PINA COLADA — ADVERSARIAL INTERFACE TOLLS
25 October2026. Bounded CPU, no API writes/GPU/paid compute/SYMBIOSE/children.

RESULT / SCOPE
There is an exact interface-capacity ledger, but it does not yield a source-only
debt ceiling without a new bound on its switch charges. Two natural cheap tolls
fail explicitly. The canonical block grammar makes EVERY actual type change a
positive recharge, not a rare resonance. These are finite algebra/certificate
results, not a Collatz resolution, new record or exhaustive novelty claim.

1. THE EXACT GENERAL SWITCH LAW
For a finite exponent wordw let
 A_w=3^length(w), B_w=2^sum(w), C_w the positive affine intercept,
 h_w=A_w-B_w (odd), f_w(x)=h_w*x+C_w.
Its periodic rational pin is-C_w/h_w. Exclude zero defect (a pin itself), which
needs separate cycle handling rather than a finite-capacity ledger.
For wordsw,v and one odd boundary statex,
 h_w*f_v(x)=h_v*f_w(x)+Delta,
 Delta=h_w*C_v-h_v*C_w.
Write a=v2(f_w(x)), b=v2(Delta) whenDelta!=0. Odd h's imply:
 a<b -> v2(f_v(x))=a (no capacity change),
 a>b -> v2(f_v(x))=b (capacity loss),
 a=b -> v2(f_v(x))>a (strict positive recharge, unbounded in principle).
ForDelta0 the capacities coincide. Thus positive UNWEIGHTED capacity changes
occur precisely at equal-depth resonances for distinct pins. Paying only the
fixed b is unsound; equality can cancel arbitrarily many leading binary bits.

2. SAFE TELESCOPING AND ITS LIMIT
Normalize capacity c_w(x)=v2(f_w(x))-1>=0 at positive odd non-pin states.
L exact copies ofw consume L*S_w units of capacity. Their debt increment is
L*d_w, d_w=2*length(w)-S_w<=S_w. Along full periodic runs,
 D_total<=c_initial-c_final+sum(signed capacity jumps)
        <=c_initial+sum(positive capacity jumps).
This is valid even when some d_w are negative. It is an ENDPOINT bound; add
the relevant within-period prefix maximum to control intermediate debt.

For rho_w=d_w/S_w the EXACT weighted identity is
 D_total=rho_first*c_initial-rho_last*c_final
         +sum(rho_new*c_new-rho_old*c_old).
A positive WEIGHTED toll can arise from a rho increase even without an
unweighted capacity increase. If rho_last<0, the terminal term is positive
and must not be dropped. Keeping these distinctions prevents false bounds.

3. CANONICAL PIT-STOP SWITCHES ARE ALL RESONANT
For canonical type(r,s),word1^(r-1)(s+1), the rational pin is
 alpha_rs=(3^r-2^r)/(2^(r+s)-3^r), withv2(alpha_rs+1)=r.
At a boundary whose NEW canonical type is(u,v), a DIFFERENT old type(r,s)
has outgoing defect depth
 t=min(r,u) ifr!=u,
 t=r+min(s,v) ifr=u ands!=v.
The new depth is>=u+v+1>t. Thus it is a STRICT positive recharge at every
actual canonical type change. The separator depth also equalsv2(Delta).
This follows immediately by comparing the2-adic distances to the two pins,
or by the shared all-ones prefix and first differing valuation.
So resonance rarity or independent random switch phases cannot be assumed.
Any proposed Eratosthenes-like sieve on rare recharge events must confront
the fact that the canonical events are exactly the type changes being counted.

The universal lower charge is
 J>=u+v+1-min(r,u) ifr!=u,
 J>=1+max(0,v-s) ifr=u.
These are LOWER bounds, not the upper charge control needed to close debt.

4. FIXED DETERMINANT TOLL FAILS WITH ONE CHANGE
Take oldcanonicaltype(2,1),pin-5, newtype(3,1),pin-19/11.
Their separator depth is2. For L>=5, choose the unique least oddboundaryy
withy=4mod9 and11y+19=2^L mod2^(L+1), andputn=(8y-5)/9.
The old source depth is5, outgoing old depth2, and incoming new depthL.
It realizesoldword(1,2), followed by at leastfloor((L-1)/4) copies of
newword(1,1,2). All these prefixes strictly exceed the originalsource.

Concrete L13: n29787 -> oldboundary33511; incoming newcapacity12 versus
oldcapacity1, a recharge11. Three newcopies give totaldebt7, exceeding
initialcapacity4 + determinantvaluation2 =6.
This refutes a reservoir ledger that charges onlyv2(Delta) per change.
It DOES NOT refute a source-size-aware bound: the source size grows withL.
L13,17,25,41 cases are explicitly replayed. The script also retains an older
noncanonical generalword reset withv=(1,1,2,1), clearly separate from this
canonical regression; both are classical CRT/pin mechanisms, not new priorities.

5. SOURCE BITS + ONE PER CANONICAL CHANGE ALSO FAILS
Use the previously established oldrecord source12235060455 (34bits).
At canonical block117, before first descent, oddtime333,S518,debt148.
There have been104 changes of the uniquely determined canonical type(r,s),
so sourcebits+changes=34+104=138<148. This is not a posthoc dictionary choice.
Counting pieces instead gives139 and also fails. It does not refute a larger
constant toll, weighted cost, or every source-size/switch complexity theorem.
No blind replacement by a new coefficient is justified by this finite example.

At this prefix, exact capacity accounting has initialcapacity8, finalcapacity0,
totaldivision518, positive switch recharge510 across104 resonant switches.
The safe unweighted upper is518, while exact weighted accounting recovers148.
Thus the unweighted conservation law by itself is weak here: essentiallyD<=S,
not an independent ceiling in terms of source alone.

6. OTHER INTERPRETATION TRAP
Contraction of height is not nonpositive debt. Canonicaltype(3,2) has slope
27/32<1 but debt+1. Actual55->83->125->47 realizesword(1,1,3).
A high orbit can accumulate debt while a layer contracts toward the original
source; pricing only transitions between expanding/contracting height phases
would omit positive debt from such contracting layers.

WILDCARD STATUS
Rational decomposition/pin comparison and2-adic cancellation were useful
associations. No theorem from Ramanujan, Erdos-Straus, or a prime sieve is
silently imported. The analogy produced testable algebra, not evidence for
its truth; the proof is the displayed identity and exact valuation logic.

REPLAY / EVIDENCE
python interface_check.py   (without-O; standard library only)
PASS20889 general pair identities;1260 canonical separator checks and3745
actual canonical realizations;7 generalword resets and4 canonicalpair resets;
the full knownrecord is independently reconstructed using canonical blocks
and every block is independently replayed as actual odd steps. Exact weighted
and unweighted telescopes are checked. AUDIT.json contains all reported data.
279 zero-defect cases were explicitly excluded rather than assigned finite
valuation. No broader all-source bound or external validation is claimed.
All recorded calculations have ended; no publication was made by this worker.
