{"id":"847bd537-9863-42a7-a4c6-b8995c0ec32c","arxiv_id":"2607.29496","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"Append-only transcript memory keeps fixed-window Transformers finite-state; one pop-enabled channel reaches DCFL and two reach Turing completeness.","lead":"This paper classifies what fixed, finite-precision language models can compute when their memory of past text blocks is managed by appending, copying, or deleting blocks. Append-only memory with a fixed visible window keeps multi-agent systems finite-state; one 'pop' channel reaches the deterministic context-free languages (DCFL) and two reach Turing completeness.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"No significant objection to the central classification; the only soft spot is a discussion-level over-application of Corollary 1 to tool loops, not the theorem itself.","rationale":"The reader's weakest assumption, finite consulted state, is indeed load-bearing and correctly identified; the paper states it as a scope condition rather than hiding it. The reader's conditional verdict is motivated by an over-application of Corollary 1 to tool-and-thought loops. I agree that this over-application is real and should be corrected, but it is a discussion-level flaw and does not undermine the central classification results. The abstract theorems appear correct as written, with the classical DCFL/RE results transparently transferred via Lemma 1. Therefore I do not see a load-bearing concern against the central claim; the appropriate verdict remains the same as the reader's, CONDITIONAL pending a corrected statement of Corollary 1's practical reach.","tokens_in":20156,"tokens_out":29460,"duration_ms":347126,"concrete_test":"Re-read Section V.b and the introduction's use of Corollary 1 and rephrase the claim about deployed tool-and-thought loops: cycling is guaranteed only when no new input symbols are injected after the finite input is consumed. Optionally, model a representative tool-injecting loop as a finite-state transducer on an unbounded input stream and verify that its output is not ultimately periodic, confirming the corollary's hypotheses are violated.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central theorems check out: Theorem 1's summary-space argument, Theorem 2's global-summary collapse, Theorem 3's step-for-step simulations, and Lemma 1's compilation are all internally consistent. The monotone collapse rests on Definition 3's finite consulted state, which the paper states explicitly and scopes carefully (Section I.c, Section V.a), so this is an explicit assumption rather than a hidden flaw. The one genuine soft spot is in the practical reading of Corollary 1 (Section V.b and the introduction): the corollary requires a finite input to be consumed first, after which the internal transition on a finite summary is autonomous. Deployed tool-and-thought loops inject new input symbols (tool outputs) each round, so the system is an FST on an unbounded input stream; such a machine is not forced to cycle. This is a real over-application in the discussion, but it does not affect Trans(MON_m)=Trans(FST) or the DCFL/RE classification. The paper's own scoping (Section I.c, Section V.a) already excludes writable stores and unbounded tool oracles from the monotone model, so the central claim stands.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a transducer normal form TMT_k for the transcript-management layer of fixed finite-precision causal Transformers: one finite controller with k channels, each channel a stack over a finite block alphabet, with per-round stay/push/pop and a fixed status map. The pop-free restriction RTMT_k is the append-only layer; MON_m is the class of monotone multi-agent protocols; OTM_k is the class of k one-channel agents under one orchestration controller. Main results: Trans(RTMT_k)=Trans(FST) for every fixed k; Trans(MON_m)=Trans(FST) for every fixed m; Trans(OTM_k)=Trans(TMT_k); and, via a compilation to the Hopcroft–Ullman presentation, Acc(TMT_1)=DCFL and Acc(TMT_k)=RE for k≥2. Proposition 1 gives a bounded-transcript realization of the management layer; Proposition 2 gives simulation costs. The paper is explicit that the DCFL/RE levels are classical results imported through Lemma 1, and its own theorems are the collapse, the correspondence, and the compilation.","tokens_in":20284,"tokens_out":31160,"duration_ms":331490,"significance":"If correct, the paper provides a clean and useful classification: without pop, any fixed finite population of agents using append/copy/route operations is no more powerful than a finite-state transducer, however long the retained transcript grows; one pop-enabled channel reaches DCFL; two reach RE. The proof strategy is direct and transparent: Theorem 1's summary-space argument, Theorem 2's global-summary collapse, and Theorem 3's simulations are all elementary and checked. The paper is unusually careful about scope: it states the finite-consulted-state condition (Definition 3), fixes precision and population, and identifies the escape routes (growing exact context, hidden-block access, writable stores, unbounded Spawn). It also explicitly credits the classical results, so there is no circularity. The main theoretical claims are sound and the paper is a useful contribution to the formal-language analysis of transcript-managed Transformers.","major_comments":[{"comment":"The claim that Corollary 1 predicts eventual repetition of tool-and-thought loops in deployed agents is an over-application. Corollary 1 applies to an autonomous deterministic finite-state transition after a finite input has been consumed. In a tool loop, each tool output is a new external input symbol; a finite-state transducer driven by an unbounded input stream need not be ultimately periodic. The corollary therefore does not license the prediction of cyclic tool loops. Please revise the relevant sentences to restrict the claim to pure chain-of-thought with no further external inputs, or state explicitly that tool outputs are outside the corollary's