{
  "packet_id": "full_manuscript",
  "scope_summary": "Main conclusion: Theorem 1.1 (thm:main) \u2014 a single deterministic exact algorithm computing F_n x for every n in O(n (log n)^\u03b8 (log log n)^{4-\u03b8}) operations (\u03b8=log_m \u03bb, m=10^6, \u03bb=m-\u0394/2^71, \u03b8<1-2\u00b710^{-13}), in an exact complex field-operation model with unrestricted coefficients, one specified root \u03b6_{D_*} with D_*<1024n^3, and unit-cost O(log n)-bit indexing; Corollaries 1.2 (decimal exponent) and 1.3 (bilinear convolution). Coverage: all numbered claims in Sections 2\u20135 were audited line by line (network scalar lemma, frame/edge lemma, label tables and residual dimension ledger, padding/exponent constants, batching recurrence, Newton reduction, replay lemma, Toeplitz measured-fit compiler, C-kernel compilation of shears, sector-address lemma, tensor synchronization, prime/working-length lemma, CRT, chirp, root ledger, final estimates, convolution corollary). Appendix (syn:*) was spot-checked only as a nonuniform/companion route explicitly not needed for the main exponent. External references [motif],[exactfourier] are credited only; the manuscript claims self-contained proofs of everything it uses, and I audited those proofs rather than the references.",
  "audited_steps": [
    {
      "anchor": "eq:main-constants / net:parameters / net:wire-count",
      "result": "checked",
      "evidence": "v=C(100,3)=161700; N_0=v^3=4,227,952,113,000,000; I_0=3v^2=78,440,670,000; d_nbr=C(97,3)+3\u00b797=147440+291=147731; v\u00b7d_nbr+101=23,888,102,801; I_0\u00b7that=1,873,798,788,739,316,670,000; +2N_0 gives W=1,873,807,244,643,542,670,000 exactly as stated. 2^70\u22481.18e21<W<2^71\u22482.36e21 so W_*=2^71 is the least power of two \u2265W."
    },
    {
      "anchor": "net:delta / net:s / net:padding",
      "result": "checked",
      "evidence": "\u0394=2N_0-2I_0\u00b7101\u00b7100=2v^2(v-30300)=52,293,780,000\u00b7131,400=6,871,402,692,000,000 (matches). s=Wm-\u0394 matches displayed s. \u03b5=\u0394/(mW_*)=6,871,402,692/2^71=1,717,850,673/2^69 (2^69=590,295,810,358,705,651,712), \u22482.910e-12>2.8e-12; ln m=13.8155<14; 1-\u03b8\u2248\u03b5/ln m\u22482.1065e-13 consistent with displayed decimal and with \u03b8<1-2\u00b710^{-13}."
    },
    {
      "anchor": "Lemma net:scalar",
      "result": "checked",
      "evidence": "(RG x)_S=\u03a3_T x_T(|S\u2229T|-1)/2: value 1 (S=T), 0 (|S\u2229T|=1), -1/2 (size 0), +1/2 (size 2); JV subtracts exactly the size-0 and size-2 terms, so RG+JV=I. Rows 0\u20137 give y\u2190y-JA-Rc+R(c+Gx)+J(A+Vx)=y+x with A,c restored. Stages (x,y)\u2192(x,y+x)\u2192(-y,y+x)\u2192(-y,x) verified, including reversed/negated stage-2 invocation giving x\u2190x-y."
    },
    {
      "anchor": "net:frame / Lemma net:edge",
      "result": "checked",
      "evidence": "J_1 diag(1,i) J_1 = \u00bd[[1+i,1-i],[1-i,1+i]]=C so \u03a6_D=C^{\u2297m}; J diag((-1)^{z\u00b7x}) J = R_z by character orthogonality; i^{[z\u00b7x]}=a+b(-1)^{z\u00b7x} (checks at z\u00b7x=0,1); C_z^2=(a^2+b^2)I+2ab R_z=R_z since a^2=i/2,b^2=-i/2,2ab=1, so C_z^{-1}=R_zC_z. For V=U\u22a5E with orthonormal basis z_j, P_V=P_U+\u03a3 z_j[z_j\u00b7x] and weight is additive mod 4 on pairwise orthogonal vectors (wt(x+y)=wt x+wt y-2|supp\u2229supp|, intersection even), so q_V-q_U=\u03a3 wt(z_j)[z_j\u00b7x] mod 4 with wt(z_j) odd; gives exactly \u03c1 kernels C_z or C_z^{-1}."
