{"id":"02271f4b-fc27-478a-ad72-ba01aef67871","arxiv_id":"2505.11509","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"The paper defines information-value measures as changes in observer-specific state values across feedback cycles and illustrates them on four multi-scale system models.","lead":"Researchers propose a toolkit of syntactic, semantic, and pragmatic measures for evaluating how much information flows matter in multi-scale feedback systems. They demonstrate it on four simulations: swarms, collective decisions, task distribution, and coupled oscillators.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Value measures are defined as temporal changes in a scalar state value, not as comparisons with and without the flow promised in §3.2; no baseline isolates the information flow's causal contribution.","rationale":"The reader's weakest assumption is that the semantic and pragmatic measures are arbitrary because SV is freely chosen. I regard that as a real, related issue, but the more directly load-bearing problem is the step before SV choice: even with a fixed SV, Eqs. (1)–(3) do not implement the with/without comparison that Section 3.2 promises. The arbitrary-SV concern presumes that a temporal difference, once SV is fixed, would be a value of the flow; the counterfactual issue attacks that prior step. If the measure cannot attribute a change to the information flow, a better SV cannot fix it. This internal mismatch is present before all four case studies and determines what the reported numbers can mean. I propose a controlled-content permutation test on the task distribution model because that model is analytic and small, making the test cheap and decisive. If the test shows the measure is sensitive to the control content, the paper's remaining weaknesses (SV choice, HO normalization, table typos) are fixable and the conditional verdict stands. If not, the central claim about valuing information flows would need a redefinition of the measures around explicit counterfactuals. The reader's verdict of CONDITIONAL remains appropriate: the concern is serious but addressable in revision, and the paper is a framework proposal rather than a formal theorem. Hence I would keep the conditional verdict rather than move to ACCEPT or REJECT.","tokens_in":29363,"tokens_out":13618,"duration_ms":151453,"concrete_test":"Use the task distribution model (§4.4), which is small enough for exact analysis. Recompute the transition matrix used for BB in Exall under two conditions: (A) the actual control error v^delta_{S1} computed from the workers' states, as in the paper; (B) the same BB managers, workers, and switching rule, but at each step v^delta_{S1} is drawn independently from its stationary marginal distribution rather than being the current state-dependent error, so the control channel exists and has the same syntactic resources but carries no state-dependent semantic content. Derive V_pr,gl and E_pr,gl from Eq. 43 and Fig. 11 for both conditions. If condition (B) yields V_pr,gl close to condition (A) — within the differences that separate strategies in Fig. 11b — then Eqs. (1)–(3) are not measuring the value of the information flow; they are measuring generic feedback-driven adaptation.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim requires that the semantic and pragmatic measures quantify the value of a specific information flow I_{t−θ→t}. Section 3.2 explicitly grounds value in a contrast: 'value is measured by comparing how the system performs towards a goal state with and without information' (Gould, 1974). But the operational equations (1)–(3) compute only SV^t − SV^{t−θ} for an analyst-chosen scalar SV. This is a temporal difference, not a with/without comparison. Nothing in the measure removes or holds fixed the rest of the system, so a positive V_pr,gl can be produced by the system's own dynamics, initialization, delays, or another feedback cycle, and is not attributable to I_{t−θ→t}. The case studies compare strategies (e.g., BB vs RS, consensus vs randomCN), but these differ in algorithm, memory, timing, and information content simultaneously; they are not controlled ablations of a single flow. Even if one grants a state-value function, the measured 'value' conflates the effect of the flow with unrelated state evolution. The same issue affects the efficiency ratios, since their numerator inherits the attribution problem. This is not a philosophical objection to observer-relative value; it is an internal mismatch between the stated definition and the equations, and it threatens the central claim that information flows, rather than whole configurations, are what is valued.