Hormesis as a Geometric Necessity of Bounded Adaptive Systems: Quantitative Predictions From First Principles Authors: Daniel John Murray Published: 2026 ABSTRACT ObjectivesTo determine whether hormesis can be derived from the geometry of bounded adaptive biological endpoints, and whether the derivation yields quantitative predictions independent of curve fitting.MethodsBounded endpoints were represented in rapidity coordinates using the arctanh linearisation of the adopted Möbius composition law. Repair activation and damage accumulation were modelled as opposing bounded rapidity increments with independently measurable thresholds. The molecular-to-functional attenuation factor was derived for linear pathways from the metabolic-control summation theorem and separated from a testable multi-target extension. Predictions were compared with published H2O2, CdCl2, and heat-shock dose-response data.ResultsThe model predicts a biphasic response whenever repair activation precedes toxicity (Da < Dt) and high-dose damage ultimately exceeds bounded repair capacity. The exact hyperbolic model is the primary model; the product form is used only as a conservative analytical approximation. Aggregate predictions — peak-amplitude range, mean amplitude, and hormetic-zone width — matched the Calabrese hormesis database of more than 10,000 responses without parameter fitting; across three mechanistically distinct agents, independently published data confirmed the structural prediction that adaptive activation precedes toxicity, and pathway-specific attenuation distinguished linear NRF2-mediated responses (130–160%) from larger multi-target heat-shock responses (200–300%). The peak-location law is presented as a falsifiable prediction.ConclusionWithin the stated scope of independently composing bounded adaptive endpoints, hormesis follows from finite repair capacity plus adaptive upregulation. The framework does not claim that all biological history is axiomatic; it identifies the conditional geometry that constrains dose-response shape once a bounded adaptive endpoint is specified. 1. Introduction The hormetic dose-response has been documented in over 10,000 dose-response relationships across more than 600 substances.1,2 Its quantitative features are remarkably consistent: peak stimulation typically 130–160% of control, approximately 80% of responses below 200%, and the hormetic zone extending 5–50 fold below the toxicity threshold.3,4 These features persist across phylogenetic kingdoms, endpoints, and stressor types. No existing framework explains this consistency. Specific mechanisms have been identified for individual substances — NRF2-mediated antioxidant upregulation for oxidative stressors, 5 heat shock protein induction for thermal stress, 6 DNA repair pathway activation for genotoxic agents 7 — but the question of why mechanistically distinct responses produce quantitatively similar features remains open. The Linear No-Threshold (LNT) model, which dominates regulatory toxicology, predicts proportional harm at all doses and is, for bounded adaptive endpoints, inconsistent with the low-dose stimulation seen in hormetic data.8,9 This paper derives hormesis from first principles. The derivation is structured as follows: Section 2 establishes the mathematical framework (Aczél’s theorem applied to bounded biological observables). Section 3 derives the hormetic response as a necessary consequence of this framework. Section 4 derives the molecular-to-functional attenuation factor from metabolic control analysis. Section 5 presents five quantitative predictions with analytical expressions. Section 6 validates against published data. Section 7 discusses implications. Scope of the claim: The derivation is conditional, not universal in the unrestricted biological sense. It applies to bounded adaptive endpoints in which repair and damage contribute independently to the same measured state over the low-to-moderate dose range. Biological variability is therefore not ignored; it appears through Da, Dt, Amax, η, steepness, tissue state, and pathway architecture. The axiomatic element is the bounded-composition coordinate law, not the full evolutionary history of the organism. 2. Mathematical Framework: Bounded Composition 2.1. Bounded Observables The biological endpoints considered in this paper are bounded state variables. Enzyme activity is bounded by finite catalytic capacity, cell viability by the interval [0, 1], gene expression by transcriptional and translational limits, and concentration by finite solubility and transport constraints. This does not mean that complete biological histories are axiomatic or deterministic. Evolution, tissue state, prior exposure, and measurement context determine the numerical parameters of the response. The narrower claim is that, once a single adaptive endpoint is represented as a finite bounded observable and perturbations compose consistently on that endpoint, the admissible coordinate geometry is constrained. 2.2. Aczél’s Additive Representation Theorem Aczél proved that, for a continuous composition law on the interval (−L, L) that is associative, (δ1 ⊕ δ2) ⊕ δ3 = δ1 ⊕ (δ2 ⊕ δ3), and strictly monotone in each argument, there exists a continuous, strictly monotonic generator, unique up to a positive multiplicative factor, that linearises the composition into ordinary addition. Aczél’s theorem does not by itself select the inverse hyperbolic tangent. The present model adopts the Möbius composition law on (−L, L), whose additive generator is: ρ=arctanh(δ/L) The composition law in the original (bounded) coordinates is: δ1⊕δ2=L · tanh( arctanh(δ1/L)+arctanh(δ2/L)) In the transformed (rapidity) coordinates, composition is simple addition: ρ12=ρ1+ρ2 This construction is not a biological model. It is a mathematical representation of the adopted boundedassociative composition law. The term geometric is used in this precise sense: the bounded interval is mapped to an unbounded linear coordinate by arctanh, and composition becomes addition in that coordinate. The theorem does not specify which biological pathway is activated; it specifies the coordinate structure once a bounded observable and a consistent composition operation have been chosen. The proof is in Aczél, Chapter 6. 