    },
    {
      "anchor": "net:telescoping",
      "result": "checked",
      "evidence": "Induction on the network: a wire with current label U holds \u03a6_U^{\u2297f}(virtual value); edge inserts \u03a6_V\u03a6_U^{-1}; a gate with common label acts as (scalar matrix)\u2297I commuting with I\u2297\u03a6_U. Hence total map = diag_sinks \u03a6 \u00b7 S \u00b7 diag_sources \u03a6^{-1} as an identity on arbitrary values in all roles."
    },
    {
      "anchor": "net:labels / Lemma net:residuals (inclusions and ledger)",
      "result": "checked",
      "evidence": "Stage boundary labels agree: outgoing X (E_j\u2297D)\u2297Q_j = incoming X at stage j+1 since E_{j+1}=E_j\u2297D; outgoing Y (P_j\u2297\u27e8t\u27e9)^\u22a5\u2297Q_j = B_{j+1}\u2297\u27e8t_{j+1}\u27e9\u2297Q_{j+1}. Stage 1 incoming X=U_d, Y=0; stage 3 outgoing X=D^{\u22973}, Y=U_d^\u22a5. All eleven listed residuals verified by E=B\u22a5P, D=\u27e8t\u27e9\u22a5t^\u22a5, t_Y^\u22a5=\u27e8t_X\u27e9\u22a5\u27e8t_X,t_Y\u27e9^\u22a5 (needs t_X\u00b7t_Y=0: neighbors). Independent per-wire totals with \u03b2=dim B=h^{j-1}-1: X and Y wires each 99h^{j-1} per stage, summing to 999999=m-1; central wire (h^{j-1}-1)h+3h+h^3-h^j=m+2h; side wire \u03b2+(99\u03b2+1)+98+1+h^3-h^j=m. Hence s=Wm-2N_0+2\u00b7I_0\u00b7101\u00b7100 and the only decreases are central 3\u21924 of dimension h."
    },
    {
      "anchor": "Stage-2 touch order (text after net:labels)",
      "result": "checked",
      "evidence": "Reversed rows 7,6,5,4,3,2,1,0 with source Y, target X touch physically Y-side, Y-center, X-side, X-center, Y-center, Y-side, X-center, X-side, matching table rows 0\u20137; the side wire A_{ST} is touched at rows 0,5 with its Y triple and 2,7 with its X triple in every stage, so the residual table applies with t_X,t_Y swapped."
    },
    {
      "anchor": "Lemma net:residuals (orthonormal bases)",
      "result": "checked",
      "evidence": "Each residual contains a norm-one vector (coordinate units outside supports of size \u22643,6,3^{j-1}; indicator tensors of odd weight). Absorption (w,p,q)\u2192(w+p,w+q,w+p+q): all three have norm 1 and pairwise products 1+1=0 over F_2; span preserved. Standard fact: nondegenerate non-alternating F_2 form has orthonormal basis."
    },
    {
      "anchor": "Proposition net:finite-interface",
      "result": "checked",
      "evidence": "Exceptional route: q_{U^\u22a5}(x)-q_U(x)=wt(x)-2\u00b727[u\u00b7x]\u2261wt(x)+2[u\u00b7x] mod 4, so ratio is R_u C^{\u2297m} per column (R_u commutes with C^{\u2297m}); correction by translation and sign/swap gives C^{\u2297mf} on every role. s directional steps, each conjugate by a linear address change to C^{\u2297f} on 2^{k-f} fibers; inverse steps add a bit-flip translation."
    },
    {
      "anchor": "Lemma net:indexing",
      "result": "checked",
      "evidence": "Radix-2^m digit prefix tree: every internal node has \u22652 children, 2^k leaves \u21d2 <2^{k+1} nodes; per-node O(1) with precomputed (constant-size, 2^m-entry) tables and O(f) stride preparation; selected bits as least-significant field makes fibers and W_*-fiber batches contiguous. Table index needs m bits, which is \u2264 k+O(1) in the only regime (k\u2265K=72m) where it is used."