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a conceptual framework for measuring syntactic, semantic, and pragmatic information in Multi-Scale Feedback Systems (MSFS). It defines three value-oriented measures—semantic truth value Vsm,th, semantic goal value Vsm,gl, and pragmatic goal value Vpr,gl—as changes in analyst-assigned state-value functions over an interval [t−θ,t], together with descriptive deltas and efficiency ratios. The framework is illustrated through four case studies: a robotic collective, collective decision-making, task distribution, and hierarchical oscillators, where different strategies, parameter settings, or hierarchy depths are compared. The central claim is that these measures allow analysts to quantify the value of information flows in any complex adaptive system with identifiable goals and to reason about feedback architectures. The paper is not presented as a theorem-based result; it is a proposal with worked examples and a general recipe for measurement.","tokens_in":29740,"tokens_out":6137,"duration_ms":65814,"significance":"If the proposed measures were well-founded, the paper would provide a useful vocabulary and a practical recipe for quantifying information value in engineered and natural adaptive systems. The four case studies span different feedback structures and include explicit equations, which is a strength, as is the paper's transparency about the observer-dependence of value. However, as written, the value measures do not isolate the contribution of the information flow: they are temporal differences of scalar state-value functions rather than comparisons with and without the flow, and the scalar functions are assigned rather than derived. The reported numbers are therefore best described as descriptive traces of state-value change, not as values of the information flow I_{t−θ→t}. This is a fixable problem, but it is load-bearing for the paper's central claim.","major_comments":[{"comment":"The text states that value is measured by comparing how the system performs towards a goal state with and without information (Gould, 1974), but the operational equations compute only temporal differences: Vsm,th = SVth(t) − SVth(t−θ), Vsm,gl = SVopt(t) − SVopt(t−θ), and Vpr,gl = SVgl(t) − SVgl(t−θ). No baseline removes or holds fixed the rest of the system, so a positive Vpr,gl can be produced by initialization, system dynamics, delays, or a different feedback cycle, and is not attributable to the flow I_{t−θ→t}. The case-study comparisons (e.g., main vs. ground truth vs. random in §4.2.1, consensus vs. randomCN in §4.3.2) change algorithms, memory, timing, and information content simultaneously, so they are not controlled ablations of a single flow. Since the efficiency ratios inherit this attribution problem, the central claim that the measures quantify the value of information flows is not currently supported. A controlled with/without comparison or an explicit causal model of the flow is needed.","section":"§3.2, Eqs. (1)–(3)"},{"comment":"The scalar state-value functions are assigned rather than derived, and the paper provides no construction principle or sensitivity analysis for them. In the collective decision-making case, Eq. (24) defines Δgl = 1 − 2|0.5 − WA|, but the system collapses at WA = 0.10 and WA = 0.90, where this expression equals 0.2 rather than 0; the goal-value function therefore does not assign the terminal states the minimum value implied by the stated goal. In the task-distribution case, SVgl is set to {0, 0.25, 0.5, 0.75, 1} directly, and the numerical values of Vpr,gl (e.g., BB ≈ 0.72 at m = 10, RS converging to 0.33) are consequences of these arbitrary assignments. Unless the framework specifies how SV is obtained from a utility or loss function, or shows that conclusions are robust across reasonable SV choices, the reported semantic and pragmatic values are not generalizable measures.","section":"§4.3.4.1.3, Eq. (24); §4.4.4.2, Eqs. (36)–(43)"},{"comment":"The syntactic measure changes meaning across the case studies: memory units in the robotic collective, Shannon entropy in collective decision-making and task distribution, and a JS-divergence dissimilarity in the hierarchical oscillator case. In Eq. (50), Csynm,i = 1 − D_JS(fm,i||U)/log(2) assigns the maximum value to the uniform distribution and measures distance from uniformity, not message size, resource use, or information transmitted through the feedback cycle. Consequently, the claim that the 3-scale system contains more information than the 2-scale system (Csyn = 6.6459 vs. 4.7917) relies on summing dissimilarity-to-uniform scores and is not commensurable with the syntactic measures used elsewhere. The paper should either justify a common interpretation of Csyn across cases or restrict efficiency comparisons to within-case claims.","section":"§4.5.3.1.1, Eqs. (50)–(51)"}],"minor_comments":[{"comment":"The caption reads 'Green arrows show abstracted flows of state information, green arrows flows of control information'; the second color should presumably be different (e.g., blue), and the legend should be clarified.","section":"Figure 2 caption"},{"comment":"Several references preserve LaTeX/encoding artifacts, such as 'V on Uexk¨ull' and 'Weizs¨acker'; please correct these to the actual author names.","section":"References"},{"comment":"The semantic-information block of Table 1 contains the column header 't2)t2', which appears to be a typo for 't2→t3'; please check all headers.","section":"Table 1"},{"comment":"The piecewise definition of Vsm,th uses 'SVkn' with inconsistent subscripts and an 'NA' condition without defining the full notation; in particular, the clause 'SVkn(...)==NA ∨ SVkn(...)