{"id":"0be5f489-ac0c-4e8e-873c-f96c0165153e","arxiv_id":"2607.09211","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"low","formal_verification":"none","parameter_count":7,"one_line_summary":"Self-replication and polynomial evaluation co-evolve from random Z80 tapes, with task pressure accelerating compact replicators and spatial niches producing emergent curricula.","lead":"Random 32-byte Z80 programs spontaneously evolve both self-replication and polynomial-solving when interaction is competence-gated. The work shows task pressure reshapes how programs copy themselves and that spatial niches create an emergent curriculum for harder tasks.","discovery_kind":"extension","skeptic_critique":{"model":"grok-4.5","headline":"No significant objection identified beyond the reader's already-flagged ISA/predefined-semantics caveat.","rationale":"The paper's strongest claim is experimental and rests on clear controls: task pressure accelerates compact replicators (Fig. 2D/E), metabolic cost induces conditional HALT (Fig. 3), and spatial niches + sparse CNP produce an emergent curriculum that isolated or panmictic populations cannot match (Figs. 4–5). These results do not require the ISA to be 'minimal'; they require only that reproduction is not axiomatic and that competence merely biases interaction probability—both of which hold. The reader's weakest assumption is therefore the right one to flag, but it is already scoped by the authors and does not undermine the internal validity of the reported feedback loop. Public artifacts would raise the verdict from CONDITIONAL to ACCEPT; no further conceptual objection moves the needle. Hence the stress-test leaves the reader's CONDITIONAL/HIGH assessment unchanged.","tokens_in":17733,"tokens_out":545,"duration_ms":7130,"concrete_test":"Release the exact emulator, seed list, and analysis scripts used for Figs. 2D/E and 4; re-run the LDIR-blocked condition (Fig. 2E) and the CNP-rate sweep (Fig. 4A) under an independent implementation. If the Load-Push\to LDD transition still accelerates under task pressure and moderate CNP still yields higher-degree solutions, the headline claim is confirmed; otherwise the reported dynamics are implementation-sensitive.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim (co-evolution of spontaneous self-replication and polynomial solving under competence-gated interaction, with task pressure reshaping replicator architecture and niches producing an emergent curriculum) is internally consistent and supported by the multi-seed ablations in Figs. 2–5 and Methods §§4.1–4.8. The reader's weakest assumption correctly identifies the principal modeling limit: programs begin with a rich, fixed Z80 ISA that already encodes block-copy primitives (LDIR/LDD) and register semantics, so 'spontaneity' is relative to that substrate rather than a pure origin-of-life analogue. That caveat is already acknowledged in the Discussion and does not create an internal contradiction or circularity within the stated experimental frame; the competence gate never supplies a copy operator, and the Load-Push\to LDIR/LDD transition plus curriculum effects remain genuine outcomes of the dynamics. No stronger load-bearing flaw (e.g., hidden hard-coded reproduction, statistical artifact, or logical inconsistency) is evident from the manuscript.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.5","summary":"The paper studies co-evolution of spontaneous self-replication and polynomial evaluation in a population of random 32-byte Z80 programs under competence-gated pairwise interaction. Task success raises interaction probability but never supplies a copy primitive; replication must be discovered via assembly execution. Four findings are reported: (i) replication and task-solving co-emerge from random initialization; (ii) task pressure accelerates the Load-Push → compact LDIR/LDD transition, supported by LDIR-family ablation and mutational-robustness assays; (iii) metabolic penalties promote early/conditional HALT; (iv) spatial niches with moderate cross-niche pollination produce an emergent curriculum for higher-degree polynomials, outperforming panmictic/single-task and smooth-fitness controls. Methods detail multi-seed protocols (typically n=100–200), statistical tests, and replicator counting.","tokens_in":18093,"tokens_out":1361,"duration_ms":25807,"significance":"If the results hold, the work is a clear advance over platforms that hard-code reproduction (e.g., Avida-style copy rewards) and over prior unseeded primordial-soup models that stop at replication alone. The interactive feedback loop—task demand reshaping reproductive architecture, and spatial replication dynamics scaffolding harder functions—is of genuine interest to ALife and evolutionary computation. Strengths include multi-seed ablations (with/without tasks; LDIR blocked), Wilson-interval robustness assays with corrected Z-tests, metabolic-penalty Spearman sweeps, CNP-rate and single-task/smooth-fitness controls, and explicit acknowledgment that programs start with a fixed Z80 ISA. The experimental design is reproducible in outline and the central claims are falsifiable within the stated substrate.","major_comments":[{"comment":"Methods §4.5 / Fig. 2F: mutational robustness is measured on hand-crafted canonical seeds (LDIR 4-byte prefix + random pad; fixed LDD