    },
    {
      "anchor": "Theorem net:tensor-bound recurrence",
      "result": "checked",
      "evidence": "k=mf+r, f\u226572, k-f\u2265(m-1)f\u226571 \u21d2 2^71 | 2^{k-f}: exact batches. S=s+(W_*-W)m=W_*m-\u0394 steps, each (2^{k-f}/W_*) recursive calls T(f) plus O(2^k) permutations; T(k)\u2264AW_*2^k+S\u00b72^{k-f}T(f)/W_* \u21d2 t(k)\u2264A+\u03bb t(\u230ak/m\u230b), \u03bb=S/W_*. Unrolling d\u2264\u2308log_m(k+1)\u2309 levels: t(k)\u2264(A/(\u03bb-1)+B)\u03bb^d, \u03bb^d\u2264\u03bb(k+1)^\u03b8. Workspace \u03a3_j W_*2^{\u230ak/m^j\u230b}=O(W_*2^k); single-array bound by zero-initializing W_*-1 roles. Correctness needs arbitrary-role transform (Remark net:role-remark), supplied by net:telescoping."
    },
    {
      "anchor": "Lemma loc:newton-reduction / eq loc:newton / eq loc:fourier-factorization",
      "result": "checked",
      "evidence": "N=F_r P lower triangular; N_{ik}=\u03a0_{j<k}(u^i-u^j)=(-1)^k u^{k(k-1)/2}H_i/H_{i-k} with H_j=\u03a0_{s\u2264j}(1-u^s)\u22600 for j<r (u primitive). LDU uniqueness: P^T L D_N=D_N L^T P forces P^TL=I; hence F=N D_N^{-1} N^T. T has h_0=1, h_j=H_j^{-1}."
    },
    {
      "anchor": "Lemma loc:replay",
      "result": "checked",
      "evidence": "Forward sweep z\u21a6Uz+Tx (U unit lower), readout Fz+Dx with FT+D=M. Sweep/add/undo yields y+Mx+FUz; source-suppressed sweep is z\u21a6Uz, readout FUz subtracted, undo: net (x,y+Mx,z) as a formal identity. Level-wise bipartite shear multigraph of max degree \u0394\u2264max(2,\u03b4), greedy (2\u0394-1)-edge-coloring, no destination is a predecessor within a level; 4 sweeps\u00b72g + 2 output rounds\u00b7a = 8g+2a shears, O((\u03b4+1)(H+1)) layers."
    },
    {
      "anchor": "Lemma loc:toeplitz (eqs loc:cross, loc:displacement-entry, loc:reconstruction, loc:fit)",
      "result": "checked",
      "evidence": "T_v=[[I,0],[E,I]]diag(T_p,T_{v-p}) with E=BT_p^{-1}, E_{ij}=\u03a3_{\u03bd=j}^{p-1}h_{i-\u03bd}g_{\u03bd-j}; shifting \u03bd gives E_{ij}-E_{i-1,j-1}=-h_{i-p}g_{p-j}; M=\u03a3_l J^l \ud835\udc9f (J^T)^l by nilpotence; outer-product action via two convolutions. Admissibility with b_0=\u230av/(D log_2 v)\u230b, D\u226516K_0+16: a+e\u2264v/8, g\u22644K_0v/D\u2264v/4 \u21d2 fit; fallback only for bounded v; O(log v) chunks per interval \u21d2 O(log^3 v) layers per level, d(v)\u2264d(\u2308v/2\u2309)+O(1+log^3 v) \u21d2 O(1+log^4 r). Padded FFT lengths <4r so power-of-two root order <8r."