2.3. The Logistic Function as the Canonical Dose-Response The dose-response of a bounded observable to an increasing input (dose) corresponds to the accumulation of compositional increments. The simplest case — and the one most commonly realised in biological systems — occurs when each unit of dose contributes a fixed rapidity increment s, so that the total rapidity at dose d is: ρ(d)=s · (d−D0) where D0 is the dose at which ρ = 0 (the midpoint). This linear dose-rapidity mapping is an assumption, not a consequence of Aczél’s theorem alone. The theorem guarantees that some linearising transformation exists; the linearity of the dose-rapidity relationship is an additional modelling assumption, justified by the observation that for most biological systems, the dose-response is well-described by the logistic over several orders of magnitude. Under this assumption, the observable in physical coordinates is: δ(d)=L · tanh(s · (d−D0)) Normalising to the interval [0, 1] by the transformation σ = (δ/L + 1)/2: σ(d)=1/(1+exp(‐2s · (d−D0))) This is the logistic function, with steepness parameter 2s. It arises from combining the adopted Möbius composition law, whose additive generator is arctanh, with thelinear dose-rapidity assumption (which specifies the argument). Its ubiquity in biological dose-response data (Hill equation, Michaelis-Menten, logistic growth) is consistent with this derivation. Crucially, the central result of this paper — that bounded adaptive systems produce biphasic dose-responses — does not depend on the specific sigmoid shape. The product of two bounded increasing-then-saturating functions whose midpoints are separated — with the activating function rising at the lower dose and contributing sufficient low-dose amplitude — is biphasic and rises above the control level over a low-dose interval. The logistic is used for quantitative predictions because it is the canonical form under the adopted Möbius composition law with linear dose-rapidity mapping, but the qualitative result (biphasic shape, hormetic zone) is robust to relaxation of the linearity assumption. Even if the dose-rapidity relationship is mildly nonlinear, producing a sigmoid that deviates from the logistic, the biphasic structure persists. 2.4. The Steepness Parameter Is Constrained The steepness 2s determines how rapidly the sigmoid transitions from 0 to 1. For a system that transitions over a characteristic dose range Δ (from 10% to 90% activation), the steepness is 2s ≈ 2 · ln(9)/Δ ≈ 4.4/Δ. Since Δ is a measurable property of the activation curve, the steepness is not a free parameter but a consequence of the system’s activation kinetics. 3. Derivation of the Hormetic Response 3.1. Premises Premise 1 (Bounded repair): The repair capacity of a biological system is a bounded observable confined to [Rbase, Rmax]. This is a physical fact: no enzyme has infinite activity, no pathway has infinite throughput. Premise 2 (Adaptive upregulation): The repair system increases its activity in response to perturbation, from Rbase (constitutive level) toward Rmax (maximum inducible level). This is the definition of biological adaptation, documented across all kingdoms.10-12 3.2. Rapidity-Space Composition Under the adopted Möbius composition law, repair and damage are modelled as sequential inputscomposing additively in rapidity space on the same bounded viability observable. Define:• ρ base = arctanh(Vbase/L): the baseline viability in rapidity coordinates, where the bracketed ratio is taken on the centred bounded scale of Section 2.3 (equal to 2v − 1 for the [0,1]-normalised viability v) and therefore lies strictly within (−1, 1), keeping the baseline rapidity finite• ρ repair (d) = Aρ · σ(d, Da): the rapidity boost from repair upregulation (a bounded increase from 0 to Aρ, with half-activation at dose Da, where σ(d, D) = 1/(1 + exp(−s(d − D))) is the logistic function derived in Section 2.3)• ρ damage (d) = Bρ · σ(d, Dt): the rapidity cost of damage accumulation (a bounded increase from 0 to Bρ, with half-activation at dose Dt; the steepness parameters sa and st may differ between repair and damage, reflecting their distinct activation kinetics) • ρ base = arctanh(Vbase/L): the baseline viability in rapidity coordinates, where the bracketed ratio is taken on the centred bounded scale of Section 2.3 (equal to 2v − 1 for the [0,1]-normalised viability v) and therefore lies strictly within (−1, 1), keeping the baseline rapidity finite • ρ base = arctanh(Vbase/L): the baseline viability in rapidity coordinates, where the bracketed ratio is taken on the centred bounded scale of Section 2.3 (equal to 2v − 1 for the [0,1]-normalised viability v) and therefore lies strictly within (−1, 1), keeping the baseline rapidity finite • ρ repair (d) = Aρ · σ(d, Da): the rapidity boost from repair upregulation (a bounded increase from 0 to Aρ, with half-activation at dose Da, where σ(d, D) = 1/(1 + exp(−s(d − D))) is the logistic function derived in Section 2.3) • ρ repair (d) = Aρ · σ(d, Da): the rapidity boost from repair upregulation (a bounded increase from 0 to Aρ, with half-activation at dose Da, where σ(d, D) = 1/(1 + exp(−s(d − D))) is the logistic function derived in Section 2.3) • ρ damage (d) = Bρ · σ(d, Dt): the rapidity cost of damage accumulation (a bounded increase from 0 to Bρ, with half-activation at dose Dt; the steepness parameters sa and st may differ between repair and damage, reflecting their distinct activation kinetics) • ρ damage (d) = Bρ · σ(d, Dt): the rapidity cost of damage accumulation (a bounded increase from 0 to Bρ, with half-activation at dose Dt; the steepness parameters sa and st may differ between repair and damage, reflecting their distinct activation kinetics) The net rapidity of the system at dose d is: (1) ρnet(d)=ρbase+ρrepair(d) −ρdamage(d) The additivity of rapidities requires justification. Under the adopted Möbius composition law for the single bounded viability observable, arctanh linearisescomposition