{"id":"338b82fb-7570-4c48-891a-b6984352c966","arxiv_id":"2607.08252","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.5,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"Separating controlled divergence from evidence-governed absorption reduces persona-environment self-locking, cutting macro-theme repetition from 61.8% to 36.3% in a same-runtime 40-day A/B.","lead":"Long-running LLM persona loops collapse into familiar places, weak relationships, and deferred decisions even when daily text looks varied. AutoPersonas measures that self-locking across eight models and shows that masking context plus targeted divergence can cut theme repetition roughly in half while keeping identity continuous.","discovery_kind":"new_method","skeptic_critique":{"model":"grok-4.5","headline":"The A/B gain is measured on generation-side theme diversity, not on OSO absorption or identity-preserving State/reachability change.","rationale":"The reader correctly flags single-canon/single-run metrics and proxy adequacy, and the CONDITIONAL verdict is right. The sharper load-bearing issue is not only that macro-theme/action-category proxies are imperfect, but that they sit at the wrong causal layer for the claim as stated: they quantify generation-side anti-fixation, while the architecture’s distinctive claim is evidence-governed absorption that moves State and reachability without identity break. Section 7.2 reports void-conversion and target-validation checks, not OSO propagation or identity audits under the A/B. The three-year diagnostic and post-architecture qualitative cases address those dimensions, but they are not the same controlled comparison that carries the quantitative claim. This does not justify REJECT: claims are carefully bounded, the eight-model baseline is solid, and the A/B still supports a real mechanism gain on theme fixation. It does keep the verdict CONDITIONAL and slightly tightens the condition—accept-shaped only if mitigation is re-measured with hardening and continuity metrics, not theme diversity alone. Agreement with the reader is partial: same weak-proxy neighborhood, different precise failure point (metric layer vs OSO claim, not only variance/multi-persona).","tokens_in":29017,"tokens_out":689,"duration_ms":7552,"concrete_test":"On the same pre-redesign vs redesigned 40-day A/B logs, score each event for (i) hardening (did it produce Observation + State or reachability change within 7 simulated days?), (ii) identity continuity (semantic State-dimension drift vs ignition canon), and (iii) environment-shell reuse. If redesigned theme gains are not accompanied by higher hardening rates and stable identity scores, the central claim is overstated relative to the measured intervention.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The strongest claim is that separating controlled divergence from evidence-governed absorption reduces persona-environment self-locking while preserving identity continuity, with the same-runtime 40-day A/B (macro-theme repetition 61.8%→36.3%, themes 55→102) as the main quantitative support. That A/B, and the isolated-generator ablation, primarily measure cumulative macro-theme count and reuse under a reviewer reuse rule on generated events (Section 7.2). They do not report the paper’s own causal criteria for anti-self-locking: Occurrence→Observation→State revision→changed future possibility space (Sections 3.6, 4.5), nor identity-continuity checks under the redesigned conditions. The paper itself treats occurrence hardening, environment watermark shells, recursive indecision, and relationship persistence as core failure modes (Sections 6–7.3), yet the redesigned A/B does not score those dimensions. Masking + per-sample targeting can therefore raise theme variety without proving that novelty hardens into State/reachability movement rather than decorative events, and without showing that identity continuity is preserved rather than merely asserted. The eight-model baseline is strong for mode-lock; the load-bearing gap is that the mitigation metric is upstream of the OSO claim.","agreement_with_reader":"partial"},"referee_report":{"model":"grok-4.5","summary":"The paper identifies self-locking as a runtime failure in long-term persona-life loops: locally plausible events continue while trajectories collapse into familiar environments, weak relationships, deferred decisions, and stale stages. It attributes this to model-side channel convergence and system-side context gravity, and proposes AutoPersonas—an OSO (Occurrence–Observation–State) multi-timescale life-environment engine that separates controlled divergence from evidence-governed absorption. Evaluation is diagnostic rather than benchmark-competitive: a three-year compressed simulation surfaces failure modes; an eight-model 40-day direct-loop stress test (1,600 events) reports 95.2%–97.6% mean rolling 5-day action-category repetition and 79.0%–88.0% macro-theme repetition; a same-runtime A/B with context-slice masking plus per-sample divergence targeting reduces macro-theme repetition from 61.8% to 36.3% and roughly doubles theme count; a juvenile-goblin fictional-world run reproduces the anti-fixation regime. The bounded claim is that separating controlled divergence from evidence-governed