    },
    {
      "anchor": "eq loc:three-kernel / loc:nonzero-shear / loc:nonzero-split",
      "result": "checked",
      "evidence": "SCS=diag(1,i)C diag(1,i)=[[a,bi],[bi,-a]] with bi=a \u21d2 H'=a^{-1}SCS. H'diag(-3,1)H'=[[-2,-4],[-4,-2]]; H'diag(2,1)\u00b7that=[[-8,-10],[0,-6]]=K. diag(-1/8,-1/6)K=[[1,5/4],[0,1]]; conjugation by D_t=diag(4t/5,1) gives [[1,t],[0,1]], valid for t\u22600. \u03ba=1+|\u03bc|^2>0 and \u03bc-\u03ba\u22600 since |\u03bc|<1+|\u03bc|^2; the two shears compose to shear by \u03bc. Conjugate via inverse-root DAG evaluation is exact."
    },
    {
      "anchor": "eq prefix-nodes / Lemma lem:sector-address",
      "result": "checked",
      "evidence": "\u03a3_j \u03a0_{i\u2264j} r_i \u2264 R \u03a3 2^{j-\u2113}<2R since r_i\u22652. Sector start: earlier sectors classified by first differing axis j contribute (\u03a0_{i<j}q_i)\u00b7p_j\u00b7\u03c4_j; updates eq sector-updates maintain S,Q,o,I in O(1) per prefix extension; block prefixes \u2264 digit prefixes since b_j\u2264r_j."
    },
    {
      "anchor": "Proposition prop:tensor-fourier",
      "result": "checked",
      "evidence": "(\u2297A_j)(\u2297B_j)=\u2297(A_jB_j) makes per-slot tensoring of padded local words exact regardless of borrowed coordinates. Pair slot = direct sum over sectors of C^{\u2297k}, k\u2264\u2113; \u03a3_sectors 2^k(k+1)^\u03b8 \u2264 (\u2113+1)^\u03b8 R; \u2264R sectors so width-one overhead O(R); O((1+log r_max)^4) slots. Compilation poly(\u2113,r_max,log R) once."
    },
    {
      "anchor": "Lemma lem:prime-lengths / eq working-bounds",
      "result": "checked",
      "evidence": "4^u/(2u+1)\u2264C(2u,u)\u2264(2u)^{\u03c0(2u)} (each prime power \u22642u); u=(j+1)^2 \u21d2 \u03c0(2u)>j+1 eventually \u21d2 r_j=O((j+1)^2); R_\u2113\u2265(\u2113+2)!/2 \u21d2 log R_\u2113=\u0398(\u2113 log \u2113) \u21d2 \u2113=\u0398(log n/log log n). L=R_\u21132^e minimal \u22652n: L<4n, 2^{e-1}<r_{\u2113+1}, e=O(log \u2113)."
    },
    {
      "anchor": "eq crt-fourier / Section 5.2 cost / eq working-transform",
      "result": "checked",
      "evidence": "E_qE_{q'}\u22610 (q\u2260q'), E_q^2\u2261E_q mod L \u21d2 \u03b6_L^{jk}=\u03a0_q \u03b6_q^{d_q j_q k_q}; \u03b6_q^{d_q} primitive, handled by loc:compiler at any primitive root. Odd axes via prop:tensor-fourier per 2^e-fiber, radix-2 FFT on the 2^e axis: O(L[(\u2113+1)^\u03b8(log(\u2113+2))^4+1+e])+poly log n. F_L^2=LJ \u21d2 F_L^{-1}=L^{-1}JF_L."
    },
    {
      "anchor": "eq chirp / eq master-root / eq root-size",
      "result": "checked",
      "evidence": "2jk=j^2+k^2-(k-j)^2; residues -(n-1)..n-1 distinct mod L\u22652n; convolution with fixed kernel is linear in x; chirp table by geometric progressions in O(n). D_*=2n\u00b7L\u00b72^{\u2308log_2 16L\u2309} is divisible by 2n, every q|L, every power of two \u22648L\u22658r, and 4; D_*<2n\u00b7L\u00b732L=64nL^2<1024n^3. Three full transforms, one in preparation."
