← Research index

Publication records

100 of 170 catalogue documents have a recorded YouTube link; 70 have no YouTube link. A recorded link may refer to a video or a community post.

170 documents · 170 GitHub catalog entries · 170 GERO · 170 LinkedIn · 170 Hugging Face · 144 Zenodo records.

Distinct documents are ordered newest first. Counts describe documents, not distinct defects. Historical collections and additional posts are preserved but are not counted as new documents.

Archive coverage was reconciled on 16 September 2026: the same 99 earlier publication IDs are available on GitHub, GERO, Hugging Face and Zenodo. One historical Zenodo collection is counted separately. These archive additions do not represent new numerical experiments.

The Collatz ribbon atlas and Ingenium narrated tour are available on GERO/GitHub, with files on Hugging Face, a Zenodo2.0.0 archive and YouTube publication. The full conjecture remains unproved.


DatePublication on GEROGitHubLinkedInZenodoHugging Face
2026-10-25SciPy exact-test boundaries: floating-point ties can change p-valuesGitHubLinkedInZenodoReport
2026-10-25Tax-Calculator: Schedule D and AMT rule alignmentGitHubLinkedInZenodoReport
2026-10-25Luxon: skipped midnights and clamped month-end differencesGitHubLinkedInZenodoReport
2026-10-25decimal.js: rounding boundaries in log2 and log10GitHubLinkedInZenodoReport
2026-10-25ethers: negative FixedNumber rounding can change integral valuesGitHubLinkedInZenodoReport
2026-10-24pvlib: clock time and elapsed time diverge on DST-transition daysGitHubLinkedInZenodoReport
2026-10-24lifelines: keep fixed-time variance constant between observed eventsGitHubLinkedInZenodoReport
2026-10-24statsmodels: continuity correction and power inversion regressionsGitHubLinkedInZenodoReport
2026-10-24Formula.js: seven proposed fixes for blanks, financial formulas and datesGitHubLinkedInZenodoReport
2026-10-24Two reported tax-model discrepancies corrected in PolicyEngine-US 2.60.3GitHubLinkedInZenodoReport
2026-10-23PoseLib: valid camera geometry lost through coordinate singularitiesGitHubLinkedInZenodoReport
2026-10-23OpenCV RQ decomposition: a scale-aware zero checkGitHubLinkedInZenodoReport
2026-10-23OpenCV iterative PnP: solve near the object instead of a distant originGitHubLinkedInZenodoReport
2026-10-23OpenCV SQPnP: conditioning across units and coordinate originsGitHubLinkedInZenodoReport
2026-10-23OpenCV Rodrigues: accuracy at small angles and near piGitHubLinkedInZenodoReport
2026-10-23OpenCV RANSAC: preserve the camera-pose initial guessGitHubLinkedInZenodoReport
2026-10-22Augmented Black–Litterman: historical prior must retain the joint stateGitHubLinkedInZenodoReport
2026-10-22Riskfolio Bayesian Black–Litterman: equilibrium weights use excess returnsGitHubLinkedInZenodoReport
2026-10-22Missing observations are not zero returns: a PyPortfolioOpt correctionGitHubLinkedInZenodoReport
2026-10-22Market-neutral portfolios rejected by a fully-invested risk boundGitHubLinkedInZenodoReport
2026-10-22Black–Litterman weights: excess returns and the singular fallbackGitHubLinkedInZenodoReport
2026-10-22Provably fair does not mean a constant expected returnGitHubLinkedInZenodoReport
2026-10-22Historical IRS Direct File source: a joint-filer credit and a rounding witnessGitHubLinkedInZenodoReport
2026-10-22PolicyEngine UK: Child Benefit charge rounding now corrected upstreamGitHubLinkedInZenodoReport
2026-10-22web3-ethereum-defi: zero depth inside an active liquidity intervalGitHubLinkedInZenodoReport
2026-10-22penaltyblog: positive expected log growth at a ruin boundaryGitHubLinkedInZenodoReport
2026-10-22shin: a default probability result can hide non-convergenceGitHubLinkedInZenodoReport
2026-10-22uniswap-python: integer price flooring and odd-tick liquidity, with a public patchGitHubLinkedInZenodoReport