absorption can reduce persona-environment self-locking while preserving identity continuity.","tokens_in":29437,"tokens_out":1641,"duration_ms":17681,"significance":"If the result holds, the paper supplies a useful systems framing for open-ended persona agents distinct from task agents and closed sandbox societies, plus a concrete, reproducible demonstration that direct recursive persona loops rapidly mode-lock across current foundation models. Strengths include the eight-model stress test, temperature probe, semantic re-keeping, same-runtime A/B (better than cross-setting anecdote), public-safe aggregate artifacts, and an action-repetition evaluator. The life-environment / OSO separation is a clear architectural contribution for reward-free open evolution under context gravity. The main significance risk is that the strongest quantitative mitigation evidence measures generation-side theme diversity rather than the paper’s own causal criteria for anti-self-locking (hardening into State and reachability change with identity continuity).","major_comments":[{"comment":"§7.2 and Abstract: the load-bearing claim is that separating controlled divergence from evidence-governed absorption reduces self-locking while preserving identity continuity. The same-runtime A/B and isolated-generator ablation report cumulative macro-theme count and repeat ratio under a reviewer reuse rule on generated events (61.8%→36.3%, 55→102 themes). They do not score the paper’s own OSO criteria in §§3.6 and 4.5—Occurrence→Observation→State revision→changed future possibility space—nor identity-continuity checks under redesigned conditions. Masking + targeting can raise theme variety without proving novelty hardens rather than remaining decorative (the occurrence-hardening gap of §7.3). Please report, for the A/B arms, at least proxy rates of State/reachability movement, hardening success/failure, and a continuity audit (e.g., canon-consistency or stale-state regression), or narr","section":"§7.2 Quantitative anti-fixation validation"},{"comment":"§7.2 and §14: the redesigned A/B and isolated ablation use one primary complex persona canon with one run per condition; the paper correctly flags missing variance bands and multi-persona replication. A single-run effect size this large is suggestive but not yet adequate for the systems claim as stated. At minimum, report 2–3 independent seeds or a second real-world canon under the same A/B protocol, with uncertainty on repeat ratio and theme count, or restate results as a single-trajectory mechanism demonstration.","section":"§7.2 / §14 Limitations"},{"comment":"§6.1 and §7.1: action-channel metrics rely on a rule-based keyword taxonomy over title/action fields (with Other residual), and theme metrics on a semantic keeper under a reuse rule. Stricter variants still show early convergence, which is reassuring for the baseline, but the mitigation A/B is only reported in the theme vocabulary. Please (i) apply the same action-channel metric to the redesigned full-runtime A/B arms, and (ii) state inter-rater or stability checks for the macro-theme grouping, so the anti-fixation gain is not keeper-specific.","section":"§6.1 Experimental setup; §7.1"},{"comment":"§4 and §7.2: AutoPersonas is defined as a multi-mechanism engine (conditional variation, context governance, information orthogonality, progressive causal propagation, trajectory monitoring, multi-timescale revision). The quantitative intervention tests only context-slice masking plus per-sample divergence targeting (and a narrative-arc targeting-only lane). That is a valid mechanism A/B for divergence under context gravity, but it does not validate the full OSO absorption stack. Either add at least one absorption-side ablation (e.g., hardening/review disabled vs enabled with divergence fixed) or explicitly bound the claim to the divergence/governance pair rather than the full engine.","section":"§4 AutoPersonas; §7.2 Stage 2"}],"minor_comments":[{"comment":"Abstract vs §7.1: abstract says all models crossed 90% by day 11; body also states all crossed 80% by day 9. Keep thresholds consistent in the abstract.","section":"Abstract"},{"comment":"§1 mentions a ~60% history/state token-budget threshold as a design finding from researcher judgment. Flag more clearly as non-benchmarked and avoid implying a universal law (the paper already hedges; a single sentence in Limitations would help).","section":"§1 Introduction"},{"comment":"Figures 3–6 are public-safe storyboards; ensure captions state that panels are illustrative compressions, not raw logs, so readers do not over-read them as quantitative evidence.","section":"§7.4–7.6"},{"comment":"Related work on Generative Agents / Agentopia is appropriately complementary; a short explicit non-claim that AutoPersonas is not evaluated on society-level believability metrics would reduce misreading as a sandbox competitor.","section":"§8 Comparison"},{"comment":"Terminology density (self-locking, watermark shell, occurrence hardening, current-state authority) is high; a one-page glossary early (partially present in §3.4) would help non-systems readers.","section":"§3.4"},{"comment":"Kimi temperature forced to 1.0 and mixed OpenRouter/Moonshot route is disclosed; consider a footnote in the main results table so readers do not treat that row as fully protocol-matched.","section":"§7.1 Table"}],"recommendation":"major_revision","confidential_remarks":"The eight-model mode-lock result is the most publishable piece and is largely independent of the full OSO story; if the authors cannot close the absorption/identity measurement gap in revision, the paper can still stand as a strong empirical systems note on recursive persona mode-lock plus a partial mitigation, but the current abstract over-reaches. Scope fit for a serious AI systems venue is reasonable if claims are tightened; commercial disclosure boundaries are acceptable given ancillary aggregates and evaluator code, but reviewers may still press for more absorption metrics that do not require private recipes."