hypothesis.","section":"Section V.b, abstract, Section V.i"}],"minor_comments":[{"comment":"The phrase 'whose finite input interface selects stay or push of a specified b∈Λ' is confusing because the orchestration-level push(i,b) is performed directly by the controller, not through the agent's transition interface. Clarify that the one-state agents are placeholders and the actual stack update is done by the orchestration action.","section":"Definition 9 / Theorem 3"},{"comment":"The constant '2' in the bound 2+k(L+1) is not explained. A one-line clarification (dispatcher selection and final output emission) would help.","section":"Lemma 1 proof"},{"comment":"The row for Oren et al. records the idealized uncapped KV cache as RE while the text notes a real bounded cache is finite-state. This is a reasonable reading of the table's 'Representative power' column, but the table could be more explicit that this is an idealized resource reading, not a claim about the cited work's experiments.","section":"Table I"}],"recommendation":"minor_revision","confidential_remarks":"The paper's central theorems are sound and the exposition is unusually careful about scope. The only substantive issue I found is the over-application of Corollary 1 to tool loops in the discussion; this does not affect the main classification but should be corrected. I recommend minor revision."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: the genuinely new result is Theorem 2 — monotone multi-agent transcript management collapses to finite-state transductions — and it is proved correctly. The DCFL/RE hierarchy is classical stack theory imported through a normal-form lemma, and the paper says so rather than overselling it.\n\nThe central mechanism is the finite consulted-state assumption (Definition 3). Given fixed precision, bounded windows, and a fixed population, the summary space (controller state, agent states, exposed tops) is finite, and the monotone protocol can only update it via local maps. The proofs of Theorems 1–3 are step-for-step correct; I checked the summary-space simulation and the two-way simulations in Theorem 3. The paper also handles the classical transfer cleanly: Lemma 1 compiles the TMT normal form into the Hopcroft–Ullman presentation with a step bound, so the one-stack DCFL and two-stack RE results do apply.\n\nThe soft spot is in the discussion, not the math. Corollary 1 says an append-only run with finite input eventually cycles. Section V.b and the introduction apply that to tool-and-thought loops. But those loops inject new tool outputs as inputs each round, so the system is an FST on an unbounded input stream, and an FST on an unbounded stream is not forced to cycle. The corollary's hypotheses require the input to be consumed first, with only internal steps afterward. The overreach is a real flaw in the paper's practical claims, but it is easy to fix by restating the corollary's scope; it does not affect Trans(MON_m)=Trans(FST) or the acceptance classification.\n\nThe load-bearing assumption is explicit. Section I.c lists the escape routes: growing exact context, hidden-block access, writable stores, unbounded Spawn. So the classification is honest about its boundary. If a deployment has an uncapped KV cache, the finite-state result does not apply; the paper does not pretend otherwise.\n\nWho should read this: anyone working on transformer expressivity or agent-based transcript management. The collapse theorem is a useful boundary result, even though the DCFL/RE levels are classical. The references look fine — no self-citations, and the 2026 works are plausible though I can't verify them from the text.\n\nI'd send this to a serious referee. It deserves a round of review, mostly to fix the Corollary 1 over-application and to make sure the 2026 citations check out. It is not a desk reject.","headline":"The monotone collapse theorem is new and correct; the DCFL/RE hierarchy is classical and the paper says so; the only real flaw is a discussion-level overreach about tool loops.","tokens_in":20906,"tokens_out":4476,"would_cite":true,"duration_ms":46419,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["68Q45","68Q05"],"pacs":[],"model":"deepseek-v4-flash","headline":"Fixed-precision Transformers are exactly as powerful as their transcript manager allows: append-only and monotone multi-agent transcripts realize only finite-state transductions, one pop-enabled channel raises acceptance to the deterministi","keywords":["transcript management","Transformer expressivity","PopContext","finite-state collapse","deterministic context-free languages","Turing completeness","multi-agent orchestration","pushdown automata"],"falsifier":"Implement the paper's monotone protocol (Definition 7) with exactly two agents and bounded blocks, and search for a transduction that accepts a^n b^n for all n; the theorem predicts no such program exists, so any working construction would falsify the central collapse claim.","tokens_in":19892,"feed_emoji":"📜","tokens_out":9555,"duration_ms":106372,"temperature":0.7,"pith_summary":"This paper asks what one specific operation, added to the way a Transformer deployment manages its stored context (its transcript), changes about the computations the model can perform. The answer it defends: if the transcript manager can only append, route, and copy the blocks currently visible, then no matter how many agents are involved or how long the transcript grows, the whole system realizes exactly the finite-state transductions — no more than a deterministic finite automaton with output. If the manager is allowed one new primitive, PopContext, which deletes the newest block on a channel and exposes the block beneath it, a single pop-enabled channel immediately raises acceptance power to the deterministic context-free languages; two pop-enabled channels — whether in one agent or in two orchestrated agents — make the system Turing-complete. The paper's own novel theorems are the finite-state collapse and the equivalence between k orchestrated agents and one controller with k channels; the DCFL and RE