Proof of both the all-order polynomiality statement and all four exact formulas. Put
Pn(q)=21−nT2(n;q)=r≥0∑ar(n)qr.
The source's free, energy-preserving interface involutions identify Pn(q) with the generating polynomial of canonical quotient triangles. Their unique ordered direct-sum decomposition gives
F(z,q):=n≥0∑Pn(q)zn=1−C(z,q)1,(1)
where C counts nonempty irreducible canonical triangles.
Let bj be the j-th bottom color and Aj its active set. Then
∣Aj∣=j,ej=#{c∈Aj:c>bj},E=j∑ej.
Set
D=j∑(j−bj)+=j∑(bj−j)+=i∑hi,hi=#{j≤i:bj>i},rj=#(Aj∖[j]).
Then
D≤E,rj≤ej+(bj−j)+,j∑rj≤2E.
If hi=ri=ri+1=0, the canonical interface is an ordered direct-sum cut. Therefore every noncut is counted by a positive hi or is adjacent to a positive rj, giving
#{noncuts}≤D+2j∑rj≤5E.
An irreducible size-s configuration has s−1 noncuts. Thus
C(z,q)=z+r≥1∑cr(z)qr,cr(z)∈Z[z],degcr≤5r+1.(2)
This proves a genuine finite cutoff, rather than an extrapolation from observed data.
For completeness, all irreducibles through energy five can now be enumerated by an exact finite-state transfer. A state before bottom position j consists of (U,A,e,d), where U is the used bottom-color set, A the active set, e accumulated energy, and d accumulated positive displacement. Choose b∈A∖U and set
U′=U∪{b},e′=e+#{c∈A:c>b},d′=d+(j−b)+.
For j<n, choose an increasing interface pair u<v outside A−=A∖{b} and set
A′=A−∪{u,v}.
Reject precisely those interfaces satisfying
U′=[j],A−∪{u}=[j],
since these and only these are direct-sum cuts. Discard e′>5 or d′>5, which is rigorously safe because D≤E. By (2), only sizes n≤26 can contribute.
The complete resulting coefficient table, with columns [q]C,[q2]C,[q3]C,[q4]C,[q5]C, is
nhline234567891011121314151617≤n≤26113100000000000002053636101000000000030150414710433120171000000040047923517111436111155414151625324100005001011521709674462174057224681161330699271959636284353110.(3)
Vanishing for all larger sizes follows from the proved bound (2).
Write
[qk]F(z,q)=(1−z)k+1Nk(z),N0(z)=1.
Equation (1) yields the exact integer-polynomial recurrence
Nk(z)=j=1∑kcj(z)Nk−j(z)(1−z)j−1.(4)
Using the complete table (3),
N2N3N4N5=5z3+32z4+6z5−15z6−3z7,=z3+48z4+325z5+25z6−368z7+4z8+81z9+9z10,=47z4+784z5+2670z6−110z7−5608z8+757z9+2928z10−456z11−360z12−27z13,=10z4+1112z5+12652z6+15341z7−9578z8−56080z9+6607z10+64728z11−21569z12−15849z13+4239z14+1431z15+81z16.
Since degNk=3k+1, coefficient extraction gives, for n≥2k+1,
k!ak(n)=s∑[zs]Nk(z)j=1∏k(n−s+j).
Expanding gives exactly
a2(n)a3(n)a4(n)a5(n)=225n2−49n−116,=6125n3+15n2−3104n+1980,=24625n4+2650n3−36877n2−31390n+244728,=1203125n5+32500n4−290925n3−1585240n2+7120060n+5588400,n≥5,n≥7,n≥9,n≥11.
Each threshold is sharp: at n=4,6,8,10, respectively, the actual values are 47,1825,64622,2207081, whereas the extended polynomials give 44,1816,64595,2207000.
Finally, (2) and (4) prove that every fixed ak(n) is eventually polynomial of degree at most k. Source Proposition 3.5 gives c1(z)=z2+3z3+z4; the maximal pole in (1) therefore has numerator value
Nk(1)=c1(1)k=5k=0.
Hence its degree is exactly k, with leading coefficient 5k/k!, for every k≥0. This proves the complete conjecture, including all four separate formulas and their exact onset thresholds.
Source: Blitvić and Petrov, Colored interlacing triangles and Genocchi medians, Conjecture 4.2, https://arxiv.org/abs/2602.04390.