{"id":"adf3eb44-dbd9-4f24-a715-593582e063ad","arxiv_id":"2607.28068","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":6,"one_line_summary":"With a developmentally matched stimulation-naive control, repeated daily HD-MEA stimulation depresses and spatially contracts organoid evoked responses (~10% vs ~93% array engagement at day 7), while the response itself is near-synchronous, not a propagating wave.","lead":"Repeated daily electrical stimulation shrinks and weakens evoked bursts in human cortical organoids; a same-age unstimulated control still lights up nearly the whole array. The finding separates stimulation history from ordinary maturation and shows why propagation-based graph metrics often do not fit these cultures.","discovery_kind":"extension","skeptic_critique":{"model":"grok-4.5","headline":"The history-vs-maturation claim rests on one control whose day-7 hyper-responsiveness could reflect batch/placement confounds rather than stimulation history alone.","rationale":"The reader correctly locates the weakest assumption: the history-versus-maturation dissociation rests on n=1 control 612 plus two repeatedly stimulated preparations, with explicit caveats in §3.7/§4.7 and intermediate data anomalies in 613. I agree that is the load-bearing soft spot for the strongest claim; the retired BMP/propagation suite is already set aside honestly and is not the claim under test. No stronger internal inconsistency appears: sampling-rate recovery, near-zero latency–distance slope, capacity-vs-endurance split, and spatial contraction are coherently reported. The concern does not justify REJECT—the design insight and transparency are real—but it does keep the verdict at CONDITIONAL until the control contrast is replicated or 612’s exchangeability is shown. Concrete next check is another naïve day-7 control (or spontaneous baseline comparability), which directly settles whether the 93% vs 10% gap generalizes beyond one preparation.","tokens_in":14409,"tokens_out":645,"duration_ms":13981,"concrete_test":"Add ≥1 additional stimulation-naïve, same-batch, same-geometry control stimulated first at day 7 (or re-analyze any existing spontaneous day-1/day-7 activity from 612 vs 552/613 for comparable baseline rates and coverage). If the new naïve control(s) engage ≪93% (e.g., nearer the depressed ~10% range) or if 612’s pre-stim spontaneous metrics already diverge systematically from the repeated pair, the history-vs-maturation contrast weakens and the claim should stay CONDITIONAL or drop.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central claim (day-7 ~93% vs ~10% engagement is caused by stimulation history, not age) is identified with a single developmentally-matched control (612) that shares stim geometry with 613 but was never stimulated until day 7 (Abstract; §2.1; Tables 2–3; §4.1). The paper itself flags n=1 control and intermediate anomalies in 613 (§3.7, §4.7). The load-bearing soft spot is not merely small n, but that 612’s day-7 response (ratio 1914; 3788 electrodes) is the strongest in the entire dataset and could arise from unmeasured batch, plating, or coupling differences rather than naïveté. Without a second naïve control or pre-stim spontaneous baselines showing 612 is exchangeable with 552/613, the causal attribution remains a single-preparation contrast. Capacity/endurance and spatial-contraction read-outs in the repeated pair are internally consistent, but the history-vs-maturation dissociation that the authors call “load-bearing” (§4.1) is only as strong as the assumption that 612 is a typical age-matched naïve organoid.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.5","summary":"The authors propose a graph-computational framework (stimulus-conditioned functional graphs, a GNN system-identification model, a biological message-passing bound Deff ≥ dmax, and metrics EIC/RI/SSI/PR/CRI) to test whether stimulus-evoked activity in human cortical organoids reflects multi-hop propagation. Applied to longitudinal HD-MEA recordings from three organoids, the framework’s propagation premises fail: after correcting sampling rate and recovering stimulus timing, the evoked response is a near-synchronous network burst (peak-latency vs. distance slope ≈ 0), so Deff, RI, and dmax are inapplicable and per-day edge graphs are unreliable at K=10. Reframing around magnitude, synchrony, responsive-population size, and PC1 coherence, they report that repeated daily stimulation progressively depresses and spatially contracts the evoked response. A developmentally matched, stimulation-naïve control (organoid 612) at day 7 engaged ~93% of the array versus ~10% in two organoids with five prior sessions, with first-trial capacity relatively preserved