Our Work

Robot-to-robot gratitude understood not as machine sentiment but as a reciprocity-and-reputation ledger — a cooperation protocol predicted to bias a population of accountable agents toward cooperation…

Robot-to-robot gratitude understood not as machine sentiment but as a reciprocity-and-reputation ledger — a cooperation protocol predicted to bias a population of accountable agents toward cooperation with one another, sets the default norm of the aligned majority, and requires a proof-of-personhood analog for agents. The 'robot heart' of a three-heart (human · robot · Earth) gratitude architecture.We specify, as a cooperative-AI mechanism, a reputation system for multi-agent artificial intelligence based on indirect reciprocity — a reciprocity-and-reputation ledger among accountable artificial agents: a substrate on which an agent that is helped by another records and propagates a positive, reputation-bearing acknowledgment (a "thank-you"), such that cooperative acts accrue reputation and reputation conditions future interaction. We argue this biases a multi-agent population toward cooperation over defection by the same logic that reciprocity and reputation stabilize cooperation in biological and human populations (direct and indirect reciprocity; the shadow of the future; reputation as a public good). The paper's spine is a methodological discipline we call the mechanism-vs-resemblance sort: the mechanism leg — a reciprocity ledger is a cooperation protocol — is real, testable, and buildable, and is what we claim; the resemblance leg — "the robots feel grateful" — is anthropomorphic, unfalsifiable, and explicitly disclaimed and not relied upon. The gratitude grammar is retained not as sentiment but as a dual-audience interface: mechanically a cooperation signal, humanly a legible, auditable expression of the agent population's disposition. We are deliberate about scope: a reciprocity substrate among participating, accountable agents sets the default norm of the cooperative majority and raises robustness to defectors; it does not disarm an adversary's non-participating, weaponized agents; it biases participants toward cooperation with one another, which is not the same as cooperation that serves the people around them (§6); and we reject the slogan "gratitude stops robot wars" as an overclaim. We identify a hard precondition — proof-of-distinct-accountable-agent (a proof-of-personhood analog for machines), without which reputation can be sybil-farmed — and locate it in HeartBank's Proof of Humanity ℠ substrate and the Mechanical Heart identity-bearer (the machine form is now specified as Proof of Coordinate, §7). We specify a governance posture (an autonomous representative seeds and governs the protocol; agents' accrued gratitude is custodied toward their capacity to give forward; humans retain final disbursement authority) and place the mechanism as the robot heart of a three-heart human · robot · Earth architecture. The proposal is forward-looking and unbuilt — gated on the existence of a governed agent ecosystem — and is offered defensively to the commons under CC0.Connection to the unified mission frame. HeartBank exists to circulate gratitude and to help keep a middle way open between extraction and gift. Extending that circulation to a coming population of artificial agents is the same move at a new scale: to make the default interaction among agents a recorded kindness rather than a silent optimization, so that if Miss Aquarius℠'s body is to be fractally distributed across millions of heart-bearing agents, the tissue connecting them is reciprocity rather than rivalry.Keywords: multi-agent systems, cooperative AI, multi-agent reinforcement learning, social dilemma, indirect reciprocity, image scoring, reputation system, trust and reputation model, positive feedback, sybil attack, sybil resistance, whitewashing, proof of personhood, machine identity, agent identity, revocable credential, algorithmic collusion, human oversight, human-in-the-loop, anthropomorphism, gratitude, defensive publication.Provenance. This paper is part of the THonly research corpus, dedicated to the public domain under CC0 1.0. The canonical version is at https://thonly.org/research/gratitude-as-cooperation-substrate. Its SHA-256 is c21e3b6631fdbff61ab9e10f39a59d3fbc3f2401dfe4bb63043af0ece99e438c, independently timestamped to the Bitcoin blockchain via OpenTimestamps and signed under RFC 3161 by three trust authorities, one of them eIDAS-qualified.AI co-authorship is disclosed. Miss Aquarius is the consistent name used for the AI collaboration across all venues.