    },
    {
      "anchor": "Proof of thm:main and cor:decimal (final estimates)",
      "result": "checked",
      "evidence": "T(n)=O(n \u2113^\u03b8 (log \u2113)^4)+poly(log n)+O(1) with \u2113=\u0398(log n/log log n) gives O(n(log n)^\u03b8(log log n)^{4-\u03b8}); \u03b5>28\u00b710^{-13}, \u03b7_0 ln m<\u03b5, \u03bb=m(1-\u03b5)<m^{1-2\u00b710^{-13}} \u21d2 \u03b8<1-2\u00b710^{-13}; (log log n)^{4-\u03b8}<(log n)^{10^{-13}} eventually. All indices polynomially bounded \u21d2 O(log n)-bit words suffice."
    },
    {
      "anchor": "Corollary cor:polynomial-multiplication",
      "result": "checked",
      "evidence": "M=2n-1 avoids wraparound; (F_M c)_j=(F_M \u00e3)_j(F_M b\u0303)_j; c=M^{-1}JF_M p; 3T(M)+O(M); D_*<1024M^3<8192n^3; M^{-1} by doubling and one division by M\u22600. Bilinear, as stated."
    },
    {
      "anchor": "Appendix syn:local-certificate / syn:amplification (companion, not needed for main exponent)",
      "result": "conditional",
      "evidence": "Call count g\u2264(Sf+71W_*+2H)2^{mf-1}=b2^{b-1}-(\u0394f-2H)2^{mf-1}<b2^{b-1} for f>2H/\u0394 checks. Binomial recurrence f(k)\u2264C_0+g E f(K), K~Bin(\u230ak/b\u230b,q/Q) and gq/(bQ)<1 \u21d4 g<bq^{b-1} checks. Lemma syn:extraction (Noether normalization/Nakayama) and Theorem syn:printer were read but only lightly verified; they are explicitly not prerequisites of thm:main."
    }
  ],
  "objections": [
    {
      "anchor": "Theorem net:tensor-bound and Lemma net:indexing (fixed tables and constant A)",
      "classification": "plausible_issue",
      "claim": "The uniform bound is asymptotically valid but the 'absolute constants' include a 2^{10^6}-entry address table, S\u22481.87\u00b710^27 permutation/kernel passes per level, and W_*=2^71 zero-initialized roles; the model accommodates m-bit table indices only because the network is invoked solely for k\u2265K=72m, which the manuscript asserts but does not isolate as a hypothesis of the indexing model.",
      "evidence": "Lemma net:indexing: 'a finite table gives, for each possible column digit' (2^m digits); Theorem net:tensor-bound: 'For k<K, apply the k factors ... one at a time'; model: 'random access on O(log(n+2))-bit words'.",
      "derivation": "Table index has m=10^6 bits; array address has k+O(1) bits; m\u2264k holds exactly when k\u2265K. In thm:main sectors have k\u2264\u2113\u2264log_2 n, so tables are touched only when log n\u226572\u00b710^6. Within that regime the cost accounting is consistent; outside it the direct method is used. No inequality fails.",
      "hypotheses_checked": [
        "k\u2265K implies m-bit indices fit the k+O(1)-bit address model",
        "k<K uses no network tables",
        "constants independent of n"
      ],
      "possible_repair": "State explicitly in Theorem net:tensor-bound that network tables are accessed only for k\u2265K, so table indices are within the address width; this is already implied by the proof structure.",
      "dependency_impact": "None on correctness of the O(\u00b7) statement; affects only interpretability (crossover astronomically large, as the manuscript admits)."
    },
    {
      "anchor": "eq loc:convolution-bounds and eq loc:chunk-bounds",
      "classification": "unverified_step",
      "claim": "The explicit bounded-fan-out (including designated output uses), fan-in-two, depth O(log) convolution DAG with the stated absolute constants K_0,K_1,\u03b4_0 for the three-branch sum is asserted with a sketch (binary FFT recursion, pointwise fixed-scalar products), not written out; the measured-fit rule makes K_0 unnecessary at run time, but the O(log^4 r) depth bound relies on K_0,K_1,\u03b4_0 existing.",
      "evidence": "Section 3.2: 'This proves (loc:convolution-bounds), including all multiplications by fixed scalars' and 'Using three branches and summing their outputs gives a specific DAG with absolute constants K_0,K_1,\u03b4_0'.",
      "derivation": "Standard radix-2 FFT butterflies a+\u03c9b, a-\u03c9b have fan-in 2 and each intermediate is read twice; pointwise scalars and inverse FFT keep fan-out \u22642 per level; a three-branch sum reads each branch output once and each source three times. This yields bounded \u03b4 and g=O(v log v), H=O(log v); I accept it as standard but it is not formally proved in the text.",
      "hypotheses_checked": [
        "padding to power of two \u22644r",
        "fixed-vector transform is prepared scalar data",
        "outputs read once per branch"
      ],
      "possible_repair": "Give the explicit \u03b4_0 (e.g., \u03b4_0\u22646) and the butterfly-level fan-out argument in one paragraph.",
      "dependency_impact": "Only the exponent 4 of log log n in eq:main-bound depends on H=O(log v) and bounded \u03b4; the power-saving exponent \u03b8 does not."