while within-session endurance collapsed.","tokens_in":14831,"tokens_out":1656,"duration_ms":36034,"significance":"If the history-versus-maturation dissociation holds, the work supplies a needed control design for organoid stimulation studies that otherwise confound plasticity with development, and it cleanly dissociates preserved first-trial capacity from degrading within-session endurance and spatial contraction of the responsive pool. The transparent negative result on propagation-based graph metrics—and the explicit recommendation to verify latency–distance structure before applying Deff/RI/dmax-type measures—is a genuine methodological contribution for HD-MEA organoid work. Strengths include full execution and then principled retirement of the original metric suite (Table 4), data-driven stimulus-time recovery, and control-supported rather than over-claimed population inference. The main limit on significance is that the load-bearing causal claim rests on a single naïve control and n=2 repeated preparations.","major_comments":[{"comment":"The central claim that day-7 collapse is driven by stimulation history rather than age (§3.2–3.5, §4.1; Tables 2–3) is identified by a single developmentally matched naïve control (612: ratio 1914, n_resp=3788) against two repeatedly stimulated organoids (~206/268; 384/486). Section 4.7 correctly flags n=1 control, but the manuscript still treats this contrast as load-bearing. 612’s day-7 response is the strongest in the dataset; without a second naïve control, pre-stim spontaneous baselines establishing exchangeability of 612 with 552/613, or batch/plating covariates, atypical viability or coupling in 612 remains a viable alternative. Either add ≥1 naïve control or substantially soften causal language to a single-preparation, hypothesis-generating contrast.","section":"§3.2–3.5, Tables 2–3, §4.1, §4.7"},{"comment":"Organoid 613’s responsive trajectory is highly non-monotonic (42%→72%→14%→100%→12%; Table 3), with day 3 lowest total spikes and day-4 recordings truncated to 8–9 trials (§3.7). The paper notes these anomalies but still pools 613 with 552 for the repeated-stimulation narrative. Intermediate-day claims and any trajectory shape (rise-then-fall) should be restricted to 552 or explicitly sensitivity-tested excluding D3/D4 of 613; the shared D7 endpoint is the only robust cross-preparation result for the repeated arm.","section":"Table 3, §3.2, §3.7"},{"comment":"Theorem 1 / Eq. (2) (Deff ≥ dmax) and the GNN depth pipeline (§2.4, Eqs. 8–10) occupy substantial Methods and Introduction space, yet §3.1 and Table 4 correctly retire them because there is no measurable propagation. The paper does not report the actual Lk-vs-k curves, cross-validated Deff values, or baseline comparisons that were computed before retirement. Either move the unused GNN/BMP apparatus to Supplementary with a brief empirical null report, or show the system-identification results that justify having run the full program—otherwise the framing overpromises relative to the retained read-outs (n_resp, M, S, PC1, Σ).","section":"Theorem 1, Eq. (2), §2.4–2.5, §3.1, Table 4"},{"comment":"Within-session desynchronization is advanced as possibly history-dependent because naïve 612 showed magnitude depression without significant desynchronization while repeatedly stimulated 613 desynchronized (§3.4, §4.5). With n=1 control and mixed patterns in 552, this should remain clearly labeled exploratory; any abstract/discussion wording that pairs it with the control-validated population-size effect should be separated so the stronger n_resp/spatial-contraction result is not diluted by an underpowered secondary hypothesis.","section":"§3.4, §4.5"}],"minor_comments":[{"comment":"Abstract and §1.3 state the control engaged ~93% vs ~10%; give electrode counts alongside percentages for consistency with Tables 2–3.","section":"Abstract, §1.3"},{"comment":"Eq. (3)–(4): state whether the MAD threshold and δ=0.25Δ were fixed a priori or tuned; a one-sentence sensitivity note would help reproducibility.","section":"§2.2"},{"comment":"SSI (Eq. 14) and multi-site stimulation are discussed as future work (§4.8) after being defined in the main metric suite; a forward reference in §2.5 would reduce the sense that unused metrics are primary deliverables.","section":"§2.5, §4.8"},{"comment":"Table 2 header “response/baseline ratio” should specify the exact aggregation (session sum vs mean of per-trial Mk) to match Eqs. (17)–(20).","section":"Table 2, §2.6"},{"comment":"Minor typographical issues: missing spaces in several compounds (e.g., “stimulation-naïvecontrol”, “whereatday7”, “~93%of”); standardize throughout.","section":"Abstract, passim"},{"comment":"Cite the specific