{"id":"48bd7e1f-5598-40af-8e4f-2516ba14c524","arxiv_id":"2601.09485","paper_version":2,"verdict":"CONDITIONAL","confidence":"LOW","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"New lower bounds are proven for P(H ≥ E[H]) in the hypergeometric distribution, including a simple bound of k/n when n ≥ 8k and a variance-based bound under additional conditions.","lead":"This paper derives explicit lower bounds on the probability that a hypergeometric random variable meets or exceeds its expectation under sampling without replacement. The bounds apply when the population is sufficiently large relative to sample size and may aid analysis in statistics and algorithms.","discovery_kind":"new_method","skeptic_critique":{"model":"grok-4.3","headline":"No significant objection identified","rationale":"The reader's identification of the parameter restrictions as the weakest assumption matches the structure of the proofs; the full text confirms the claims are narrowly scoped and the conditions are necessary for the stated inequalities. No additional load-bearing gap is present that would require changing the CONDITIONAL verdict.","tokens_in":1801,"tokens_out":315,"duration_ms":25095,"concrete_test":"Pick parameters satisfying the conditions with equality, e.g., n=80, k=10, i=40 (so E(H)=5 exactly). Compute the exact hypergeometric pmf and verify whether P(H ≥ 5) ≥ 10/80 = 0.125; repeat for the second bound at a point where E(H)=3 and (n-i)(n-k)/n >1.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claims are lower bounds on P(H ≥ E(H)) under explicitly stated restrictions (n ≥ 8k for the first; 1 ≤ E(H) ≤ min{i,k}-1 and (n-i)(n-k)/n > 1 for the second). These restrictions are used to control the regime where the hypergeometric is sufficiently far from boundary effects and the sampling fraction is small. The auxiliary bounds on tail conditional expectation and mean absolute deviation are invoked in a manner that directly supports the main inequalities without circularity or unstated approximations. No internal inconsistency appears in the derivation steps or the handling of the discrete nature of H.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.3","summary":"The paper establishes two explicit lower bounds on P(H ≥ E(H)) for a hypergeometric random variable H ~ Hypergeometric(n, i, k). The first states that P(H ≥ E(H)) ≥ k/n whenever n ≥ 8k. The second, under the restrictions 1 ≤ E(H) ≤ min{i,k}−1 and (n−i)(n−k)/n > 1, gives P(H ≥ E(H)) ≥ (e^{−1/12}/(4√2)) ⋅ √((n−1)/n) ⋅ √Var(H) / (1 + √(1 + ((n−1)/(n−k)) Var(H))). Auxiliary results include an upper bound on the tail conditional expectation E[H | H ≥ E(H)] and a lower bound on the mean absolute deviation E[|H − E(H)|].","tokens_in":1918,"tokens_out":659,"duration_ms":45478,"significance":"If the derivations hold, the results supply concrete, non-asymptotic lower bounds on the central probability mass of the hypergeometric distribution under explicit sampling-fraction restrictions. Such bounds are useful in combinatorial probability, sampling theory, and algorithmic applications where one needs guaranteed mass away from the lower tail without asymptotic approximations. The auxiliary bounds on conditional expectation and MAD are stated in closed form and may be reusable in other hypergeometric analyses.","major_comments":[{"comment":"§3, proof of Theorem 1: the derivation of the factor 8 in the hypothesis n ≥ 8k proceeds from the inequality E[H | H ≥ E(H)] ≤ E(H) + 1/2 together with a crude estimate on the MAD; the resulting constant appears loose and the proof does not indicate whether a smaller multiplier (e.g., n ≥ 4k) can be recovered by tightening the MAD lower bound.","section":"§3, Theorem 1"},{"comment":"§4, display (12): the second main inequality is obtained by combining the MAD lower bound with a one-sided Chebyshev-type estimate; the factor (n−1)/(n−k) inside the square root is introduced without an explicit justification that it remains bounded under the stated hypothesis (n−i)(n−k)/n > 1, which could affect the numerical sharpness of the constant e^{−1/12}/(4√2).","section":"§4, (12)"}],"minor_comments":[{"comment":"Abstract, second displayed inequality: the placement of the square-root term √((n−1)/n) is slightly ambiguous in the LaTeX; it should be clarified that it multiplies the entire fraction involving √Var(H).","section":"Abstract"},{"comment":"§2, notation paragraph: the range of i is stated as {1,…,n} but the later conditions require i ≥ E(H) + 1; a single sentence reconciling the global parameter range with the local hypotheses would improve readability.","section":"§2"}],"recommendation":"minor_revision","confidential_remarks":null},"author_rebuttal":{"model":"grok-4.3","summary":"We thank the referee for the careful reading and constructive comments. We respond to each major comment below.","responses":[{"response":"The factor of 8 is obtained by substituting the bound E[H | H ≥ E(H)] ≤ E(H) + 1/2 into the relation between the central probability and the MAD lower bound. The referee correctly notes that this choice is conservative. Our MAD lower bound is already stated in closed form for potential reuse, and tightening it further would require additional assumptions not present in the theorem. We will add a remark after Theorem 1 noting that the constant 8 is sufficient under the given hypotheses but may not be sharp, and that numerical evidence suggests n ≥ 4k often holds though it is not recovered by the current proof technique.","revision_made":"partial","referee_comment":"[§3, Theorem 1] §3, proof of Theorem 1: the derivation of the factor 8 in the hypothesis n ≥ 8k proceeds from the inequality E[H | H ≥ E(H)] ≤ E(H) + 1/2 together with a crude estimate on the MAD; the resulting constant appears loose and the proof does not indicate whether a smaller multiplier (e.g., n ≥ 4k) can be recovered by tightening the MAD lower bound."