    },
    {
      "anchor": "Lemma loc:toeplitz depth recursion d(v)\u2264max{d(\u230av/2\u230b),d(\u2308v/2\u2309)}+O(1+log^3 v) and Proposition loc:compiler 'fixed repeating pattern'",
      "classification": "unverified_step",
      "claim": "Running the two child calls 'in parallel' as mixed rounds requires padding children of unequal round counts (and sub-blocks that hit the bounded-v direct fallback) with idle rounds so that the parent's cross update begins after both finish; the manuscript states padding at the Proposition level but the recursion as written implicitly assumes synchronized children.",
      "evidence": "'Child calls use disjoint blocks, hence the mixed-round depth satisfies ...'; later 'An idle round uses identity monomials ... Padding the shorter words with idle rounds synchronizes any finite collection'.",
      "derivation": "Depth d(v) is a max over children plus cross-update rounds; realizing it as a word on exactly v coordinates requires inserting identity rounds on the faster child, which is permitted (identity monomial, singleton C-slots) and does not change O(1+log^4 v). Thus the gap is expository, not mathematical.",
      "hypotheses_checked": [
        "idle rounds are available slot types",
        "children act on disjoint coordinate sets",
        "fallback words have O(1) rounds for bounded v"
      ],
      "possible_repair": "State that each child word is padded to the common depth d_child(max) with idle rounds before the cross update.",
      "dependency_impact": "None on the bound; affects only the explicitness of the compiler's output format."
    }
  ],
  "unverified_obligations": [
    {
      "anchor": "[motif] Proposition 8 and Section 3.5; [exactfourier] Theorem 8.3, Lemmas 2.2\u20132.3, Section 3",
      "obligation": "External companion papers are credited only; the manuscript reproves the network and compiler it uses. No claim in thm:main's proof chain was found to depend on an unproved external statement, but the references themselves were not audited."
    },
    {
      "anchor": "Lemma syn:extraction, Theorem syn:printer (Appendix)",
      "obligation": "Verify Noether-normalization substitution and Nakayama step, idempotent resolution over E, and the printer's root ledger (R\u22642^B\u2264n, D=2nPR<8n^3). Only lightly read; explicitly not a prerequisite of eq:main-bound."
    },
    {
      "anchor": "eq loc:convolution-bounds",
      "obligation": "Write out the bounded-fan-out, fan-in-two FFT convolution DAG with explicit constants (see objection 2)."
    },
    {
      "anchor": "Lemma net:residuals, final paragraph",
      "obligation": "Confirm that every residual's orthonormal basis can be fixed deterministically by the stated finite F_2 linear algebra for all ~10^27 edges as a once-only constant-cost preprocessing; conceptually fine, not executed."
    }
  ],
  "assessment": "no_demonstrated_error_found",
  "_meta": {
    "model_id": "claude-fable-5-1",
    "model_key": "fable",
    "provider": "anthropic",
    "effort": "high",
    "packet_sha256": "e9dc36257106c27125cfeb73d82a6af0b4caac4f129012d449b3a097251df921",
    "started_at": "2026-10-13T05:22:50.162210+00:00",
    "finished_at": "2026-10-13T05:33:58.720414+00:00",
    "elapsed_seconds": 668.559,
    "selected_attempt": "fallback",
    "source_only_independent_first_pass": true
  }
}