and composition in rapidity space is addition. Equation (1) applies this as follows: viability is the single bounded observable; repair and damage are two sequential inputs composing on that observable. The system first receives a repair boost (which increases its rapidity above baseline) and then receives a damage load (which decreases it). Because both inputs compose on the SAME bounded observable (viability) through the SAME composition operation, the theorem applies: each input contributes an additive rapidity increment (positive for repair, negative for damage), and these increments sum. This additive structure holds under a specific scope condition: the repair and damage pathways must not directly chemically interact at the point of composition. If the damage product inhibits the repair enzyme, or the repair enzyme is consumed by the damage reaction, then the rapidity contributed by each depends on the current state of the other, and the system requires coupled composition equations rather than simple rapidity addition. For most hormetic systems in the low-to-moderate dose range — where the hormetic zone resides — repair and damage operate on different molecular targets (e.g., NRF2-mediated transcription vs. direct oxidative damage to lipids) and this scope condition is satisfied. At very high doses, where molecular systems are simultaneously overwhelmed and degraded, the independence breaks down; but this regime is beyond the hormetic zone and does not affect the predictions of interest. Scope condition (stated explicitly): The rapidity-addition model applies when the repair and damage pathways compose independently on the bounded viability interval — i.e., when the rapidity increment contributed by repair does not depend on the current damage load, and vice versa. This condition is evaluated explicitly for the agents examined in Section 6: it holds for the hydrogen peroxide and heat-shock endpoints and is endpoint-dependent for cadmium — satisfied for the proliferation endpoint but not for the metabolic-viability endpoint that cadmium impairs directly. More generally it is expected to hold in the hormetic (low-dose) regime where repair activation precedes significant damage. 3.3. The Response Is Necessarily Biphasic The physical observable (viability, normalised to control) is: (2) Response(d)={1+tanh[ρnet(d)]} / {1+tanh(ρbase)} Now consider the structure of ρnet(d):• At d = 0, the logistic tails satisfy ρrepair(0) = Aρ·σ(0, Da) and ρdamage(0) = Bρ·σ(0, Dt). These are generally small but not identically zero. Thus ρnet(0) is approximately ρbase when saDa and stDt are sufficiently large, and the zero-dose response is correspondingly approximately the control level.• At low d (d near Da, well below Dt): ρrepair rises (repair activates), ρdamage ≈ 0 (damage has not engaged). So ρnet > ρbase. Response > 1. This is the hormetic zone.• At d between Da and Dt: ρrepair approaches its maximum Aρ, ρdamage begins to rise. ρnet peaks and begins to decline.• At high d (d >> Dt): ρdamage dominates. ρnet < ρbase. Response < 1. This is the toxicity zone. • At d = 0, the logistic tails satisfy ρrepair(0) = Aρ·σ(0, Da) and ρdamage(0) = Bρ·σ(0, Dt). These are generally small but not identically zero. Thus ρnet(0) is approximately ρbase when saDa and stDt are sufficiently large, and the zero-dose response is correspondingly approximately the control level. • At d = 0, the logistic tails satisfy ρrepair(0) = Aρ·σ(0, Da) and ρdamage(0) = Bρ·σ(0, Dt). These are generally small but not identically zero. Thus ρnet(0) is approximately ρbase when saDa and stDt are sufficiently large, and the zero-dose response is correspondingly approximately the control level. • At low d (d near Da, well below Dt): ρrepair rises (repair activates), ρdamage ≈ 0 (damage has not engaged). So ρnet > ρbase. Response > 1. This is the hormetic zone. • At low d (d near Da, well below Dt): ρrepair rises (repair activates), ρdamage ≈ 0 (damage has not engaged). So ρnet > ρbase. Response > 1. This is the hormetic zone. • At d between Da and Dt: ρrepair approaches its maximum Aρ, ρdamage begins to rise. ρnet peaks and begins to decline. • At d between Da and Dt: ρrepair approaches its maximum Aρ, ρdamage begins to rise. ρnet peaks and begins to decline. • At high d (d >> Dt): ρdamage dominates. ρnet < ρbase. Response < 1. This is the toxicity zone. • At high d (d >> Dt): ρdamage dominates. ρnet < ρbase. Response < 1. This is the toxicity zone. The transition from Response > 1 to Response < 1 is guaranteed within the stated scope whenever: (a) ρrepair activates before ρdamage (Da < Dt), AND (b) ρdamage eventually exceeds ρrepair (Bρ > Aρ at high dose) Condition (a), Da < Dt, is an explicit scope condition defining the class of adaptive endpoints considered here: the repair-activation midpoint precedes the damage midpoint, providing the threshold ordering used to derive a resolved low‐dose stimulatory interval. It does not follow from boundedness or adaptation alone and must therefore be assessed empirically for each endpoint, as is done using independent published evidence for the three agents in Section 6.3. Condition (b) states that the damage increment ultimately exceeds the repair increment, Bρ > Aρ. In the present model both terms are bounded sigmoids, so this does not follow from Premise 1 alone; it is an independent high-dose toxicity condition, satisfied for any endpoint represented by this model whose high‐dose response ultimately falls below baseline. The biphasic result is conditional on this inequality. Therefore, any bounded adaptive endpoint satisfying Premises 1 and 2, under the stated scope condition of independent composition, must produce a biphasic dose-response. The result is not a claim that every biological system is simple or history-free; it is a claim that bounded adaptive composition forces a low-dose region in which repair can exceed damage before high-dose damage dominates. This is hormesis as a conditional geometric consequence, not as an empirical curve assumed in advance. 