2026-10-22Curo: fortnightly payments collapse onto monthly datesGitHubLinkedInZenodoReport
2026-10-22actuarialmath: the variance selector rejected before its own branchGitHubLinkedInZenodoReport
2026-10-22bond_pricing: perpetuity-rate inversion fixed in release 1.0.1GitHubLinkedInZenodoReport
2026-10-22Samsung ONE: Gather slices and Mean axes changed by compiler passesGitHubLinkedInZenodoReport
2026-10-22Empyrical: constant returns and floating-point Sharpe ratiosGitHubLinkedInZenodoReport
2026-10-22Riskfolio BOP: correcting the shrinkage target denominatorGitHubLinkedInZenodoReport
2026-10-22PyPortfolioOpt: a ticker-order mismatch breaks the allocation budgetGitHubLinkedInZenodoReport
2026-10-17A mid-month rate increase disappears from chainladder on-level factorsGitHubLinkedInZenodoReport
2026-10-17A missing drawdown and a positive score for a loss in quantstatsGitHubLinkedInZenodoReport
2026-10-17A shared constant flips two convolution biases in coremltoolsGitHubLinkedInZenodoReport
2026-10-07Frappe Lending quarterly day counts: three days instead of 91, and the v16.6.1 correctionGitHubLinkedInZenodoReport
2026-10-04A zero guarantee with a minus-200 outcome: penaltyblog partial hedgingGitHubLinkedInZenodoReport
2026-10-03A one-payment loan breaks IRR: an incorrect derivative in loan-calculatorGitHubLinkedInZenodoReport
2026-10-03When gross-up does not return the requested net: two loan-calculator defectsGitHubLinkedInZenodoReport
2026-09-30actuarialmath whole_life_annuity ignores benefit b at zero interestGitHubLinkedInZenodoReport
2026-09-30actuarialmath Uniform shortcuts use the wrong conditioning age and reverse limited-life weightsGitHubLinkedInZenodoReport
2026-09-27amortization 3.0.0: small positive rates can erase interest or alter the rounded paymentGitHubLinkedInZenodoReport
2026-09-24MLX arctan: finite weighted derivatives collapse to zero after intermediate overflowGitHubLinkedInZenodoReport
2026-09-24actuarialmath direct endowment second moments omit the benefit powerGitHubLinkedInZenodoReport
2026-09-23bond_pricing 0.7.3 ignores immediate-start annuity paymentsGitHubLinkedInZenodoReport
2026-09-23FinancePy Merton jump pricing: a zero probability can erase an option valueGitHubLinkedInZenodoReport
2026-09-23Apache Commons Statistics: finite bounds, NaN probabilities, and a changed decisionGitHubLinkedInZenodoReport
2026-09-21MLX log10: finite gradients, nonfinite updates under extreme loss scalingGitHubLinkedInZenodoReport
2026-09-20actuarialmath continuous annuity: dropping elapsed years changes a value and a decisionGitHubLinkedInZenodoReport
2026-09-20FinancePy floating lookback calls: a finite price becomes unavailableGitHubLinkedInZenodoReport
2026-09-20nntrainer CentroidKNN chooses an unobserved class over an exact centroid matchGitHubLinkedInZenodoReport
2026-09-19Same Itô integral, fewer operations: a reproducible computational study
43-second English video		GitHubLinkedInZenodoReport
2026-09-19MLX erf: a finite gradient becomes infinity before damping is applied
32-second English video		GitHubLinkedInZenodoReport
2026-09-19Endowment variance scales the benefit twice in actuarialmath
32-second English video		GitHubLinkedInZenodoReport
2026-09-19Strata: a zero-volatility Hessian prevents custom-model option pricing
English video (36 seconds)		GitHubLinkedInZenodoReport
2026-09-18
Updated 2026-09-19		PyLoan replaces an explicit 100 payment with a 919.10 final payment
v1.2: accepted fix in official 0.7.3, independently verified; first video pending		GitHubLinkedInZenodoReport
2026-09-18Strata returns zero implied volatility for a positive option price
33-second video		GitHubLinkedInZenodoReport
2026-09-18MLflow NDCG: irrelevant retrievals score above zero after a document-ID collision