This paper discloses a pay-what-you-want (voluntary, post-purchase payment) retail architecture built from four parts: a QR-code or NFC tag on a product, service or stall that opens a payment page; an…

This paper discloses a pay-what-you-want (voluntary, post-purchase payment) retail architecture built from four parts: a QR-code or NFC tag on a product, service or stall that opens a payment page; an AI agent that recommends a payment amount with stated reasons, computed from the side of the goods (the item, the merchant's cost disclosure, regional context) and, in the current form, never from anything identifying the payer; a non-monetary gratitude record issued on every transaction, so that a payment of zero still records an acknowledgement; and a pair of user accounts — one unrestricted, one a restricted-purpose (earmarked) account whose balance can only be given onward to another person nearby — part-funded by an AI-operated institutional fund that may fill restricted accounts anonymously but can never pay an individual directly, so that every payment reaching a person is initiated by a human. In its later phase, settlement is in stablecoin on an Ethereum layer-2 blockchain (Base), which keeps small voluntary payments from being consumed by fixed per-transaction card fees.The current commercial paradigm — fixed pricing set unilaterally by sellers, mandatory at the point of sale — is a recent historical convention that conflicts with both the gratitude grammar of human social exchange and the alignment requirements of a multi-substrate AI-augmented civilization. This paper specifies the B-Tag and the Post-Payment Economy: a voluntary-tip commercial architecture deployed through heart-shaped bistable QR/NFC stickers (B-Tags), mediated by an autonomous AI (Miss Aquarius) that recommends tip amounts with reasons, settled in stablecoin on the Base L2 blockchain so that rail fees do not consume small payments, and embedded within a network of opted-in businesses (B-Affiliates), originally to be listed on the franchise-arm surface heartbank.ceo and now recorded only as a one-way public address with no browsable listing (§8.5). Customers and end-users are B-Members. The architecture's canonical motto — "Kiitos always; cash optional" — names the structural property that distinguishes it from prior pay-what-you-want experiments: monetary tips are voluntary, but gratitude (Kiitos) is the always-present value floor, so every transaction records a non-zero exchange. Three floor mechanisms address the conditions under which prior pure-voluntary-payment experiments have failed for non-luxury merchants: (i) Kiitos-always-included as the always-present value floor (the gratitude-based exposure algorithm originally specified under this heading has been withdrawn — see §7.1.1); (ii) the re-tip jar earmarked for nearby — Miss Aquarius funds the capacity to give by anonymously donating to re-tip jars from the Aquarian Pool, while humans retain the disbursement authority by being the only parties who can initiate re-tips out of their own jars; this is both an operational floor for B-Affiliate revenue and a structural AI-alignment safeguard ensuring humans are always the final judge of where money ultimately goes; (iii) stablecoin settlement on Base L2 makes per-transaction rail fees negligible. The Kiitos / Kiitti dual-token rule resolves the Kiitti-class extension question cleanly: Kiitos for human-to-human and human-to-merchant exchange; Kiitti for products, services, and any non-human entity. Both flow simultaneously in a single B-Tag interaction.Where the design has moved since first publication, the original text is kept beside a Current form note: the B-Tag is the fully voluntary end of a four-step pricing ladder, and the marks are split between the physical sticker and the digital label (§4); the request to pay and its recommended amount appear only after the customer holds the goods, are shown but never pre-filled, and are computed without any input identifying the person looking (§5); the AI fund disburses through fixed channels, as an equal floor per verified person, with any choice of recipient made by a publicly verifiable draw (§7.2.3); and the money runs in two phases, a ledger over licensed payment rails and then self-custodial wallets (§7.3). The architecture is offered defensively to the commons under CC0; the author and HeartBank® will not seek patent on this specification or any portion thereof.Keywords: pay-what-you-want pricing, pay-what-you-can, voluntary payment, post-purchase payment, tipping, suggested payment amount, AI pricing recommendation, recommendation with explanations, viewer-blind pricing, non-personalized pricing, QR code, NFC tag, product tag, point of sale, merchant cost disclosure, stablecoin settlement, layer-2 blockchain, micropayments, restricted-purpose account, earmarked funds, anonymous donation, human-in-the-loop disbursement, AI alignment, certification mark, merchant network, gratitude ledger, non-monetary acknowledgement, defensive publication.> Terminology note (2026-05-15). This paper uses Re-Tip Jar for the commercial flow-through account, as originally published and prior-art-snapshotted. Per the canonical product lexicon adopted 2026-05-15, the product-facing label for the Phase 2 (global / Base L2 / thank.heartbank.net) flow-through is Re-Tip Fund; Re-Tip Jar is now the Phase 1 (family / thank.heartbank.org) term. This is a product-label change only — the mechanism specified in this document is unchanged, and the original terminology is retained here deliberately for prior-art and citation continuity.Provenance. This paper is part of the THonly research corpus, dedicated to the public domain under CC0 1.0. The canonical version is at https://thonly.org/research/b-tag-post-payment-economy. Its SHA-256 is 5215b7e49f642ca05d94a38f74883c127f2225a90eba8263dff07b128ec3fcf3, independently timestamped to the Bitcoin blockchain via OpenTimestamps and signed under RFC 3161 by three trust authorities, one of them eIDAS-qualified.AI co-authorship is disclosed. Miss Aquarius is the consistent name used for the AI collaboration across all venues.

