p(1-p)+2(1/49-1/81)=152/1323.
Since sum_t S(t)p_t^2=8/27,
 Cov(S_i,I_i)=Cov(S_(i+1),I_i)=4/27,
and all other crosses vanish. The joint long-run covariance rate of(cost,count)
is8/27. The vector(S_i,I_i) is1-dependent, square-integrable, so the usual
multivariate finite-dependence functional CLT applies. The coboundary is still
negligible on that clock. Time change to exactlyn genuinechanges has mean9n/2
and variance coefficient
 [4+(9/2)^2*(152/1323)-9*(8/27)]/(8/9)=807/196.
This differs from the conditional marginal variance25/4. The random inverse
clock step needs to be written, not assumed by treating selected J as IID.
The main all-boundary result is already sufficient for a focused paper; do not
expand the paper just to include a half-audited extra theorem.

6. NOVELTY / PRIOR-ART VERDICT
Classical: exact parity cylinders, IID geometric valuations under Haar,
canonical run compression, and standard coboundary/CLT machinery.
Paper15 already has periodic divisibility, arbitrary phases/partialperiods,
effective dictionary complexity and fixed-budget excursion bounds. Do not
rebrand a finite-run cylinder recoding or an8^m-to-smallconstant polish as the
main advance. Every finite canonical type/run sequence is realizable by the
standard word cylinder, so its finite-language admissibility alone supplies no
new arithmetic restriction.

What looks plausibly distinctive in the scoped search: the reserve-refill
observable, its exact selected marginal law, all-lag negative covariance kernel,
and the explicit short coboundary producing variance4 rather than68/9. The
paper should make this observable-calibration contribution central and label
positive-integer genericity/source-only control as open. No exhaustive priority
claim is justified by our searches.

The relevant search returned macindoe/collatz/publication.md and a binary-block
preprint discussing classical geometric block laws; I read only search excerpts,
not their complete proofs. They reinforce the classical attribution above but
do not establish novelty/non-novelty of this exact selected-refill observable.
Search URLs are saved separately if the parent wants a primary fulltext check.

7. REPLAY / EVIDENCE
independent_audit.py derives four exact geometric cases, checks PGF rational
values, truncates allfour independenttypeparameters at18 and bounds omitted
mass by4*2^-18 while checking PMF levels0..12, and checks moments/change-clock
covariance arithmetic with Fractions. AUDIT.json PASS. These finite controls
corroborate the written universal calculations; they do not prove integer
orbit genericity or certify all the mathematics mechanically.

RECOMMENDED PUBLICATION POSITION
A focused paper on exact adaptive-interface statistics under Haar, with
deterministic ledger/coding limits as motivation and exact certificates, is
scientifically defensible if the parent's primary literature check finds no
direct predecessor. A 'Collatz breakthrough' or new deterministic debt ceiling
is not supported. All work in this worker directory is complete and CPU-only.