34-second English video		GitHubLinkedInZenodoReport
2026-09-17MLX rsqrt: finite directional derivatives lost to an intermediate ratio
32-second English overview		GitHubLinkedInZenodoReport
2026-09-17FinancePy BondZero: verifying the upstream fix for a 10,000-fold duration error
33-second English overview		GitHubLinkedInZenodoReport
2026-09-17actuarialmath Beta: a missing survival factor changes expected insurance benefits
29-second English overview		GitHubLinkedInZenodoReport
2026-09-17FinancePy adjusted binomial: invalid support points produce negative probabilities and shift the mean
34-second English overview		GitHubLinkedInZenodoReport
2026-09-17FinancePy: premature series truncation produces an invalid option price
31-second English overview		GitHubLinkedInZenodoReport
2026-09-17nntrainer SplitLayer leaves later channels unwrittenGitHubLinkedInZenodoReport
2026-09-16FinancePy FXForward converts today's NPV using a future exchange rateGitHubLinkedInZenodoReport
2026-09-16Apple MLX sigmoid derivatives lose a representable positive tailGitHubLinkedInZenodoReport
2026-09-16FinancePy Vasicek pricing loses the small-mean-reversion limit
34-second English overview		GitHubLinkedInZenodoReport
2026-09-16Where We Stand on the Collatz Conjecture
Research in progress; full conjecture unproved. Ribbon atlas · Narrated tour		GitHubLinkedInZenodo · ZenodoReport
2026-09-16FinancePy annuity pricing depends on prior payment callsGitHubLinkedInZenodoReport
2026-09-15FinancePy CIR pricing loses its finite range and small-volatility limitGitHubLinkedInZenodoReport
2026-09-15MLX median overflows while averaging finite central valuesGitHubLinkedInZenodoReport
2026-09-15ConstantForce whole-life insurance ignores the benefit amountGitHubLinkedInZenodoReport
2026-09-15FinancePy Merton calibration depends on the monetary unitGitHubLinkedInZenodoReport
2026-09-14MLX CPU quantized matmul: avoidable accumulation error in float16 and bfloat16
Evidence archive		GitHubLinkedInZenodoReport
2026-09-14nntrainer DivideLayer: finite forward values, incorrect denominator gradients
Evidence archive		GitHubLinkedInZenodoReport
2026-09-14QuantLib: deterministic ITM probabilities and cap/floor deltas
Evidence archive		GitHubLinkedInZenodoReport
2026-09-14actuarialmath whole-life variance squares the mean twiceGitHubLinkedInZenodoReport
2026-09-14FinancePy cash-settled swaption annuity counts N−1 payments instead of NGitHubLinkedInZenodoReport
2026-09-13A constant forward, a NaN gradient: nntrainer PowLayer at exponent zeroGitHubLinkedInZenodoReport
2026-09-13FinancePy misprices three FX digital payout-currency pathsGitHubLinkedInZenodoReport
2026-09-13FinancePy Act/365L can fail when an optional date is omittedGitHubLinkedInZenodoReport
2026-09-13Integer overflow reverses MLX isclose and allclose resultsGitHubLinkedInZenodoReport
2026-09-13PyXIRR RATE can return an unfinished calculation as a resultGitHubLinkedInZenodoReport
2026-09-13FinancePy vector pricing uses the last expiry for every optionGitHubLinkedInZenodoReport
2026-09-13Parallel vectors produce nonzero cross products in Apple MLXGitHubLinkedInZenodoReport
2026-09-13Apple MLX: integer norms can become zero, negative, or NaNGitHubLinkedInZenodoReport
2026-09-13FinancePy mortgage payments fail at zero and very small ratesGitHubLinkedInZenodoReport
2026-09-13Apple MLX: small matrices lose finite log determinantsGitHubLinkedInZenodoReport
2026-09-13Finite ONNX Normalizer inputs overflow, underflow, or violate the output contractGitHubLinkedInZenodoReport
2026-09-13MLX LayerNorm: an optional unit weight changes bias precisionGitHubLinkedInZenodoReport
2026-09-12Finite FP16 Inputs Become Zero in MLX GroupNormGitHubLinkedInZenodoReport