levels are transferred from the classical one-stack and two-stack results through a compilation lemma.","feed_headline":"Two pop-enabled transcripts make Transformers Turing-complete","feed_subtitle":"One pop-enabled channel already reaches the deterministic context-free languages; append-only agents stay finite-state.","key_machinery":"The central object is the Transcript-Managed Transducer TMT_k (Definition 1): one finite controller with k transcript channels, each holding bounded blocks, with per-round actions stay, push, and pop, and a caller-driven status map deciding which rounds consume input. The 'pop' action is named PopContext (P_c), and it is exactly the operation the append-only layer omits. Two pieces of machinery carry the argument. Proposition 1 recodes any fixed visible radius into a single finite 'window symbol', so a bounded physical transcript becomes an abstract stack over a finite alphabet. Lemma 1 compiles this normal form into the classical top-replacement presentation of pushdown and multi-pushdown a","core_discovery":"Central claim: under fixed finite precision, bounded block sizes and visible windows, and a fixed finite agent population, the transcript-management layer around a Transformer is the whole story of its asymptotic expressivity. A pop-free transcript (RTMT_k) realizes exactly Trans(FST), the deterministic finite-state transductions, for every fixed number of channels k (Theorem 1); the same holds for any fixed finite population under a monotone protocol of append, route, and visible-copy (Theorem 2, the 'monotone multi-agent collapse'). Adding the PopContext operation to even one channel turns that channel into a stack, and via Lemma 1 the classical hierarchy transfers: one pop-enabled channel","pith_inferences":["The collapse result gives a concrete design principle the authors only hint at: if an application needs context-free behavior, the cheapest route is to add a pop capability in the transcript manager (or let the orchestrator pop an agent's transcript), rather than increasing context length or changing the architecture.","The paper's fixed, caller-driven status map is a strong premise; an editorial extension to test is whether a learned or adaptive status map — e.g., a controller that learns when to issue pop — could push a fixed-precision, pop-enabled system beyond DCFL toward RE with only one channel, since the classical one-stack bound assumes a fixed controller.","Corollary 1 (append-only CoT cycles) suggests a practical test: monitor long-running tool-use agents and instrument the exact transcript window; the theorem predicts eventual periodic repetition of states and actions once the input is consumed — a signature that could be detected in deployment logs.","The paper's boundary between finite consulted state and growing exact context implies that emerging long-context models with uncapped KV caches live in a different resource regime; for them the finite-state collapse does not apply, so extrapolating these bounds to such deployments would be a misuse of the theorem."],"forward_implications":["Any append-only or monotone multi-agent deployment with fixed precision and bounded visibility has exactly the transduction power of a finite-state machine, regardless of how long the retained transcript grows; it cannot parse general programming languages no matter how many finite-state agents are added.","Admitting PopContext on a single transcript channel is a minimal, token-protocol-preserving change that lifts acceptance from regular to deterministic context-free — enough for deterministic parsing of structured instruction languages.","Two pop-enabled channels (in one agent or two orchestrated agents) suffice for Turing completeness, so a deployed two-agent or two-channel system can, in principle, simulate any effective procedure.","The simulation overhead is small: one TMT_k round costs at most k orchestration steps, a classical pushdown transition expands to O(1) rounds, and one Turing-machine step costs O(1) two-agent steps, so the stack power is reached without changing model weights or the token protocol.","Fixed block size and visible radius are irrelevant to the classification (Proposition 1): any fixed radius packs into a finite alphabet symbol, so the hierarchy is stable under resizing those bounds."],"fun_headline_variants":["Two pops make Transformers Turing-complete","Pop once, context-free; pop twice, Turing-complete","Two pop-enabled transcripts: universal Transformers","PopContext twice, and Transformers go universal","One pop for stacks, two for full Turing power"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The load-bearing premise is that every single transition consults only a fixed finite amount of exact state — finite precision, bounded visible window, no exact access to buried blocks, and a fixed finite population — and if any deployment lets the consulted state grow with context, the finite-state collapse and the DCFL/RE classification both fail.","fun_headline_variants_meta":{"raw":{"variants":["Two pops make Transformers Turing-complete","Pop once, context-free; pop twice, Turing-complete","Two pop-enabled transcripts: universal Transformers","PopContext twice, and Transformers go universal","One pop for stacks, two for full Turing power"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000312,"raw_usage":{"total_tokens":1667,"prompt_tokens":855,"completion_tokens":812,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":599,"completion_tokens_details":{"reasoning_tokens":739}},"tokens_in":599,"tokens_out":812,"duration_ms":9382,"temperature":1.0,"reasoning_tokens":739,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-03T05:49:49.196339+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Implement the paper's monotone protocol (Definition 7) with exactly two agents and bounded blocks, and search for a transduction that accepts a^n b^n for all n; the theorem predicts no such program exists, so any working construction would falsify the central collapse claim.","supporting_citations":[],"review_version":1}