loop; pure Load-Push repeats), not on the evolved population distribution. The hierarchy LDIR > LDD > Load-Push is therefore a property of these seeds under the interaction register init, not a direct measurement of the lineages that actually take over. Because the paper attributes the Load-Push\to LDIR/LDD transition primarily to this robustness hierarchy (with tasks only accelerating it), please either (a) re-run the assay on samples drawn from evolved grids at intermediate epochs, or (b) clearly qualify that the hierarchy is a seed-level proxy and discuss other factors (discovery rate, compatibility with task code, stack/register interference) that could drive takeover.","section":"Methods §4.5 / Fig. 2F"},{"comment":"§2.5, Fig. 4–5, Abstract: the phrase “spontaneous self-replication generates an emergent learning curriculum” risks overstating what is spontaneous. Niche partition, polynomial assignment, and CNP rate π are experimenter-designed; what emerges is the genealogical use of simpler solutions as stepping stones under that structure. The single-task and CNP-rate controls are strong, but the abstract and results framing should distinguish designed spatial structure from the emergent stepping-stone dynamics, and state that the curriculum is contingent on graded task niches plus sparse migration rather than on replication alone.","section":"§2.5 / Abstract"}],"minor_comments":[{"comment":"Methods §4.4: byte-substring replicator counts can miss NO-OP-interleaved variants and can double-count tapes carrying multiple patterns. The authors note the sum approximates population size; still, state the false-negative/false-positive risk explicitly in the figure caption or methods and, if feasible, report a short sensitivity check (e.g., allowing one intervening byte).","section":"Methods §4.4"},{"comment":"Fig. 1E colormap and “green bias for validation success” are described only briefly; a short legend or supplementary panel mapping byte patterns to colors would help readers interpret homogenization vs. niche solutions.","section":"Fig. 1E"},{"comment":"Table 1 and Algorithm 1: the realized fraction of programs selected per epoch (~56% after de-duplication) is mentioned in text but not in the parameter table; adding it would aid reimplementation.","section":"Table 1 / Algorithm 1"},{"comment":"Discussion: the predefined-semantics caveat is appropriately noted; consider also flagging that LDIR/LDD are unusually powerful single-instruction block-copy primitives relative to a more minimal ISA, so the compact-replicator attractor may be substrate-specific.","section":"Discussion"},{"comment":"Author contributions list “C.K.” who does not appear in the author list; please reconcile.","section":"Author Contributions"},{"comment":"Minor typography: “immediate no means” (§2.2) → “no immediate means”; consistent hyphenation of “cross-niche” / “Load-Push”; arXiv date line says July 13, 2026 while v1 is 10 Jul 2026—align.","section":"Throughout"}],"recommendation":"minor_revision","confidential_remarks":"Solid experimental ALife paper that cleanly extends the same group’s 2024 primordial-soup work with competence gating and curricula. Novelty is real but incremental relative to that prior substrate; the main risk for a top venue is overclaiming “spontaneity” / origin-of-life framing given the rich Z80 ISA. The two major comments are clarification/qualification, not redesign. Fit is good for ALife / evolutionary computation venues; borderline for a general Nature/Science-style outlet without stronger multi-ISA or multi-task generalization."},"author_rebuttal":null,"desk_editor":{"model":"grok-4.5","letter":"The useful takeaway is that competence-gated interaction (raise interaction odds if you evaluate a polynomial correctly, never hand out a copy opcode) is enough for self-replication and task-solving to co-evolve from random 32-byte Z80 tapes, and that the tasks actively accelerate the shift from bulky Load-Push replicators to compact LDIR/LDD ones while spatial niches plus sparse pollination produce an emergent curriculum for higher-degree polynomials.\n\nWhat is new relative to their 2024 soup paper is the competence gate, the metabolic-penalty conditional-halting result, the mutational-robustness hierarchy (LDIR > LDD > Load-Push), and the clean demonstration that panmixia or single-task isolation fails where moderate cross-niche pollination succeeds. The experimental package is careful: n=100–200 seeds, task-off controls, LDIR-family ablation, Wilson intervals + corrected Z-tests, Spearman sweeps on the metabolic coefficient, and the smooth-fitness / single-task comparisons. The gate never injects a copy primitive, so the co-evolution claim is not circular inside the stated frame.\n\nSoft spots are real but proportionate. The Z80 already ships block-copy instructions and register semantics; “spontaneous” is relative to that substrate, not a pure origin-of-life analogue—the Discussion already says so. Parameters (p_base, π, C, etc.) are hand-chosen and code is not yet public. Tasks stay inside one polynomial family and 32-byte tapes. None of these overturn the reported dynamics.