3.4. Product Form as Leading-Order Approximation The exact hyperbolic model (Equations 1–2) is the fundamental model. The product form below is a leading-order analytical approximation valid when repair and damage perturbations are moderate relative to ρbase. It is retained because it makes peak location, zone width, and amplitude dependencies transparent and because it is conservative: across the parameter range tested in the Appendix, it preserves biphasic structure and slightly underestimates amplitudes relative to the exact hyperbolic model. (3) Response(d)≈(1+Amax · σ(d,Da))×(1−σ(d,Dt)) where Amax is the fractional increase in protective capacity and σ is the logistic function (Section 2.3). Equation 3 is not used as the proof of hormesis; the proof is the rapidity-space result in Equations 1–2. The approximation is used for closed-form predictions, and the exact model is used as the reference model for numerical verification. Thus any discrepancy between Equation 2 and Equation 3 affects numerical precision, not the existence of the biphasic response. 4. Derivation of the Molecular-To-Functional Attenuation Factor 4.1. The Measurement Problem Standard biological assays measure molecular upregulation: fold-change in mRNA, protein expression, or enzyme concentration. These molecular fold-changes systematically overestimate the protective function because the mapping from enzyme concentration to pathway flux is not 1:1. 4.2. Derivation From the Summation Theorem Kacser and Burns 13 and Heinrich and Rapoport 14 independently proved the summation theorem of metabolic control analysis (MCA): ΣiCiJ=1 where CiJ is the flux control coefficient of enzyme i — the fractional change in pathway flux J per fractional change in enzyme i concentration, holding all other enzyme concentrations constant. The sum is over all enzymes in the pathway. For a pathway with n rate-limiting steps of comparable importance, each enzyme has: Ci≈1/n When a single enzyme in the pathway is upregulated by a fractional amount (Amax,mol), the resulting fractional change in pathway flux (Amax,func) is: (4) Amax,func≈Ci · Amax,mol=Amax,mol / n The molecular-to-functional attenuation factor is therefore: (5) η=Amax,mol / Amax,func≈n For protective pathways with 2–3 rate-limiting enzymatic steps, η ≈ 2–3. This is not a fitted parameter. It is derived from the summation theorem, a proven result of metabolic control analysis. The specific value of η for any given pathway can be determined from its architecture: Linear pathways (NRF2-antioxidant, DNA repair): The protective chain typically involves 2–3 sequential bottlenecks: transcription factor activation → gene expression → enzyme maturation and localisation. Each step shares control approximately equally, giving η ≈ 2–3. For example, upregulating NQO1 mRNA 2-fold increases functional antioxidant protection by approximately 1/2 to 1/3 of that fold-change, because the downstream steps (translation, post-translational modification, substrate access) are not co-upregulated. Multi-target effectors (HSP70, certain cytokines): The summation theorem applies rigorously to linear flux pathways. A molecule such as HSP70 is different because the same induced effector can contribute to several survival-relevant processes at once, including protein refolding, membrane stabilisation, anti-apoptotic signalling, and proteostasis. The present manuscript therefore separates two claims. The rigorous claim is the linear-pathway attenuation η ≈ n. The multi-target claim is an architecture-dependent, independently testable extension: when one effector acts on m partly independent survival nodes, the effective attenuation should be smaller than in a single linear pathway because one molecular increase contributes to multiple functional outputs. This pathway-dependent η makes a specific prediction that is not required for the core proof of biphasic shape: hormetic peak amplitude should be larger for agents whose protection is mediated by multi-target effectors than for agents acting through single linear pathways. This prediction can be tested directly by measuring molecular upregulation and functional protection in the same experimental system before fitting any hormetic curve. 4.3. Numerical Verification For a 2-enzyme linear pathway (n = 2, C = 0.5): doubling one enzyme (molecular fold = 2×, Amax,mol = 1.0) increases flux by 50% (Amax,func = 0.5). η = 2.0. For a 3-enzyme linear pathway (n = 3, C = 0.33): doubling one enzyme increases flux by 33% (Amax,func = 0.33). η = 3.0. For a multi-target effector acting on three survival nodes through approximately two-step pathways, the predicted effective control is higher than for one two- or three-step linear pathway, giving η ≈ 1.3–1.5. This numerical estimate is an explicit hypothesis about architecture, not an additional axiom. It is included because it explains why heat-shock responses can exceed the usual 130–160% chemical-stressor range and because it provides a falsifiable test of the model. 4.4. Pathway-specific η Values for Validation Agents For linear pathways, η values are derived from the summation theorem (Table 1). For multi-target effectors, η values are predicted by architecture and must be independently tested by measuring both molecular upregulation and functional protection in the same system. The core geometric predictions of peak location and zone width do not depend on this multi-target extension.Table 1.Predicted molecular-to-functional attenuation factor (η) for each validation agent, based on protective-pathway architecture.AgentProtective pathwayArchitecturePredicted ηH2O2NRF2 → NQO1/SOD/GPxLinear, 2–3 steps2.0–3.0CdCl2NRF2 → NQO1/HO-1Linear, 2–3 steps2.0–3.0Heat shockHSP70 → folding + membranes + anti-apoptosisMulti-target, 3–4 nodes1.3–1.5 (predicted)Linear NRF2-mediated pathways (H₂O₂, CdCl₂) involve 2–3 rate-limiting steps (η ≈ 2–3); the multi-target HSP70 heat-shock response acts on 3–4 survival nodes, giving a predicted η ≈ 1.3–1.5. Table 1.Predicted molecular-to-functional attenuation factor (η) for each validation agent, based on protective-pathway architecture.AgentProtective pathwayArchitecturePredicted ηH2O2NRF2 → NQO1/SOD/GPxLinear, 2–3 steps2.0–3.0CdCl2NRF2 → NQO1/HO-1Linear, 2–3 steps2.0–3.0Heat shockHSP70 → folding + membranes + anti-apoptosisMulti-target, 3–4 nodes1.3–1.5 (predicted)Linear NRF2-mediated pathways (H₂O₂, CdCl₂) involve 2–3 rate-limiting steps (η ≈ 2–3); the multi-target HSP70 heat-shock response acts on 3–4 survival nodes, giving a predicted η ≈ 1.3–1.5. Predicted molecular-to-functional attenuation factor (η) for each validation agent, based on protective-pathway architecture. Linear NRF2-mediated pathways (H₂O₂, CdCl₂) involve 2–3 rate-limiting steps (η ≈ 2–3); the multi-target HSP70 heat-shock response acts on 3–4 survival nodes, giving a predicted η ≈ 1.3–1.5. 