},{"response":"The factor (n−1)/(n−k) enters through the exact variance formula Var(H) = k(i/n)((n−i)/n)(n−k)/(n−1) when the one-sided Chebyshev inequality is applied to the MAD. The hypothesis (n−i)(n−k)/n > 1 guarantees that the denominator in the final bound is positive and that Var(H) is sufficiently large, but we agree an explicit sentence justifying the appearance and boundedness of the factor was omitted. We will revise the proof of the second bound to insert a short paragraph deriving the factor from the variance expression and confirming that the numerical constant e^{−1/12}/(4√2) remains valid under the stated restrictions.","revision_made":"yes","referee_comment":"[§4, (12)] §4, display (12): the second main inequality is obtained by combining the MAD lower bound with a one-sided Chebyshev-type estimate; the factor (n−1)/(n−k) inside the square root is introduced without an explicit justification that it remains bounded under the stated hypothesis (n−i)(n−k)/n > 1, which could affect the numerical sharpness of the constant e^{−1/12}/(4√2)."}],"tokens_in":1537,"tokens_out":565,"duration_ms":43613,"standing_objections":[]},"desk_editor":{"model":"grok-4.3","letter":"The main results are a lower bound of k/n on the probability that the hypergeometric count meets or exceeds its mean when n is at least 8k, and a second bound that scales with sqrt(Var(H)) under conditions keeping the mean away from the edges and the sampling fraction small. The auxiliary upper bound on tail conditional expectation and lower bound on mean absolute deviation are presented as separate contributions and appear to support the main inequalities directly from distribution properties and standard tools like variance calculations or basic inequalities. No fitted parameters or circular steps show up in the claims. The explicit constants and the variance form are the clearest additions relative to usual hypergeometric concentration results. The conditions are stated clearly, which is helpful for knowing the regime. The main limitation is how restrictive those conditions are: n ≥ 8k forces a small sampling fraction, and the second bound needs 1 ≤ E(H) ≤ min(i,k)-1 plus (n-i)(n-k)/n >1. These avoid boundary problems but exclude many practical cases where k is comparable to n/2 or the mean sits near zero or the upper limit. The derivations seem internally consistent from the abstract and stress-test notes, with no visible contradictions or post-hoc adjustments. This work is aimed at people who need concrete, non-asymptotic tail lower bounds for hypergeometric sampling in discrete probability or applications like finite-population inference. It is narrow but the explicit forms and auxiliaries give it value for someone already working in that area. I would send it to peer review because the claims are grounded enough and the auxiliaries add something reusable, even if revisions will likely tighten the conditions or add examples.","headline":"The paper gives two explicit lower bounds on P(H ≥ E(H)) for hypergeometric H, one simple and one variance-linked, plus some auxiliary tail and MAD bounds that look independently usable.","tokens_in":2377,"tokens_out":413,"would_cite":false,"duration_ms":16993,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":{"model":"grok-4.3","evidence":[{"relation":"echoes","rs_module":"IndisputableMonolith/Foundation/RealityFromDistinction (8-tick period)","rs_theorem":"reality_from_one_distinction","paper_passage":"P(H≥E(H))≥k/n , when n≥8k"},{"relation":"unclear","rs_module":"IndisputableMonolith/Cost/FunctionalEquation","rs_theorem":null,"paper_passage":"E(|H−ik/n|)≥ e^{-1/8}/(2√2) ⋅ √((n−1)/n) ⋅ √Var(H)"}],"headline":"Hypergeometric tail lower bounds with n≥8k condition; 8 appears but no J-cost, φ-ladder or RS forcing chain","alignment":"orthogonal","rationale":"Paper proves P(H ≥ E(H)) ≥ k/n for hypergeometric H when n≥8k (Theorem 1.1) and a MAD-based bound under E(H)∈[1,min{i,k}−2] and (n−i)(n−k)/n>1 (Theorem 1.2), using tail conditional expectation and mean absolute deviation auxiliaries. The constant 8 echoes the RS 8-tick period (reality_from_one_distinction, AlexanderDuality for D=3 via 2^D=8), yet the machinery is classical probability (stochastic orders, Robbins-Stirling, Ehm TV bound) with no J-cost, cosh-cost, ratio symmetry, golden-ratio identities or parameter-free constant derivations. Domain is discrete probability inequalities; RS has no opinion.","tokens_in":51195,"confidence":"moderate","tokens_out":372,"duration_ms":30345,"cache_read_input_tokens":128,"cache_creation_input_tokens":0},"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.3","headline":"Hypergeometric