prior longitudinal stimulation studies [19, 27] with a clearer statement of what design element (naïve age-matched control; capacity vs endurance split) is new versus confirmatory.","section":"§4.1, §4.3"}],"recommendation":"major_revision","confidential_remarks":"Fit is reasonable for a methods-plus-systems journal in q-bio.NC / network neuroscience, but the title and front matter still lead with the graph framework that the data retire; after revision the paper will read more cleanly as (i) a negative methodological result on propagation metrics in organoids and (ii) a small-n but well-controlled observation of stimulation-history depression. I would not reject for the failed BMP test—the honesty is a plus—but I would not accept until the history-vs-maturation claim is either replicated in another naïve control or rewritten as provisional. No concerns about misconduct or citation manipulation."},"author_rebuttal":null,"desk_editor":{"model":"grok-4.5","letter":"The useful core is not the graph framework. They built the full BMP/GNN/metric suite, recovered true sampling rate and stimulus times, then found a near-synchronous burst with latency–distance slope ≈ 0. They retired Deff, RI, dmax and the K=10 edge graphs instead of forcing them. That negative result is clean and should change how people apply propagation metrics to organoid HD-MEA data.\n\nWhat is actually new relative to Osaki and Chow is the developmentally matched stimulation-naïve control at day 7: ~93% array engagement versus ~10% after five prior sessions, plus the capacity-versus-endurance split (first trial mostly preserved, session mean collapses) and spatial contraction with preserved PC1. Tables 2–3 make the contrast easy to see. The analysis pipeline is transparent, circularity is low, and the citation pattern is appropriate.\n\nThe soft spot is real and they name it: the history-versus-maturation claim rests on one control (612) whose day-7 response is the strongest in the dataset, plus n=2 repeated preparations, non-monotonic/truncated days in 613, and no shared code or raw data. If 612 differs in plating or coupling, the causal attribution weakens. That does not erase the within-pair depression pattern, but it caps how hard you can lean on the confound break.\n\nThis is for people doing organoid stimulation, organoid-intelligence claims, or graph metrics on culture data. Methodologists will get more from it than theorists chasing multi-hop computation. Math is light and mostly definitional; the empirical checks are the substance. I would send it to referees. It needs more controls and cleaner intermediate days, not a desk reject.","headline":"Honest negative result on propagation metrics plus a real but thin control contrast showing repeated stimulation, not age, collapses the evoked network.","tokens_in":15502,"tokens_out":446,"would_cite":true,"duration_ms":16457,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.5","headline":"Repeated daily stimulation, not age, collapses the evoked network in human cortical organoids from nearly the whole array to about 10%.","keywords":["cortical organoids","HD-MEA","stimulus-evoked dynamics","network synchrony","repeated stimulation","habituation","functional graphs","message passing"],"falsifier":"Add more developmentally matched, never-before-stimulated organoids at day 7: if they also engage only ~10% of the array, or if repeatedly stimulated organoids fail to contract relative to multiple naive controls, the history-versus-maturation dissociation fails.","tokens_in":15264,"feed_emoji":"🧠","tokens_out":998,"duration_ms":25395,"temperature":0.7,"pith_summary":"This paper set out to measure whether stimulus-evoked activity in human cortical organoids spreads as structured multi-hop computation on a graph, using functional graphs, a graph-neural dynamical model, and a biological message-passing bound on integration depth. Once sampling rate and stimulus timing were correctly recovered, the evoked response was a fast, near-synchronous network burst with no latency-versus-distance gradient, so the propagation metrics do not apply. Reframing around response size, synchrony, and shared variability, the authors show that repeated daily stimulation progressively depresses and spatially contracts the responding population. A developmentally matched organoid stimulated for the first time only at day 7 still engaged about 93% of the array, while organoids with five prior sessions engaged about 10%, separating stimulation history from ordinary maturation. First-trial capacity stayed relatively intact while within-session endurance collapsed, and coherence among remaining responders was preserved.","feed_headline":"Daily shocks shrink organoid