5. Quantitative Predictions 5.1. Analytical Derivation of Peak Dose For analytical transparency, the peak of the product approximation (Equation 3) occurs where dR/dd = 0: (6) A · σ′(d*,Da) · [1−σ(d*,Dt)]=[1+A · σ(d*,Da)]· σ′(d*,Dt) where σ′(d, D) = s · σ(d, D) · [1 − σ(d, D)] is the logistic derivative. For well-separated thresholds (Dt >> Da), at the peak dose d* the damage sigmoid is still near zero: σ(d*, Dt) ≈ 0 and σ′(d*, Dt) is small. The peak is therefore governed primarily by the repair curve’s dynamics. Numerical evaluation of the exact hyperbolic model and the product approximation across the biologically plausible parameter space (Dt/Da = 5–20, Amax = 0.3–0.8) yields the same peak-order prediction: (7) d* / Da≈1.7 to 2.4 The ratio increases with Dt/Da (wider threshold separation pushes the peak rightward because damage engages later) and with Amax (stronger repair extends the beneficial zone). For the typical range Dt/Da = 10–20, the peak sits at approximately 2.0–2.4 × D a . Prediction 1: For any substance showing hormesis, the peak dose should correlate with the independently measured repair activation EC50, with a proportionality constant of approximately 2. 5.2. Analytical Derivation of Zone Width The hormetic zone boundaries occur where Response = 1, i.e., where (1 + A · σ(d, Da)) · (1 − σ(d, Dt)) = 1. The lower boundary dlo occurs at low dose where repair is just beginning to exceed unity; the upper boundary dhi occurs where damage overtakes the repair boost. Numerical evaluation yields zone widths (dhi/dlo) of 8–80 fold across the biologically plausible parameter space, overlapping the Calabrese database range of 5–50 fold. Prediction 2: The zone width should scale approximately with the ratio Dt/Da. 5.3. Peak Amplitude From Section 4: (8) Peak amplitude≈(1+Amax,mol / η)×100% of control where η ≈ n (number of rate-limiting enzymes in the protective pathway). For Amax,mol = 0.8–1.5 and η = 2–3: peak amplitude = 127–175% of control. Mean ≈ 145%. Prediction 3: The peak amplitude is determined by the functional repair capacity Amax,func = Amax,mol/η, not by the molecular fold-change alone. 5.4. Repair Capacity Determines Zone Width Prediction 4: Cell types with higher Amax,func should show wider hormetic zones. Testable by comparing NRF2-competent versus NRF2-knockout cells, or young versus aged cells. 5.5. Molecular-To-Functional Attenuation Is Pathway-Determined Prediction 5: The ratio η = Amax,mol/Amax,func is determined by the architecture of the protective pathway, not by the stressor. Specifically, η ≈ n for linear pathways with n rate-limiting steps, and η < n for multi-target effectors acting on m survival nodes, where distributing a single molecular increase across m partly independent survival outputs reduces the effective attenuation below the single-pathway value (predicted η ≈ 1.3–1.5 for HSP70; see Section 4.4). This is testable by independently measuring both molecular upregulation and functional protection in the same system. The prediction distinguishes between linear-pathway agents (η ≈ 2–3, peak amplitude 130–160%) and multi-target agents (η ≈ 1.3–1.5, peak amplitude 170–200%+). 6. Validation Against Published Data 6.1. Aggregate Agreement With the Hormesis Database Beyond the global database, the subclass of responses most relevant to this framework — NRF2-mediated adaptive responses — has been characterised independently. Calabrese and Kozumbo 15 identify NRF2 activation as a general mechanism of the hormetic dose-response and note that the resulting stimulatory response is characteristically constrained to a modest magnitude across diverse agents, consistent with the amplitude bound derived in Section 4. In a complementary high-resolution analysis, Bogen 16 re-examined NRF2-ARE activation across fifteen chemicals at twelve concentrations each in human HepG2 cells and reported a bounded, J-shaped dose-response governed by an ultrasensitive molecular switch — independent empirical support for the sigmoidal composition law of Section 2. 6.2. Molecular-To-Functional Attenuation as an Account of a Robust Empirical Mismatch A recurring but rarely-explained feature of the adaptive-response literature is a mismatch of scales: molecular induction of protective effectors is frequently several-fold, whereas the corresponding functional hormetic peak is modest (typically 130–160% of control). Sulforaphane, the most extensively characterised NRF2 inducer, illustrates this directly — it raises NQO1 and NRF2 expression several-fold across a low-micromolar range, yet the associated functional stimulation peaks only modestly before toxicity engages at higher concentrations. 