2026-09-10One FP16 Batch Corrupts MLX BatchNorm Running VarianceGitHubLinkedInZenodoReport
2026-09-10Finite Gradients Become Zero: MLX Gradient ClippingGitHubLinkedInZenodoReport
2026-09-10Same Ratio, Wrong Derivative: MLX Division at Extreme ScalesGitHubLinkedInZenodoReport
2026-09-10Scale Changes the Gradient: MLX arctan2GitHubLinkedInZenodoReport
2026-09-10Finite Values, Missing Derivatives: MLX arcsinh and arccoshGitHubLinkedInZenodoReport
2026-09-10A Nonzero Derivative Becomes Zero: MLX expm1GitHubLinkedInZenodoReport
2026-09-10Same Inputs, Wrong Weights: MLX logcumsumexp Gradients at Large OffsetsGitHubLinkedInZenodoReport
2026-09-10Missing Curvature: MLX logcumsumexp at a Zero Incoming GradientGitHubLinkedInZenodoReport
2026-09-10Zero Mask, Missing Gradient: MLX block_masked_mmGitHubLinkedInZenodoReport
2026-09-09A Missing Transpose in MLX Quantized GradientsGitHubLinkedInZenodoReport
2026-09-09Sorted Indices, Wrong Gradients: MLX GatherGitHubLinkedInZenodoReport
2026-09-09Orthogonal Is Not Symmetric: MLX Hadamard GradientsGitHubLinkedInZenodoReport
2026-09-09A Constant, a Polynomial, a NaN: MLX power Derivatives at ZeroGitHubLinkedInZenodoReport
2026-09-09An Unused Value Gets a Gradient: MLX Masked AssignmentGitHubLinkedInZenodoReport
2026-09-09A Missing Gradient in MLX Block UpdatesGitHubLinkedInZenodoReport
2026-09-09When SAME Padding Makes Average-Pooling Backward Skip WindowsGitHubLinkedInZenodoReport
2026-09-09When Dropout Sends Two Gradients to One InputGitHubLinkedInZenodoReport
2026-09-09Finite Loss, Wrong Gradient: Gaussian NLL in MLXGitHubLinkedInZenodoReport
2026-09-09Same Gradients, Half the Step: MLX Muon and Convolution ShapesGitHubLinkedInZenodoReport
2026-09-09A Real Loss, a Reversed Gradient: Complex Autodiff in MLXGitHubLinkedInZenodoReport
2026-09-09A Smooth Polynomial, a NaN Hessian: MLX cumprod at ZeroGitHubLinkedInZenodoReport
2026-09-09ONNX ReferenceEvaluator turns infinite ReduceLogSumExp slices into NaNGitHubLinkedInZenodoReport
2026-09-09ONNX ReferenceEvaluator misidentifies BatchNormalization opset-9 modesGitHubLinkedInZenodoReport
2026-09-09ONNX ReferenceEvaluator gathers unsorted Unique slices from the wrong axisGitHubLinkedInZenodoReport
2026-09-09ONNX ReferenceEvaluator rejects valid GatherElements subshapesGitHubLinkedInZenodoReport
2026-09-07When Cosine and LayerNorm Break Their ContractsGitHubLinkedInZenodoReport
2026-09-07Research acceleration needs an outcome-weighted evidence contractGitHubLinkedInZenodoReport
2026-09-07A machine-checked proof still needs a statement contractGitHubLinkedInZenodoReport
2026-09-07When Causal Attention Sees the FutureGitHubLinkedInZenodoReport
2026-09-07ONNX reference LpNormalization uses a signed sum for p=1GitHubLinkedInZenodoReport
2026-09-07ONNX GroupNormalization reverses channel/group divisibility in its documentationGitHubLinkedInZenodoReport
2026-09-07When Small Losses and Gradients DisappearGitHubLinkedInZenodoReport
2026-09-07Finite Inputs, Non-finite Results: An Activation Stability AuditGitHubLinkedInZenodoReport
2026-09-06ONNX reference LogSoftmax and cross-entropy return infinity for finite representable resultsGitHubLinkedInZenodoReport
2026-09-06Dinero.js misallocates safe integers after an unsafe intermediate productGitHubLinkedInZenodoReport
2026-09-06Dinero.js toDecimal accepts non-decimal bases and returns plausible wrong stringsGitHubLinkedIn · LinkedInZenodoReport
2026-09-06Zero Observed False Positives Is Not a Zero False-Positive RateGitHubLinkedInZenodoReport
2026-09-06When the reference also fails: a reproducible ONNX numerical snapshotGitHubLinkedInZenodoReport
2026-09-06ReduceMean Without an Axis Could Not InitializeGitHubLinkedInZenodoReport
2026-09-05ONNX Runtime float16 QuantizeLinear truncates instead of roundingGitHubLinkedInZenodoReport