},"author_rebuttal":null,"desk_editor":{"model":"grok-4.5","letter":"The thing worth knowing is that this paper makes self-locking concrete and measurable. Across eight models and 1,600 events, direct persona loops slam into ~95–98% rolling action-category repetition by day 11, and semantic re-keeping still shows 79–88% macro-theme reuse. A temperature probe does not open the repertoire. That baseline is the real contribution, and it is cleaner than most long-horizon agent anecdotes.\n\nWhat is new relative to Generative Agents / Agentopia is the object: open persona–life-environment co-evolution under context gravity, not closed sandbox society simulation with reward. The OSO split (Occurrence / Observation / State), the life-environment layer, and the diagnostic taxonomy (watermark shells, hardening gaps, recursive indecision) are useful systems language. The same-runtime A/B is better than a cross-setting story: masking plus per-sample divergence targeting cuts macro-theme repetition from 61.8% to 36.3% and roughly doubles theme count. The goblin run is a decent generality check, not decoration. Claims are carefully bounded, and the citation pattern is fair.\n\nThe soft spot is real but proportional. The load-bearing mitigation metric is generation-side theme diversity under a reuse rule on one primary canon with one run per redesigned condition. The paper’s own causal criteria for anti-self-locking are Occurrence → Observation → State revision → changed reachability, plus identity continuity. The A/B does not score those dimensions; it scores upstream variety. So the strongest claim slightly outruns the measured object. Production internals are withheld, variance bands and multi-persona replication are missing, and relationship persistence remains qualitative. That is a gap, not a collapse of the argument.\n\nThis is for people building persistent companions and open-ended agent loops, not for pure theory or closed-world society simulators. The failure evidence is solid enough that I would bring it to reading group and cite the mode-lock numbers. A serious editor should send it to referees; they will demand multi-run stats and absorption metrics, but the paper already earns that time.","headline":"Strong multi-model evidence for persona mode-lock, plus a real same-runtime diversity gain; the soft spot is that the A/B measures theme variety more than OSO absorption or identity continuity.","tokens_in":29999,"tokens_out":525,"would_cite":true,"duration_ms":6158,"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":"Separating controlled divergence from evidence-governed absorption can reduce persona-environment self-locking while preserving identity continuity.","keywords":["persona agents","self-locking","life-environment","OSO loop","open-ended evolution","context gravity","diversity collapse","multi-timescale simulation"],"falsifier":"A multi-persona, multi-run full-runtime A/B in which enabling context-slice masking and per-sample divergence targeting fails to cut macro-theme repetition or raise cumulative themes relative to the same-runtime baseline, or raises theme counts without corresponding diagnostic movement in State, relationship function, occurrence hardening, or later reachability.","tokens_in":29882,"feed_emoji":"🎭","tokens_out":1193,"duration_ms":25269,"temperature":0.7,"pith_summary":"Long-term persona agents are meant to keep living—entering new situations, forming relationships, revising concerns—while remaining recognizable. This paper identifies self-locking: a runtime failure in which locally plausible events keep appearing while the generated life collapses into familiar environments, weak relationships, suspended decisions, and stale life stages. The authors trace the failure to two coupled pressures: model convergence onto high-probability behavioral channels, and system-level context gravity from State, memory, history, and environment summaries. AutoPersonas is a multi-timescale life-environment engine whose OSO loop separates environment-side Occurrences, accumulated Observations, and persona State, so divergent future-facing material can enter while only evidence-governed absorption may revise State or reachability. Across eight models, direct recursive loops showed extreme action- and theme-channel repetition; a same-runtime A/B found that context-slice masking plus per-sample divergence targeting cut macro-theme repetition from 61.8% to 36.3% and roughly doubled cumulative