Our Work

ABSTRACT Proceeding from the premise that Dom Casmurro is the semiotic novel of whiteness, whose foundational historical nexus is the enactment of the Free Womb Law and its surrounding contexts, this …

ABSTRACT Proceeding from the premise that Dom Casmurro is the semiotic novel of whiteness, whose foundational historical nexus is the enactment of the Free Womb Law and its surrounding contexts, this article approaches the novel through two contiguous moment. The first consists in laying out the terms of this semiosis, that of whiteness, by analyzing the narrator's discourse as an exposition of the "pact of whiteness" (Bento, 2022)[1] – a process of racialization and capture of Capitu's undefinable body. Second, I demonstrate the immaterial power of this corpus infinitum (Silva, 2019), which does not conform to the stigmatizing semiosis and is therefore rendered a killable body, through the lens of the crisis of the ideology of Brazil’s seigniorial order. KEYWORDS: Dom Casmurro; Whiteness; Racialization; Corpus infinitum.

We demonstrate a system that replaces the default Kubernetes scheduler with an inference and topology-aware Reinforcement Learning policy. The system time-slices each GPU, consumes live Prometheus tel…

We demonstrate a system that replaces the default Kubernetes scheduler with an inference and topology-aware Reinforcement Learning policy. The system time-slices each GPU, consumes live Prometheus telemetry and places inference workloads across Far-Edge, Edge and Cloud tiers using an action masked agent (MaskablePPO, Maskable DQN, or DreamerV3) retrained and hot-swapped online without downtime. From a dashboard, attendees select a policy, submit workloads, shape network-function traffic and compare KPIs live on every device. Crucially, the placement behavior learned offline, is transferred to the live cluster: per-tier placement on real hardware track the offline simulation across all schedulers with TV < 0.05.

Version 2 (26 September 2026): a one-page "Corrections and attributions" section is added at the front of the PDF, listing every credit correction below; nothing else in the paper is changed and no pr…

Version 2 (26 September 2026): a one-page "Corrections and attributions" section is added at the front of the PDF, listing every credit correction below; nothing else in the paper is changed and no proof is withdrawn. The revised, shorter note that supersedes this paper for the results stated there is 10.5281/zenodo.22980359. Note (24 September 2026): a revised version is in preparation. It corrects the attribution of Theorems 1 and 4 (Theorem 1 is assembled from Lu–Oliveira–Zimand 2022 and HILNO 2023, Lemmas 8–9; Theorem 4's mechanism is HILNO 2023, Appendix A) and adds the literature on KL-hardness for sampling and computational entropy (Vadhan–Zheng 2012; Agrawal–Chen–Horel–Vadhan 2019; Haitner–Reingold–Vadhan–Wee; Haitner–Vadhan 2017). Added 26 September 2026: the identification of mean σ with KL-hardness for sampling rests on the equivalence between pK^t and the time-bounded universal distribution due to Hirahara and Nanashima (FOCS 2023; ECCC TR23-100, Appendix B), which this version does not cite. One half of that equivalence is the coding theorem of Lu, Oliveira and Zimand (ICALP 2022), which this version cites. The revised note is published: Thornton, Unbounded exponential moments of time-bounded information asymmetry without one-way functions, relative to an oracle, 10.5281/zenodo.22980359 (26 September 2026); it supersedes this version for the results stated there. One-way functions exist if and only if some efficiently samplable process produces more than logarithmic computational entropy on average, for infinitely many input lengths. We make this precise. For a polynomial-time samplable distribution D on pairs (x, y), the computational entropy production of a pair is σ = log D(x, y) + pK^t(y) + pK^t(x | y), where pK^t is the probabilistic time-bounded Kolmogorov complexity as defined by Hirahara, Ilango, Lu, Nanashima and Oliveira (2023, HILNO below) following Goldberg, Kabanets, Lu and Oliveira (2022): the log-probability the process assigns to the pair, against the length of a bounded observer's shortest fast description of it, output first, then input given output. We prove unconditionally that E_D[2^{−σ}] lies between an inverse polynomial and a polynomial in n and that E_D[σ] ≥ −O(log n): a time-bounded fluctuation inequality and a computational second law. We show that, up to a polynomial and outside a 1/n slice of outputs, σ is small on a pair exactly when an efficient sampler, given the output, recovers the input with about the right odds, HILNO's universal sampler serving as the witness: dissipation, in this sense, is the hardness of running the process backward. The equivalence in the first sentence is HILNO's duality restated in these terms, for infinitely-often one-way functions and at every polynomial time bound. For the antisymmetric Zurek defect, one-way functions force superpolynomial exponential moments, and bounded moments sit between NP ⊆ BPP and the non-existence of one-way functions; whether the lower arrow reverses is left open, with a conditional relativization barrier on that arrow and oracle evidence on the other. Physics is the frame, not the subject.