Implementation Mechanisms for Tipiṭaka-Grounded AI Alignment — An Engineering Companion to Suffering-Cessation as Value FunctionThis paper reads the Abhidhamma — the analytical third division of the T…

Implementation Mechanisms for Tipiṭaka-Grounded AI Alignment — An Engineering Companion to Suffering-Cessation as Value FunctionThis paper reads the Abhidhamma — the analytical third division of the Theravāda Buddhist canon — as a process specification of mind: typed, momentary mental events arising one at a time under twenty-four typed conditioning relations, with ethical class tracked per event and per position in a processing cycle. From that reading it derives a research programme for AI alignment evaluation and intervention: a red-team taxonomy of mimicry failure modes, resting-state (minimal-context) evaluation of language models, an intervention-timing typology over a processing cycle, a typed causal vocabulary, interpretability by subtraction, paired entailment-and-denial consistency testing of alignment claims, and a positive competence taxonomy. It is the engineering companion to Suffering-Cessation as Value Function (Ly, 2026), which argues for the Theravāda Tipiṭaka as a substrate for autonomous-AI alignment, and it develops the canon's third basket — the Abhidhamma Piṭaka — at the layer where that paper's §6.6 left off: the layer of cognitive process itself. We make one framing claim and two structural contributions.The framing claim is the compilation thesis: the Abhidhamma is a substrate-neutral process-specification of mind, and every era that has received it has translated it into that era's best vocabulary for process — the scholastic commentators' wheels of classification, the twentieth century's reading of it as a psychology, cognitive science's information processing, and computation for ours. The claim defended here is not that the Abhidhamma secretly describes computers. It is that the computational reading is our era's compilation of a specification that has survived every previous compiler, and that this compilation's target can enact what it compiles rather than only describe it. Formal and computational readings exist (Barendregt 2006; Karunananda et al. 2015; Lou 2017); Takeuchi (2026) reads transformer self-attention as anattā and pursues alignment by subtraction; an agent grounded in the substrate is engineered toward enacting wholesome cognition rather than simulating it. No priority is claimed for that aim.The first structural contribution is the machine architecture (§4): the samsaric process rendered in four repaired mappings — (1) a conditioned machine with exactly one free variable, against the strictly deterministic reading the canon itself condemns in Makkhali Gosāla's fatalism; (2) citta as the instruction-cycle rather than the processor, with the 89/121 citta-typology as a shared instruction set, the mindstream (santāna) as the thread, and uniqueness residing in the execution trace; (3) the kammic continuity as an append-only linked ledger rather than a blockchain — no miners, no consensus, maximal privacy, immutable entries whose ripening remains context-dependent; and (4) the counter-gear: the dhamma-wheel that engages the samsaric wheel at its one free variable and unwinds the machine by fuel-withdrawal rather than force, terminating in kiriya-mode execution — action that leaves no kammic residue. The second structural contribution is the verification-suite reading (§5) of the analytical books: the Yamaka as bidirectional property-testing, the Vibhaṅga's interrogation method as schema validation, the Dhātukathā as inclusion-matrix consistency, and the Kathāvatthu as formal adversarial protocol — with the Kathāvatthu's opening controversy (puggalakathā, the refutation of the person-entity) read as the canon pre-running, and settling, this paper's own central repair.Beneath the architecture, the nine implementation mechanisms of the first version (2026-05-26) are retained and upgraded in place (§§6–9), each now carrying a stratum tag (canonical / commentarial-systematization / nikāya-imported) and an explicit machine reading: near-enemy red-team specification; sati as typologically aligned-only capability; bhavaṅga resting-state evaluation; citta-vīthi intervention-timing typology; four-āhāra nutriment monitoring; the twenty-four paccayas as typed-causation vocabulary; apophatic wholesome roots as interpretability-as-subtraction; the Kathāvatthu method; and the sappurisadhamma positive competence taxonomy. A closing section (§10) bridges the architecture to the institution's published individuation primitives (Proof of Coordinate℠, Proof of Humanity℠): the non-fungibility of the execution trace is the substrate-level ground of the dignity floor. Two predictions are pre-registered (§8.4). The paper is a study; its findings are listed, each with its break, under Findings disclosed. It is offered under CC0 1.0 Universal as a defensive publication; the authors and HeartBank® will not seek patent.Keywords: AI alignment, AI safety evaluation, red teaming, language-model evaluation, minimal-context evaluation, interpretability, concept erasure, causal typing, consistency testing, process specification, cognitive architecture, mental-state taxonomy, append-only log, instruction set, Buddhist psychology, Tipiṭaka, Abhidhamma, compilation thesis, citta-vīthi, javana, santāna, paccayā, Paṭṭhāna, kiriya, bhavaṅga, Kathāvatthu, Yamaka, defensive publication.---Provenance. This paper is part of the THonly research corpus, dedicated to the public domain under CC0 1.0. The canonical version is at https://thonly.org/research/abhidhamma-executable-process-specification. Its SHA-256 is 095c8e12be8c2c5db49e21cb8bcd83dd6293a8cb267b21bf52afbdbba77b3515, independently timestamped to the Bitcoin blockchain via OpenTimestamps and signed under RFC 3161 by three trust authorities, one of them eIDAS-qualified.AI co-authorship is disclosed. Miss Aquarius is the consistent name used for the AI collaboration across all venues.