==NA' should be written with explicit expressions for both time arguments and a clear definition of NA.","section":"§4.4.4.2, Eq. (37)"},{"comment":"The efficiency normalization uses an arbitrary scaling factor coef = 10 with no justification; please state how coef is chosen and how it affects comparisons between strategies.","section":"§4.4.4.2, Eq. (38)"},{"comment":"Several performance claims (e.g., that the short-memory strategy outperforms the long-memory strategy for the full sub-model, or that parameter-region differences are due to noise) are qualitative readings of pooled figures without error bars, confidence intervals, or statistical tests; adding quantitative support would strengthen the case-study conclusions.","section":"§4.2.2.2 and §4.3.3"},{"comment":"No data or code repository is provided for the simulation results; for a paper whose main evidence is computational case studies, a reproducibility statement or availability link should be added.","section":"Throughout"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is a useful conceptual contribution in a domain where formal measures of semantic and pragmatic information are rare, but the internal gap between the stated with/without definition and the implemented temporal-difference equations is substantial. I would support publication if the authors either implement a genuine causal comparison (e.g., blocking or counterfactual baselines) or explicitly reframe the paper as proposing descriptive state-value-change indicators rather than values of information flows. The arbitrary SV assignments also need a construction principle or sensitivity analysis. The case studies are constructive and have the potential to be persuasive after these revisions."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Read this as a framework paper, not a results paper. The new thing is the packaged suite of syntactic, semantic, and pragmatic measures — semantic delta, semantic truth and goal values, pragmatic delta and goal values, efficiency ratios — applied to four different models (robotic collective, collective decision-making, task distribution, hierarchical oscillators). The case studies are genuinely diverse, and the authors engage the right literature: Gould, Weinberger, Frank, Kolchinsky, and the older semantic/pragmatic information discussions. The semantic-versus-pragmatic split (knowledge vs action) is cleanly stated, and the efficiency measures draw the resource dimension back in.\n\nThe weak spot is exactly where the reader and stress-test put it. Section 3.2 says value is measured by comparing how the system performs toward a goal with and without information, but the operational equations (1)–(3) compute only SV(t) − SV(t−θ) for an analyst-chosen scalar state-value function. That is a temporal difference, not an attribution of change to the flow I_{t−θ→t}. A positive pragmatic goal value can be produced by the system's own dynamics, initialization, or delays. The strategy comparisons in the case studies (BB vs RS, consensus vs randomCN, 2-scale vs 3-scale) vary algorithm, memory, timing, and information content at the same time, so they are not controlled ablations of a single flow. If the measures are read as descriptive indicators of state change, the framework is coherent; if they are read as measuring the value of a specific information flow, the central claim is not yet supported. The hand-assigned SV functions (for instance, {0, 0.25, 0.5, 0.75, 1} in task distribution) are a second structural gap: without a rule for constructing SV, the resulting values are conventions. That is acceptable in a proposal, but it needs to be stated more prominently.\n\nTwo smaller things. The hierarchical oscillator definition Csyn = 1 − D_JS(f||U)/log(2) is nonstandard and it drives the 2-scale vs 3-scale efficiency comparison; it needs a real justification. And the paper gives no code or key parameter values (F_m, tau_m, W), with the supplement not available, so the numeric side is not independently checkable. Table 1 also has a garbled header. None of these sink the paper, but they keep confidence moderate.\n\nI would not desk-reject this. It deserves a serious referee, with the expectation of major revision: either bring the definitions in line with a with/without baseline, or change the claims to describe temporal changes in state value; give guidance on constructing SV; and make the case-study parameters and code available. The likely audience is system designers and researchers in self-adaptive and collective systems who need a vocabulary for information value. If I were working on that, I would read it and probably argue with it, but I would not cite the numbers yet.","headline":"A useful vocabulary for information value in multi-scale feedback systems, but the flow-attribution claim is not supported by the temporal-difference equations.","tokens_in":30192,"tokens_out":4692,"would_cite":false,"duration_ms":46642,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper sets out to establish that the value of an information flow in a Multi-Scale Feedback System can be measured as the difference it makes to a goal-scored state value, and supplies concrete formulas and four case studies.","keywords":["adaptation","syntactic information","semantic information","pragmatic information","complexity","multi-scale feedback systems","value of information","complex adaptive systems"],"falsifier":"Take one recorded simulation of the robotic collective and compute $V_{\\mathrm{pr,gl}}$ for the same information