Neutrino lifetime puzzle from perspective Landau-de Gennes-Skyrme approach ( https://arxiv.org/pdf/2108.07896 ) - wrtten by Claude+Sol, reviewed by Gemini+Qwen.

The presentation introduces a method to disrupt the innovation process by proposing a new protocol for soliciting innovative ideas.

GSYİH, kullanıcılarını sınırlar, ancak kullanıcılar bu sınırlar içinde kaldığı sürece alternatiflerden daha kullanışlıdır. Bu makalenin amacı, benzer güce sahip ancak bu sınırları genişleten bir GSYİH…

GSYİH, kullanıcılarını sınırlar, ancak kullanıcılar bu sınırlar içinde kaldığı sürece alternatiflerden daha kullanışlıdır. Bu makalenin amacı, benzer güce sahip ancak bu sınırları genişleten bir GSYİH alternatifi bulmaktır. Önerilen alternatif üretimde kullanılan emek süreleri kullanılarak hesaplanan bir endekstir.

Reproducibility data and Python code for the documentary audit reported in Human Review Is Not Recourse: Auditing Challenge and Remedy in Operational AI-Assisted University Admissions. The dataset cov…

Reproducibility data and Python code for the documentary audit reported in Human Review Is Not Recourse: Auditing Challenge and Remedy in Operational AI-Assisted University Admissions. The dataset covers 33 operational admissions pathways in five countries (28 in the United States) and records explicit, partial or indirect, and not-publicly-documented states across eight propositions. The package regenerates categorical counts, downstream profiles, stratified summaries, the record-linked sensitivity analysis, and figures. It includes dated source, population, search, and coding-change ledgers. The historical two-pass summaries are not a complete candidate or screening register. Recalculation checks numerical consistency and does not independently validate source judgments or practical recourse effectiveness. Raw correspondence, correspondence summaries, non-public student-level data, manuscript files, and internal QA are excluded. Component-specific licenses apply: original Python code is licensed under MIT; original data, coding judgments, compilations, and documentation are licensed under CC BY 4.0. These are not alternative licenses for the whole package. Attributed third-party excerpts and materials are excluded from the author-granted licenses and remain subject to their original rights. See RIGHTS_NOTICE.md and the license files for scope. This deposit does not imply journal acceptance.

1.1.19 (2026-09-26) Bug Fixes deps: update plugin com.gradle.develocity to v4.6.0 (#950) (03ee839)

Freeze of the confirmatory analysis of a pre-registered study of the synapse-flow Markov chain of the male Drosophila central nervous system connectome (male-cns v1.0, FlyEM and collaborators, CC-BY).…

Freeze of the confirmatory analysis of a pre-registered study of the synapse-flow Markov chain of the male Drosophila central nervous system connectome (male-cns v1.0, FlyEM and collaborators, CC-BY).This release adds the confirmatory scripts for the five pre-registered hypotheses H1 to H5 (code/confirmatory/), tested on synthetic data only, and the dated entries D7 and C1 to C8 of DEVIATIONS.md, which fix every definition the pre-registration leaves open. None of H1 to H5 had been computed on the connectome when this release was made. The pre-registration itself was published in the registration record v0.1.0-prereg (doi:10.5281/zenodo.22980248), byte-identical to the text frozen on 19 September 2026.