2026-09-05MoneyPHP loses largest-remainder ordering above float precisionGitHubLinkedIn · LinkedInZenodoReport
2026-09-05Apache Fineract nop does not invert pmt for non-zero ratesGitHubLinkedIn · LinkedInZenodoReport
2026-09-05When Reciprocal Multiplication Changes the Answer: Two Swift Numerics Contract TestsGitHubLinkedInZenodoReport
2026-09-05The Fast Path Put a Complex Number in the Wrong Quadrant: A Swift Numerics Sign AuditGitHubLinkedIn · LinkedInZenodoReport
2026-09-05How Numerical Contracts Break: Five Cases in Swift Numerics and nntrainerGitHubLinkedInZenodoReport
2026-09-04MLX affine quantization: a documentation/runtime contract mismatchGitHubLinkedInZenodoReport
2026-09-04What Simple Mathematical Invariants Reveal About AI SoftwareGitHubLinkedInZenodoReport
2026-09-04A KL-Divergence Backward Pass That Changes Its Own TargetGitHubLinkedInZenodoReport
2026-09-04Four Loss Layers, One Missing Scale: A Mixed-Precision Contract FailureGitHubLinkedInZenodoReport
2026-09-04Fractional Projective-Gap Inequalities for Harmonic Gaussian ChaosesGitHubLinkedInZenodoReport
2026-09-03A one-line contradiction in Swift Numerics' tie-to-zero documentationGitHubLinkedIn · LinkedInZenodoReport
2026-09-03An exact fifth root exposes a one-ULP accuracy gap in Swift NumericsGitHubLinkedIn · LinkedIn · LinkedInZenodoReport
2026-09-03A scale-dependent zero point silently corrupts UINT8 affine quantization in nntrainerGitHubLinkedIn · LinkedInZenodoReport
2026-09-02Where Mathematics Meets AI Reliability: The Problems Frontier Labs Are Naming ThemselvesGitHubLinkedInZenodoReport
2026-09-02AI Can Generate New Mathematics. Who Certifies the Proof?GitHubLinkedInZenodoReport
2026-09-02The Most Important AI Feature Is Not an Answer. It Is a Boundary.GitHubLinkedInZenodoReport
2026-09-02A Constant Left Open in 1973, and What It Took to Close ItGitHubLinkedInZenodoReport
2026-09-02A Retraction Log: Six Collatz Claims and What Killed ThemGitHubLinkedInZenodoReport
2026-09-02The 60-Pair Smoke Benchmark, With the Data and the CommandGitHubLinkedInZenodoReport
2026-09-02Hidden reasoning is an observability problemGitHubLinkedInZenodoReport
2026-09-02Attribution graphs still need a semantic contractGitHubLinkedInZenodoReport
2026-09-02Research mathematics needs proof assurance, not a single judge scoreGitHubLinkedInZenodoReport
2026-09-02Monitoring reasoning is a stability problemGitHubLinkedInZenodoReport
2026-09-02Research mathematics is a generator–verifier–reviser control loopGitHubLinkedInZenodoReport
2026-09-02Self-correction is two problems, not oneGitHubLinkedInZenodoReport
2026-09-02A gold-level benchmark result is not a research reliability certificateGitHubLinkedInZenodoReport
2026-09-14FinancePy BAW pricing divides by zero at a zero rateGitHubLinkedInZenodoReport
2026-09-14FinancePy bond principal does not scale with face amountGitHubLinkedInZenodoReport
2026-09-14ONNX LpNormalization loses finite vectors through intermediate rangeGitHubLinkedInZenodoReport
2026-09-14ONNX ReduceL2 loses finite norms at extreme scalesGitHubLinkedInZenodoReport
2026-09-14ONNX ReduceMean overflows before reaching a finite meanGitHubLinkedInZenodoReport
2026-09-14GERO financial calculation checks: a scoped Apache Fineract case studyGitHubLinkedInZenodoReport
2026-09-14Spark-X2.5-1.7B: exact fee bases and half-up rounding studyGitHubLinkedInZenodoReport
2026-09-14Obstructions to Low-Dimensional Natural Matrix Interpretations for a Mixed-Base Collatz Rewrite SystemGitHubLinkedInZenodoReport
2026-09-14Axisymmetric p-harmonic modes in circular cones: nodal classification and two-term asymptoticsGitHubLinkedInZenodoReport
2026-09-14pyliferisk geometric valuations discard supplied mortality assumptionsGitHubLinkedInZenodoReport

Machine-readable publication records · Prepared archive checksums