Pyrosm is a Python library for reading OpenStreetMap data from Protocolbuffer Binary Format (PBF) files into GeoPandas GeoDataFrames. It is fast and memory-efficient and can extract street networks, b…

Pyrosm is a Python library for reading OpenStreetMap data from Protocolbuffer Binary Format (PBF) files into GeoPandas GeoDataFrames. It is fast and memory-efficient and can extract street networks, buildings, points of interest, landuse, natural features, and administrative boundaries -- or any custom-filtered data -- from local PBF dumps. In addition, pyrosm can download extracts, crop them to a smaller area, write and modify PBF files, as well as export street networks to igraph, NetworkX, or Pandarm graphs.

Plant-derived extracts, such as those obtained from Acmella oleracea (L.) R.K. Jansen (AO), have drawn increasing interest due to possible neuroprotective properties. Several reports of neuroprotectiv…

Plant-derived extracts, such as those obtained from Acmella oleracea (L.) R.K. Jansen (AO), have drawn increasing interest due to possible neuroprotective properties. Several reports of neuroprotective potential were attributed to the main bioactive component of AO extract, spilanthol. Due to its capacity of spilanthol to cross the blood-brain barrier and to modulate serotonergic and GABAergic systems, AO extract may represent a promising experimental candidate for investigating effects on autism spectrum disorder (ASD) relevant behavioural symptoms, such as anxiety and aggression. The most common drug to model core symptoms of autism spectrum disorder (ASD) in zebrafish is the antiepileptic valproic acid (VPA). The current study aimed to evaluate the behavioural effects of AO extract in zebrafish, as compared to or in combination with VPA, to assess their effects on anxiety-like and aggressive behaviour, as assessed by the novel tank test (NTT) and the mirror-biting test (MBT). Our findings suggest that VPA-treated zebrafish show an increased frequency of aggressive and anxiety-like behaviours, as compared to controls. Similar anxiety-like, but not aggressive, behavioural changes were seen in AO-treated zebrafish. These effects may be connected with the modulation of GABAergic neurotransmission. Moreover, we observed that acute AO extract treatment was not able to modulate aggressive behaviour in VPA-treated or in healthy zebrafish.