15 The molecular and functional scales are typically measured in separate studies and treated as unrelated; the present framework predicts that they are not, and specifies how they are connected. The summation theorem of metabolic control analysis (Section 4) accounts for this constraint. Because the flux-control coefficients of the enzymes in a pathway sum to unity (ΣCi = 1), a multi-fold increase in any single effector is compressed into a fractional change in pathway flux by a factor η ≈ n, where n is the number of rate-limiting steps. Large molecular inductions therefore generically yield only modest functional peaks — the pattern that is observed. This is offered as an explanation of an existing empirical regularity, not as a measurement of η: a direct estimate requires molecular induction (A_max_mol) and functional protection (A_max_func) to be quantified in the same system, which we identify as a specific experimental target rather than a result claimed here. 6.3. Structural Validation Across Three Mechanistically Distinct Agents The framework’s central structural requirement is that adaptive repair activates at a lower dose than overt toxicity (Da < Dt; Section 3.3), so that a stimulatory response necessarily precedes inhibition. We tested this requirement against published dose-response data for three mechanistically distinct agents, taking parameters from primary studies that measured molecular activation and toxicity independently of any hormetic-curve fitting. Because matched, fine-grained dose-response data within the hormetic window are not uniformly available for these systems, the agent-level analysis tests the structural prediction — the ordering Da < Dt and the resulting biphasic separation — rather than agent-specific numerical peak amplitudes; the sharper quantitative claim (peak ≈ 2 × Da) is treated as a falsifiable prediction (Section 5.1). Test 1: CdCl2 in HepG2 hepatocytes (NRF2 pathway). Zhu et al 17 measured NRF2 activation and cytotoxicity in the same system. NRF2 protein rose from a control level of 0.60 to a plateau near 2.24 (relative units), with half-maximal activation between 2 and 5 μM (Da ≈ 2–3 μM); the downstream NRF2 targets were induced over the low-dose range, with HO-1 and AKR1C1 significant from 2 μM and GCLC and GCLM from 5 μM. Cell viability was essentially unchanged up to 5 μM and fell significantly only above 10 μM (Dt onset ≈ 10 μM; (here taken as the onset of significant toxicity rather than the sigmoid midpoint; the corresponding viability IC50 is approximately 60 μM) IC50 ≈ 60 μM, estimated by interpolation from the viability curve, which declined to ≈ 55% at 50 μM and ≈ 34% at 100 μM). Adaptive activation therefore precedes toxicity with a threshold separation Dt/Da ≈ 4–5, satisfying the structural requirement. Consistent with the framework, MTS viability showed at most a small low-dose elevation (≈108% of control at 1 μM, within measurement scatter); acute viability in healthy HepG2 cells is bounded near the control maximum, so the clearer stimulatory response appears instead in proliferation, where low-dose Cd increases cell number 18 ; both observations place the stimulatory response below the toxicity threshold. This agent is also informative as a boundary case. Cadmium directly impairs mitochondrial and metabolic targets, so for the metabolic-viability endpoint the independent-composition scope condition (Section 3.2) is not satisfied and no resolved viability peak is expected — consistent with the MTT data. Hormesis nonetheless re-emerges in the proliferation endpoint, where that condition holds. The framework therefore predicts not merely that hormesis occurs, but in which endpoint it appears and in which it does not — a discriminating prediction the data bear out. Test 2: H2O2 in human keratinocytes (NRF2 pathway). Yokoo et al 19 reported a hormetic functional response in NHEK-F keratinocytes: 20 μM H2O2, which was non-cytotoxic, extended replicative lifespan to ∼160% of control, and 60 μM, which was marginally cytotoxic, to ∼120%; in the same keratinocyte lineage, Han et al applied 0.3 mM H2O2 as an oxidative-stress challenge in HaCaT cells. 20 The location of the stimulatory peak (≤20 μM), the graded decline toward the cytotoxic range, and the wide separation between stimulatory and toxic doses are consistent with the predicted biphasic geometry. This system is treated as a structural and falsification test rather than a quantitative anchor for the peak-location law, because same-system NRF2-activation EC50 data are not available for the keratinocyte H2O2 model: H2O2 is conventionally applied as an acute bolus challenge rather than titrated as an adaptive-pathway inducer. A direct test of the peak ≈ 2 × Da prediction requires the H2O2 NRF2-activation EC50 in the same keratinocyte model, which is identified here as a specific experimental target (Section 5.1). Test 3: Heat shock in human fibroblasts (HSP70 pathway). Repeated mild heat stress (41°C) produces well-documented hormetic, anti-aging effects in human skin fibroblasts, whereas severe heat stress is damaging.6,21 Demirovic et al 22 showed that the dominant adaptive controller HSF1 translocates to the nucleus substantially more strongly under mild than under severe stress in young cells (∼5-fold), confirming that adaptive activation is concentrated in the mild, sub-toxic regime, consistent with Da < Dt. As predicted for a multi-target effector (η ≈ 1.3–1.5; Section 4.4), heat-shock hormesis exhibits the largest peak amplitudes of the three systems (200–300%). 6.4. Summary Across all three agents, the structural prediction — adaptive activation below the toxicity threshold, producing a stimulatory response separated from overt toxicity — is supported by real, independently published data (Table 3). The aggregate quantitative predictions (peak-amplitude range, mean amplitude, and zone width) match the database of more than 10,000 responses without parameter fitting (Table 2; Section 6.1).Table 2.Comparison of model predictions with aggregate features of the Calabrese hormesis database (>10,000 responses).FeatureModel predictionCalabrese database (>10,000 responses)Peak amplitude range127–175%130–160% typical, up to 200%Mean peak amplitude∼145%∼140%Zone width range8–80 fold5–50 foldCurve shapeBiphasicBiphasicThe model was not fitted to the database.Table 3.Structural validation across three mechanistically distinct agents.Agent/systemReal data sourceStructural prediction (Da < Dt; biphasic)H2O2/keratinocytesYokoo (2004); Han (2023)SupportedCdCl2/HepG2 (NRF2)Zhu (2017); Niture (2023)SupportedHeat shock/fibroblastsDemirovic (2014); Rattan (2009)SupportedFor each system the independently published data source, and whether it supports the structural prediction (adaptive repair activation preceding toxicity, Da < Dt, yielding a biphasic response), are shown. Table 2.Comparison of model predictions with aggregate features of the Calabrese