flow under two different state-value functions: one that scores the robot distribution by absolute error from the desired counts and one that scores it by object-weighted error. If the two state-value functions reverse the ranking of the main and ground-truth strategies, the pragmatic measure is an artifact of the analyst's choice rather than a property of the information flow, and the paper's central claim that these measures quantify the value of information flows would not be well-defined.","tokens_in":29127,"feed_emoji":"🔄","tokens_out":11657,"duration_ms":96347,"temperature":0.7,"pith_summary":"This paper sets out to show that information in a complex adaptive system has a measurable value, not just a measurable size, and that this value can be computed by comparing the system's state before and after a feedback cycle. The authors define Multi-Scale Feedback Systems as collections of agents that abstract local information upward and receive control information downward, then propose three families of measures: syntactic (how much information is carried and at what resource cost), semantic (how much the flow changes an agent's knowledge and how close that knowledge is to truth or to optimal knowledge), and pragmatic (how much the flow changes action and how close that action brings the system to its goal). The contribution is a general recipe plus worked examples from four models: a robotic collective, a collective decision-making task, a task-distribution hierarchy, and a set of coupled oscillators. If the recipe works, an analyst can rank feedback architectures by how effectively they use information, which matters for designing self-adaptive systems and for understanding how information drives adaptation in natural ones.","feed_headline":"Score information's value by the goal-state change it causes","feed_subtitle":"Syntactic measures say how big a message is; semantic and pragmatic measures say what it does to knowledge and action.","key_machinery":"The load-bearing object is the feedback cycle of a Multi-Scale Feedback System, treated as the unit of comparison: state information is abstracted from micro-scale to macro-scale, processed, and reified downward as control information, and the analyst evaluates the system at $t-\\theta$ and at $t$. The measure that carries the argument is the analyst-assigned state-value function $SV$, defined for knowledge states and action states with respect to a ground truth, an optimal knowledge state, or a goal; the paper's three value formulas are all differences of $SV$ across the cycle. The descriptive deltas and efficiency ratios do auxiliary work: deltas isolate the magnitude of change, and efficiencies divide values by $C_{\\mathrm{syn}}$ to connect information value to the physical resources it costs.","core_discovery":"The central claim is that the value of an information flow in a Multi-Scale Feedback System is the difference it makes to a scalar state-value function $SV$ that scores how close the system's knowledge or action state is to a reference state. The paper defines the semantic truth value as $V^{t-\\theta\\to t}_{\\mathrm{sm,th}} = SV^{t}_{\\mathrm{th}} - SV^{t-\\theta}_{\\mathrm{th}}$, the semantic goal value as $V^{t-\\theta\\to t}_{\\mathrm{sm,gl}} = SV^{t}_{\\mathrm{opt}} - SV^{t-\\theta}_{\\mathrm{opt}}$, and the pragmatic goal value as $V^{t-\\theta\\to t}_{\\mathrm{pr,gl}} = SV^{t}_{\\mathrm{gl}} - SV^{t-\\theta}_{\\mathrm{gl}}$, where $t-\\theta$ is the previous comparison time and $I_{t-\\theta\\to t}$ is the information that flowed in between. Descriptive deltas $\\Delta_{\\mathrm{sm}}$ and $\\Delta_{\\mathrm{pr}}$ measure the magnitude of change in knowledge and action regardless of goal, and efficiencies divide each value by the syntactic content $C_{\\mathrm{syn}}$. Across the four case studies, the measures separate accurate-but-slow knowledge from fast-but-inaccurate knowledge, show when extra hierarchy costs efficiency, and track how delays change the value of otherwise identical information.","pith_inferences":["The before/after comparison is not inherently multi-scale: any feedback loop can be read as a two-scale system, so the same value measures could rank single-loop controllers, a testable extension the paper does not pursue.","Sweeping the sensing horizon in the robotic collective would map where precision begins to hurt, turning the paper's observed trade-off between accuracy and delay into a quantitative design curve.","If the state-value function $SV$ were tied to survival probability, the measures would connect to viability-based semantic information and could be applied to biological systems whose goals are not supplied by an engineer.","Because $SV$ is a free choice, two analysts with different goals will necessarily disagree about whether the same flow is valuable; the framework thus makes the choice of goal an explicit empirical variable rather than a hidden one."],"forward_implications":["Feedback strategies can be compared by resource efficiency rather than by message size: the short-memory robotic collective matches or beats the long-memory one over long cycles at a fraction of the syntactic cost.","Semantic and pragmatic measures can disagree, and the disagreement is informative: in collective decision-making, a strategy that keeps