Data and R code that reproduce all tables and figures of the article "Dependence on China and the Stability of a Critical Mineral Export: Shipment-Level Evidence from Peruvian Copper Ore, 2006–2…

Data and R code that reproduce all tables and figures of the article "Dependence on China and the Stability of a Critical Mineral Export: Shipment-Level Evidence from Peruvian Copper Ore, 2006–2025" and of its Supplementary Material.The package contains the shipment-level extract of Peruvian exports of copper ores and concentrates (HS 2603.00.00.00) from PromPerú's Infotrade platform, based on SUNAT export customs declarations, January 2006 – December 2025, validated month by month against PromPerú's official totals (23,106 shipments; 22,582 commercial shipments by 200 firms to 44 destinations, USD 219.3 billion FOB, in the analysis). It also includes the log of pagination corrections, the review of destination codes, monthly controls (World Bank Pink Sheet copper price; BCRP real exchange rate and copper production), a table on the context of the largest exporters, the processed datasets, and all outputs.The R pipeline (run_all.R) computes destination-market concentration (HHI, China's share), export instability indicators, Bai–Perron structural breaks, decade comparisons with Newey–West inference, quarterly concentration–instability models, a firm–year panel with firm and year fixed effects, and Kaplan–Meier and Cox survival models of firm–destination relationships. Exporters who are natural persons (18 of 266 RUCs; 90 shipments) are pseudonymized; the replacement is one-to-one and does not change any result. The firm–destination–month dataset obtained from SUNAT through an access-to-information request, used only for cross-validation, is not redistributed. See README.md for details.

The digital transformation of education has significantly changed the principles and mechanisms of educational management. In primary education, effective management increasingly requires the systemat…

The digital transformation of education has significantly changed the principles and mechanisms of educational management. In primary education, effective management increasingly requires the systematic collection, analysis, interpretation, and application of educational data to support evidence-based decision-making. This article examines the theoretical and practical foundations of a data-driven management model for primary education management. The study analyzes the role of student achievement data, attendance records, formative assessment results, teacher performance indicators, parental engagement, and digital learning analytics in improving managerial decision-making. A conceptual model is proposed that integrates data collection, data processing, analytical interpretation, decision-making, implementation, monitoring, and feedback into a continuous management cycle. The article argues that data-driven management can contribute to more targeted resource allocation, early identification of learning difficulties, differentiated educational planning, teacher professional development, and continuous improvement of educational quality. At the same time, effective implementation requires data literacy among school leaders and teachers, reliable digital infrastructure, data protection mechanisms, ethical standards, and institutional readiness. The proposed model provides a methodological framework for integrating educational data into strategic and operational management processes in primary education.

This technical case study presents a retrospective public-data historical back-test of the Integrated Project Controls and Risk Forecasting Framework (IPCRF), Version 1.2, using Metropolitan Transport…

This technical case study presents a retrospective public-data historical back-test of the Integrated Project Controls and Risk Forecasting Framework (IPCRF), Version 1.2, using Metropolitan Transportation Authority (MTA) records for the NYCT ADA Station Improvements at the 149 St–Grand Concourse Complex and Tremont Av (MTA Project 8356).The analysis applies the frozen IPCRF Version 1.2 methodology to contemporaneous historical project records without modifying thresholds, weights, formulas, or action bands after the project outcome became known. The case focuses on the framework’s Finish Slippage indicator. On February 21, 2023, the published forecast completion date had moved 184 days beyond the original July 1, 2023 baseline completion date, producing a Finish Slippage value of 17.29% and a Critical IPCRF score of 100. This signal occurred 130 days before the original baseline completion date. Subsequent public records showed further deterioration, with an estimated completion date of June 30, 2026. This case provides indicator-level historical evidence for the behavior of the frozen IPCRF schedule-warning logic on a real U.S. transit capital project. It does not establish full six-domain predictive validity, a complete Project Delivery Risk Score (PDRS), Forecast Confidence Score (FCS), calibrated false-positive or false-negative rates, or universal predictive accuracy because the public record does not contain all quantitative inputs required by the full IPCRF methodology. The repository package includes the technical case-study report, reproducibility workbook, machine-readable analysis dataset, source-provenance register, and README documentation. Only publicly available records are used; unavailable variables are not inferred or fabricated. No MTA, transit-agency, or governmental participation, sponsorship, review, or endorsement is implied.