themes. Readers who build continuing companions should care because memory alone does not solve this, and uncontrolled novelty breaks identity.","feed_headline":"Divergence control cuts persona self-lock from 62% to 36%","feed_subtitle":"Eight models collapsed into repeated life themes; separating divergence from absorption roughly doubled theme variety.","key_machinery":"The OSO loop (Occurrence → Observation → State revision → future possibility space): a multi-timescale life-environment engine that keeps environment-side future-facing material, accumulated evidence, and continuity-bearing State as separate authorities. Controlled divergence (conditional variation with context-slice masking and per-sample targeting) opens plausible non-identical signals; evidence-governed absorption decides whether they harden into State or reachability. Without the divergence source the loop self-locks around old State; without OSO governance, novelty becomes drift.","core_discovery":"Self-locking is a distinct runtime failure of continuing persona-life loops, not merely forgetting or low local text quality: the loop can produce varied prose while functionally returning to the same places, unresolved routes, relationship roles, and life stage. Direct self-orchestrated loops across eight models generated 1,600 events with mean rolling 5-day action-category repetition of 95.2%–97.6% (all models past 90% by day 11) and 79.0%–88.0% macro-theme repetition. The paper’s bounded systems claim is that separating controlled divergence from evidence-governed absorption—via an OSO architecture plus context-slice masking and per-sample divergence targeting—can reverse that fixation: i","pith_inferences":["If self-locking is mainly a context-authority problem, long-horizon agent progress may come more from runtime governance than from larger base models alone.","The same separation of controlled divergence from evidence-governed absorption may transfer to other recursive generation settings where summaries re-enter generation, such as long-running story engines.","Because relationship persistence remained a hard failure even when theme variety improved, product systems that only measure event novelty may still be shipping decorative lives that never change obligations or reachability.","A natural next measurement is multi-persona variance bands that track whether route-authority and relationship-function changes move with theme counts, not only whether new themes appear."],"forward_implications":["Direct recursive persona loops without life-environment architecture will rapidly close over a small action and life-theme repertoire across current foundation models.","Memory expansion and higher sampling temperature alone will not break self-locking; divergence must be architecturally separated from State absorption.","Day-level simulation is a phase boundary for audits: coarser weekly or yearly summaries can hide deferred decisions, decorative opportunities, and environment watermark shells.","Deployed companions with independent lives require dual-stream recall—persona self-life versus user-specific relationship memory—rather than user-only personalization.","Sandbox society simulators and open-environment persona engines solve different authority problems and should not be treated as direct benchmark substitutes."],"fun_headline_variants":["OSO loop cuts persona theme lock from 62% to 36%","Controlled divergence halves self-lock in eight-model tests","Persona loops hit 95% action repeat; masking doubles themes","Separating divergence from absorption drops lock to 36%","Context-slice masking reduces fixed themes 62% to 36%"],"cache_read_input_tokens":16512,"weakest_assumption_plain":"That action-category and macro-theme repetition on one primary complex persona canon, with one run per redesigned condition, are adequate proxies for whether a persona’s functional life trajectory is actually diversifying rather than only looking varied in labels.","fun_headline_variants_meta":{"raw":{"variants":["OSO loop cuts persona theme lock from 62% to 36%","Controlled divergence halves self-lock in eight-model tests","Persona loops hit 95% action repeat; masking doubles themes","Separating divergence from absorption drops lock to 36%","Context-slice masking reduces fixed themes 62% to 36%"]},"model":"grok-4.5","effort":"low","cost_usd":0.006148,"raw_usage":{"total_tokens":1736,"prompt_tokens":971,"num_sources_used":0,"completion_tokens":89,"cost_in_usd_ticks":61480000,"prompt_tokens_details":{"text_tokens":971,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":676,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":971,"tokens_out":89,"duration_ms":6211,"temperature":1.0,"reasoning_tokens":676,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-10T10:34:20.632010+00:00","model_set":{"reader":"grok-4.5"},"falsifier":"A multi-persona, multi-run full-runtime A/B in which enabling context-slice masking and per-sample divergence targeting fails to cut macro-theme repetition or raise cumulative themes relative to the same-runtime baseline, or raises theme counts without corresponding diagnostic movement in State, relationship function, occurrence hardening, or later reachability.","supporting_citations":[],"review_version":1}