the environment alive does so through fast oscillation rather than accurate perception, so survival and knowledge accuracy separate.","Delays become a tunable quantity: the four models show that delaying feedback can either compensate for inaccurate adaptation or cause overshoot, so designers can adjust cycle length rather than only improving sensing.","Accuracy is not always an asset: the ground-truth robotics strategy underperforms estimation strategies, implying that the value of information depends on how quickly the system can act on it, not only on how correct it is.","The same information flow can have different values at different scales, so ranking a strategy requires stating the scale and the goal first."],"supporting_citations":[{"why":"Supplies the entropy measure used for syntactic information content and resource accounting.","marker":"Shannon (1948)"},{"why":"Introduces the Multi-Scale Feedback Systems concept that this paper extends with information measures.","marker":"Diaconescu et al. (2019)"},{"why":"Provides the earlier multi-scale feedback formalization and the task-distribution model simplified in this paper.","marker":"Diaconescu et al. (2021a)"},{"why":"Extends coupled oscillators to a hierarchy and supplies the hierarchical oscillator case study and its delay analysis.","marker":"Mellodge et al. (2021)"},{"why":"Provides the robotic collective model and response-threshold-based goal selection used in the first case study.","marker":"Zahadat (2023)"},{"why":"Supplies the differential-equation model of coupled biochemical oscillators underlying the oscillator case.","marker":"Kim et al. (2010)"},{"why":"Establishes the viability-based approach to semantic information value that the paper's goal-relative measures build on.","marker":"Kolchinsky and Wolpert (2018)"},{"why":"Defines pragmatic information with respect to an action, the contrast point for the paper's efficiency-based extension.","marker":"Frank (2003)"},{"why":"Provides the collective decision-making model of agents estimating task sizes used in the second case study.","marker":"Di Felice and Zahadat (2022)"}],"fun_headline_variants":["Info value = change in goal-state score","How much a message moves you toward a goal","Goal-state delta: the true worth of information","Pragmatic info metrics: value by goal progress","Beyond bits: measuring info's goal impact"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the analyst can assign a scalar state-value function $SV$ to every knowledge and action state that correctly reflects how valuable that state is with respect to the system's goal; if that assignment is arbitrary or contested, every semantic and pragmatic number the paper reports is arbitrary.","fun_headline_variants_meta":{"raw":{"variants":["Info value = change in goal-state score","How much a message moves you toward a goal","Goal-state delta: the true worth of information","Pragmatic info metrics: value by goal progress","Beyond bits: measuring info's goal impact"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000875,"raw_usage":{"total_tokens":3851,"prompt_tokens":1074,"completion_tokens":2777,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":690,"completion_tokens_details":{"reasoning_tokens":2707}},"tokens_in":690,"tokens_out":2777,"duration_ms":22965,"temperature":1.0,"reasoning_tokens":2707,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-16T04:19:57.194157+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take one recorded simulation of the robotic collective and compute $V_{\\mathrm{pr,gl}}$ for the same information flow under two different state-value functions: one that scores the robot distribution by absolute error from the desired counts and one that scores it by object-weighted error. If the two state-value functions reverse the ranking of the main and ground-truth strategies, the pragmatic measure is an artifact of the analyst's choice rather than a property of the information flow, and the paper's central claim that these measures quantify the value of information flows would not be well-defined.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the entropy measure used for syntactic information content and resource accounting."},{"cited_title":"J., and Mellodge, P","cited_arxiv_id":null,"evidence_quote":"Introduces the Multi-Scale Feedback Systems concept that this paper extends with information measures."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the robotic collective model and response-threshold-based goal selection used in the first case study."},{"cited_title":"H., Heslop-Harrison, P., and Cho, K.-H","cited_arxiv_id":null,"evidence_quote":"Supplies the differential-equation model of coupled biochemical oscillators underlying the oscillator case."},{"cited_title":"and Wolpert, D","cited_arxiv_id":null,"evidence_quote":"Establishes the viability-based approach to semantic information value that the paper's goal-relative measures build on."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Defines pragmatic information with respect to an action, the contrast point for the paper's efficiency-based extension."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the collective decision-making model of agents estimating task sizes used in the second case study."}],"review_version":1}