\n\nThis is for ALife and evolutionary-computation people who care about open-ended systems and origin-of-life analogies. It is not a theory paper and will not reorganize the field, but the feedback loop between task pressure and reproductive architecture is a clean, well-controlled result. I would send it to peer review; the evidence is strong enough to deserve referee time. Worth reading and, if the artifacts appear, citing for the curriculum and architecture-acceleration findings.","headline":"Solid ALife experiment: task pressure really does reshape spontaneous Z80 replicators and sparse niches yield a curriculum; main limit is the rich fixed ISA, already flagged by the authors.","tokens_in":18660,"tokens_out":502,"would_cite":true,"duration_ms":5575,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.5","headline":"In a digital soup of random assembly programs, self-replication and polynomial-solving co-evolve and reshape each other.","keywords":["Artificial Life","Spontaneous Replication","Computational Evolution","Automated Curricula","competence-gated interaction","Z80 assembly","metabolic constraints","spatial niches"],"falsifier":"Remove the task-validation gate (or replace it with uniform random interaction) and measure whether compact LDIR/LDD replicators still overtake Load-Push architectures on the same timescale, and whether high-degree polynomials still appear under identical niche and pollination rates.","tokens_in":18683,"feed_emoji":"🧬","tokens_out":595,"duration_ms":6469,"temperature":0.7,"pith_summary":"Traditional evolutionary algorithms hard-code how programs reproduce. This paper instead starts with random 32-byte Z80 assembly programs and requires self-replication to arise only from mutations and pairwise interactions. A validation step links the two capacities: correctly evaluating a niche polynomial raises a program's chance of interacting, but never itself supplies a copy command. From this setup the authors show four results: replication and task solutions co-emerge; the demand to compute accelerates the shift from bulky Load-Push replicators to compact LDIR/LDD block-copy machines that leave tape free for code; metabolic penalties favor conditional halting that distinguishes validation from interaction; and spatial niches with sparse cross-pollination spontaneously generate a curriculum in which simple polynomial solutions seed harder ones. The paper's claim is that heredity and function are not independent modules but co-evolve under an interactive feedback loop.","feed_headline":"Random code learns to copy itself and solve polynomials","feed_subtitle":"Task pressure reshapes how programs reproduce; niches spontaneously form curricula","key_machinery":"Competence-gated interaction: a program that correctly evaluates its niche polynomial on three sampled inputs receives elevated interaction probability (optionally discounted by a metabolic cost on instruction count), yet the system never supplies a copy primitive; replication must still be executed by the program's own assembly instructions during pairwise tape concatenation.","core_discovery":"Self-replication and mathematical problem-solving successfully co-evolve from pure random initialization under competence-gated interaction. Environmental pressure to evaluate polynomials actively reshapes reproductive architecture, accelerating compact, mutationally robust replicators that leave memory free for computation, while spontaneous replication plus spatial niches produces an emergent curriculum that lets simple solutions act as stepping stones to higher-degree polynomials.","pith_inferences":[],"forward_implications":[],"fun_headline_variants":["Random code co-evolves self-replication and polynomial solving","Task pressure forges compact replicators that free memory for math","Spatial niches turn simple solutions into polynomial stepping stones","Metabolic limits drive conditional halt in evolving replicators","Self-replication and problem-solving emerge together from random soup"],"cache_read_input_tokens":128,"weakest_assumption_plain":"The claim that self-replication is truly spontaneous rests on a fixed microprocessor instruction set whose computational meanings are already predefined, so the 'primordial soup' is not free of prior semantics.","fun_headline_variants_meta":{"raw":{"variants":["Random code co-evolves self-replication and polynomial solving","Task pressure forges compact replicators that free memory for math","Spatial niches turn simple solutions into polynomial stepping stones","Metabolic limits drive conditional halt in evolving replicators","Self-replication and problem-solving emerge together from random soup"]},"model":"grok-4.5","effort":"low","cost_usd":0.004388,"raw_usage":{"total_tokens":1302,"prompt_tokens":760,"num_sources_used":0,"completion_tokens":64,"cost_in_usd_ticks":43880000,"prompt_tokens_details":{"text_tokens":760,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":478,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":760,"tokens_out":64,"duration_ms":6027,"temperature":1.0,"reasoning_tokens":478,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-13T04:39:05.001359+00:00","model_set":{"reader":"grok-4.5"},"falsifier":"Remove the task-validation gate (or replace it with uniform random interaction) and measure whether compact LDIR/LDD replicators still overtake Load-Push architectures on the same timescale, and whether high-degree polynomials still appear under identical niche and pollination rates.","supporting_citations":[],"review_version":1}