hormesis database (>10,000 responses).FeatureModel predictionCalabrese database (>10,000 responses)Peak amplitude range127–175%130–160% typical, up to 200%Mean peak amplitude∼145%∼140%Zone width range8–80 fold5–50 foldCurve shapeBiphasicBiphasicThe model was not fitted to the database. Comparison of model predictions with aggregate features of the Calabrese hormesis database (>10,000 responses). The model was not fitted to the database. Table 3.Structural validation across three mechanistically distinct agents.Agent/systemReal data sourceStructural prediction (Da < Dt; biphasic)H2O2/keratinocytesYokoo (2004); Han (2023)SupportedCdCl2/HepG2 (NRF2)Zhu (2017); Niture (2023)SupportedHeat shock/fibroblastsDemirovic (2014); Rattan (2009)SupportedFor each system the independently published data source, and whether it supports the structural prediction (adaptive repair activation preceding toxicity, Da < Dt, yielding a biphasic response), are shown. Structural validation across three mechanistically distinct agents. For each system the independently published data source, and whether it supports the structural prediction (adaptive repair activation preceding toxicity, Da < Dt, yielding a biphasic response), are shown. The predicted class separation in amplitude is consistent with the compiled evidence: linear NRF2-mediated pathways cluster at 130–160% (η ≈ 2–3), whereas the multi-target HSP70-mediated response reaches 200–300% (η ≈ 1.3–1.5). This class separation between single-pathway and multi-target effectors — rather than an agent-by-agent numerical match — is the substantive amplitude result. The framework’s sharper quantitative claim, that the stimulatory peak occurs at approximately 2 × Da, is presented as a falsifiable prediction (Section 5.1) rather than a completed agent-by-agent validation. Testing it requires dose-resolved measurement of Da (the adaptive-marker EC50) and of the functional peak in the same system — a dedicated experiment that the datasets compiled here were not designed to provide. The validation claimed here is therefore threefold: aggregate agreement with the hormesis database (Section 6.1), the framework’s account of the molecular-to-functional attenuation that constrains hormetic amplitude (Section 6.2), and structural confirmation across three mechanistically distinct agents including a scope-condition boundary case (Section 6.3) — a deliberately conservative standard that makes no agent-by-agent numerical claim beyond what independent data support. 7. Discussion 7.1. Relationship to Existing Hormesis Models Prior quantitative models of hormesis1,23 fit biphasic curves to dose-response data using empirical parameters. These models describe the shape of hormetic curves effectively but do not explain why the shape arises or why its quantitative features are consistent across systems. The present model differs fundamentally: it derives the biphasic shape from the composition law of bounded observables 24 rather than fitting it empirically. The consistency of quantitative features follows from the summation theorem of metabolic control analysis, 13 not from parameter coincidence. No prior model, to the author’s knowledge, has derived hormesis as a geometric necessity of bounded systems or connected its amplitude to flux control coefficients. 7.2. Why Hormesis Is Ubiquitous Any system satisfying Premises 1 and 2, under the scope condition of independent composition (Section 3.2), must produce a biphasic dose-response. Since virtually all adaptive biological systems satisfy both premises, and since the scope condition holds in the low-dose regime where repair and damage pathways operate on different molecular targets, hormesis is expected to be common. Studies failing to detect it likely reflect insufficient dose resolution in the hormetic zone, systems with Da ≈ Dt (yielding an unresolvably narrow hormetic zone), or measurement noise overwhelming a small stimulatory signal. Notably, systems with Da ≈ Dt still exhibit hormesis geometrically; the biphasic shape is present but experimentally unresolvable. 7.3. Why Features Are Consistent — And When They Diverge Peak amplitudes for chemical stressors acting through linear pathways (NRF2, DNA repair) cluster in the 130–160% range because of two independent constraints: (1) molecular repair upregulation rarely exceeds 2–3 fold for any single pathway; (2) the summation theorem limits the flux effect of single-enzyme upregulation to 1/n of the molecular fold-change, with n ≈ 2–3 for typical linear pathways. Together, these yield functional Amax values of 0.3–0.6 and peak amplitudes of 130–160%. The model also predicts that agents acting through multi-target effectors (e.g., HSP70, certain growth factors) should show HIGHER peak amplitudes (170–300%), because the pathway-specific η is smaller for multi-target action. The Calabrese database is consistent with this: while 80% of hormetic responses show peaks below 200% (consistent with linear-pathway dominance), a significant minority show peaks of 200–400%, which the present model attributes to multi-target effector mechanisms with low η. This bimodal distribution of amplitudes — most clustered at 130–160%, a tail extending to 200%+ — is a novel prediction that can be tested by classifying agents according to the architecture of their protective pathways. 7.4. Morphological Adaptation and the Adaptation-Pathology Threshold The same logic applies to morphologically expressed adaptation. Hyperplasia, hypertrophy, and atrophy are not merely descriptive pathology terms; they are tissue-level examples of bounded adaptive reallocation. 