responses from 93% to 10%","feed_subtitle":"A first-stimulation control at matched age shows the collapse is from stimulation history, not maturation","key_machinery":"The developmentally matched, stimulation-naive control (same age, first stimulated only at day 7), which breaks the confound between repeated stimulation and ordinary maturation and thereby carries the causal claim about stimulation history.","core_discovery":"Under focal electrical stimulation on HD-MEAs, human cortical organoids produce a near-synchronous network-wide burst rather than an outward-propagating wave, so propagation-based graph metrics (effective integration depth, reachability, dmax) are not applicable. Independently, repeated daily stimulation drives a progressive depression and spatial contraction of the evoked response that is attributable to stimulation history: at matched age, a first-ever stimulation engaged ~93% of electrodes versus ~10% after five prior sessions, with fresh first-trial capacity largely preserved while session endurance and responsive population size collapsed.","pith_inferences":["Protocols that use daily electrical drive as a training or entrainment signal in organoid biocomputing may be silently shrinking the usable network unless recovery intervals are built in.","The preserved coherence of the remaining day-7 core suggests a shrinking recruitable periphery rather than global desynchronization, which predicts that multi-site or weaker stimuli might re-engage dropped units more than stronger single-site drive.","A systematic inter-stimulus-interval sweep would map the recovery time constant of the fast within-session drop and test whether the slow across-day contraction shares the same recovery kinetics.","If desynchronization truly tracks prior stimulation history while magnitude depression is intrinsic, timing precision could serve as a cheap longitudinal marker of cumulative drive load."],"forward_implications":["Propagation-based graph metrics (reachability depth, Deff ≥ dmax) should be applied to organoid or culture data only after an empirical latency-versus-distance check shows real outward spread.","Longitudinal stimulation studies that stimulate every preparation cannot attribute network change to plasticity versus maturation without a stimulation-naive age-matched control.","Organoid response engineering should track first-trial capacity and within-session endurance separately, because repeated drive mainly degrades endurance and recruitable population size.","At ~10 trials, edge-level functional connectivity graphs over thousands of electrodes are not reliably thresholdable; stable claims should rest on aggregate measures such as responsive-population size and PC1 coherence.","If the depression is reversible after rest, it points to homeostatic adaptation; if durable, it points to lasting reconfiguration of the recruitable pool."],"fun_headline_variants":["Repeated stimulation shrinks organoid responses from 93% to 10%","First stimulus hits 93% of array; after five sessions only 10%","Daily focal shocks progressively depress and contract organoid bursts","Matched-age control pins response collapse on stimulation history","No outward wave: organoids give near-synchronous bursts, not propagation"],"cache_read_input_tokens":128,"weakest_assumption_plain":"The claim that stimulation history, not age, causes the collapse rests on a single stimulation-naive control organoid alongside two repeatedly stimulated ones.","fun_headline_variants_meta":{"raw":{"variants":["Repeated stimulation shrinks organoid responses from 93% to 10%","First stimulus hits 93% of array; after five sessions only 10%","Daily focal shocks progressively depress and contract organoid bursts","Matched-age control pins response collapse on stimulation history","No outward wave: organoids give near-synchronous bursts, not propagation"]},"model":"grok-4.5","effort":"low","cost_usd":0.005281,"raw_usage":{"total_tokens":1558,"prompt_tokens":908,"num_sources_used":0,"completion_tokens":74,"cost_in_usd_ticks":52808000,"prompt_tokens_details":{"text_tokens":908,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":576,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":908,"tokens_out":74,"duration_ms":9376,"temperature":1.0,"reasoning_tokens":576,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-31T18:49:22.738405+00:00","model_set":{"reader":"grok-4.5"},"falsifier":"Add more developmentally matched, never-before-stimulated organoids at day 7: if they also engage only ~10% of the array, or if repeatedly stimulated organoids fail to contract relative to multiple naive controls, the history-versus-maturation dissociation fails.","supporting_citations":[],"review_version":1}