random variables satisfy P(H ≥ E(H)) ≥ k/n when the population size n is at least eight times the sample size k.","keywords":["hypergeometric distribution","tail bounds","lower bounds","sampling without replacement","mean absolute deviation","conditional expectation","variance bounds","probability inequalities"],"falsifier":"A concrete numerical counterexample with n = 8k, any valid i, where direct enumeration shows P(H ≥ E(H)) < k/n would disprove the first claim; similarly, any triple satisfying the second set of conditions where the displayed inequality fails would disprove the refined bound.","tokens_in":2708,"feed_emoji":"","tokens_out":865,"duration_ms":44082,"temperature":0.7,"pith_summary":"The paper proves explicit lower bounds on the probability that a hypergeometric random variable meets or exceeds its expectation. Hypergeometric variables arise in sampling without replacement, so these bounds quantify how likely it is to see at least the average number of successes under finite-population constraints. One result gives the simple guarantee of at least k/n whenever n is eight or more times k. A second result supplies a tighter expression involving the variance, valid when the expectation lies in a restricted interior range and a mild inequality on the remaining population holds. Supporting lemmas bound the tail conditional expectation from above and the mean absolute deviation from below.","feed_headline":"Hypergeometric P(H ≥ mean) ≥ k/n when n ≥ 8k","feed_subtitle":"Explicit lower bounds are given for the probability that the number of marked items in sampling without replacement meets or exceeds its own","key_machinery":"Lower bounds on the hypergeometric upper tail probability obtained via an auxiliary upper bound on the tail conditional expectation and a lower bound on the mean absolute deviation.","core_discovery":"Let n and k be positive integers with n ≥ k, and let H count the black marbles drawn in a sample of size k from an urn containing i black marbles out of n total. Then P(H ≥ E(H)) ≥ k/n whenever n ≥ 8k. Moreover, when 1 ≤ E(H) ≤ min{i, k} − 1 and (n − i)(n − k)/n > 1, the probability satisfies P(H ≥ E(H)) ≥ (e^{-1/12}/(4√2)) ⋅ √((n − 1)/n) ⋅ √Var(H) / (1 + √(1 + ((n − 1)/(n − k)) ⋅ Var(H))).","pith_inferences":["The same technique of controlling the tail conditional expectation could be tested on the binomial distribution to see whether analogous constants emerge.","Relaxing the n ≥ 8k threshold slightly might still preserve a comparable fractional lower bound, which could be checked by exhaustive computation for moderate k.","The explicit variance form suggests a possible route to concentration inequalities that remain valid even when the population is only modestly larger than the sample.","These lower bounds on the median-crossing probability may help calibrate error rates in combinatorial algorithms that rely on hypergeometric sampling."],"forward_implications":["When the sample size is at most one-eighth the population size, the chance of observing at least the expected number of successes is bounded below by the sampling fraction k/n.","Under interior-mean conditions the probability can be bounded from below by a multiple of the standard deviation scaled by a simple function of the variance.","An upper bound on the conditional expectation of H given H ≥ E(H) holds for the hypergeometric family.","A lower bound on the mean absolute deviation of H is available for the same family.","The auxiliary bounds may be applied independently to other tail or deviation analyses of sampling-without-replacement distributions."],"fun_headline_variants":["Hypergeometric tails bounded by k/n when n >= 8k","Hypergeo P(H >= mean) at least k/n for n >= 8k","Hypergeometric distribution tail lower bound when n >= 8k","Variance based lower bound for hypergeometric tails","Hypergeometric P(H >= E(H)) >= k/n under n >= 8k"],"cache_read_input_tokens":2112,"weakest_assumption_plain":"The condition n ≥ 8k for the first bound, and the restrictions 1 ≤ E(H) ≤ min{i,k} − 1 together with (n − i)(n − k)/n > 1 for the second bound.","fun_headline_variants_meta":{"raw":{"variants":["Hypergeometric tails bounded by k/n when n >= 8k","Hypergeo P(H >= mean) at least k/n for n >= 8k","Hypergeometric distribution tail lower bound when n >= 8k","Variance based lower bound for hypergeometric tails","Hypergeometric P(H >= E(H)) >= k/n under n >= 8k"]},"model":"grok-4.3","cost_usd":0.012082,"raw_usage":{"total_tokens":5321,"prompt_tokens":761,"num_sources_used":0,"completion_tokens":83,"cost_in_usd_ticks":120824500,"prompt_tokens_details":{"text_tokens":761,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":4477,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":761,"tokens_out":83,"duration_ms":70809,"temperature":1.0,"reasoning_tokens":4477,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-05-16T14:17:16.188981+00:00","model_set":{"reader":"grok-4.3"},"falsifier":"A concrete numerical counterexample with n = 8k, any valid i, where direct enumeration shows P(H ≥ E(H)) < k/n would disprove the first claim; similarly, any triple satisfying the second set of conditions where the displayed inequality fails would disprove the refined bound.","supporting_citations":[],"review_version":1}