25 A tissue exposed to load, injury, endocrine stimulation, nutrient limitation, or chronic stress can increase cell number, increase cell size, or reduce mass before frank failure occurs. These responses involve complex targets rather than a single linear enzyme chain: cytoskeletal organisation, 26 mechanotransduction, mitotic control, apoptosis, extracellular-matrix remodelling, and metabolic supply all contribute to the measured endpoint. In the present framework, these are multi-node adaptive responses with their own Da, Dt, and effective η values. The boundary between adaptation and pathology occurs when the protective/adaptive rapidity increment can no longer exceed the damage or dysregulation increment. Oncogenesis is therefore outside the simple single-endpoint proof, but it is naturally adjacent to it: dysregulated growth represents a case in which adaptive proliferative machinery, genomic instability, apoptosis evasion, 27 and tissue constraints no longer compose as a stable bounded repair response. This addition clarifies why the model predicts ordinary adaptive morphology before breakdown while not claiming that multistage cancer risk is reducible to a single hormetic endpoint. These morphological responses are instructive precisely because they are multi-target in the sense of Section 7.3, and the framework therefore makes a specific qualitative prediction about them. Compensatory hyperplasia (for example, the proliferative response of hepatocytes to partial hepatectomy or of epithelium to chronic irritation), physiological hypertrophy (such as load-induced enlargement of cardiac or skeletal muscle), and adaptive atrophy (such as disuse- or nutrient-limitation–driven reduction of tissue mass) are each governed not by a single linear enzyme chain but by the joint action of cytoskeletal organisation, mechanotransduction, mitotic control, and apoptosis. Because control over the measured endpoint is distributed across many contributing nodes, the pathway-specific attenuation factor η is small, and the amplitude argument of Section 7.3 predicts that such morphologically expressed adaptations should occupy the higher-amplitude regime — a larger adaptive reserve between baseline and the adaptive ceiling — than adaptations mediated by a single linear pathway. The adaptation–pathology threshold is then reached when the distributed adaptive capacity can no longer raise the protective rapidity faster than the damage or dysregulation it offsets: at that boundary, ordinarily reversible hyperplasia, hypertrophy, or atrophy gives way to the deregulated growth characteristic of oncogenesis, in which apoptosis evasion and genomic instability remove the very bounds the composition law assumes. The framework thus places these familiar morphological categories on the same axis as molecular hormesis — as bounded, multi-target adaptive responses with their own activation and toxicity doses — while locating oncogenesis just beyond the point where the bounded-composition premise fails. 7.5. Implications for the LNT Model The LNT model assumes proportional harm at all doses. The present derivation shows that this assumption is geometrically incompatible with a single bounded adaptive endpoint when repair activates before damage dominates. In that scoped case, the dose-response cannot remain linear and harmful down to arbitrarily low dose, because the repair term initially contributes a positive rapidity increment before the damage term overtakes it. This conclusion applies specifically to single adaptive endpoints such as SOD activity, cell viability, DNA repair capacity, or stress-protein-mediated survival under the independence conditions stated in Section 3.2. It does not by itself invalidate every regulatory use of LNT, especially for multistage carcinogenesis, population-level precautionary policy, or endpoints in which adaptive repair is absent, delayed, or inseparable from damage. The claim is narrower and stronger: for bounded adaptive endpoints satisfying the stated premises, LNT is excluded by the composition law. 7.6. Implications for Pharmaceutical Dosing The model predicts that the beneficial-effect peak for agents acting through adaptive pathways occurs at approximately 2 × Da, not the maximum tolerated dose. This is consistent with evidence for metronomic chemotherapy outperforming maximum-dose regimens 28 and with the biphasic dose-response of many chemotherapeutic agents.8,29 8. Conclusion Hormesis is a geometric necessity of bounded adaptive endpoints under the condition that repair and damage compose independently on the bounded viability interval. The derivation proceeds from two biological premises (finite repair capacity and adaptive upregulation), one scope condition (independent composition of repair and damage over the low-to-moderate dose range), one theorem (Aczél’s additive representation theorem), and one established result (the summation theorem of metabolic control analysis). No substance-specific curve shape is assumed. No parameters are fitted to hormetic data. Each parameter — Da, Dt, η — is either measured independently or derived from pathway architecture. The five quantitative predictions — peak dose location (≈2 × Da), zone width (scaling with Dt/Da), peak amplitude (determined by Amax,mol/η), repair-capacity dependence of zone width, and pathway-specific η — are each determined by independently measurable biological quantities. The pathway-specific derivation of η makes a novel prediction: agents acting through multi-target effectors should show systematically larger hormetic peaks than agents acting through linear pathways. This is consistent with the observed distribution in the Calabrese database. Validation comprised two parts: aggregate agreement with the Calabrese database of more than 10,000 responses — peak-amplitude range, mean amplitude, and zone width — without parameter fitting, and confirmation across three mechanistically distinct agents, using independently published data, that adaptive activation precedes toxicity, the structural condition for hormesis. Pathway-specific attenuation distinguished linear NRF2-mediated responses (130–160%) from larger multi-target heat-shock responses (200–300%); the peak-location law (peak at approximately twice the repair-activation dose) is presented as a falsifiable prediction, and the multi-target amplitude case as a testable architectural extension rather than part of the core proof. The result is a precise conditional claim: bounded adaptive composition forces hormesis when repair activation precedes toxicity and high-dose damage eventually exceeds bounded repair.