Upcube / Ethen Research Lab
Researching what comes next.
Ethen Research Lab studies intelligence, computing, science and frontier technology. It publishes methods, protocols and results with an explicit evidence status, and it builds the knowledge and capability that strengthen Ethen and the wider mission.
Ethen Research Lab is Upcube’s research organization and public publication program. Each publication is labelled with its evidence status. Looking for the AI research workspace? That is the Ethen Research product, which is separate from this publication record.
Catalog countCatalog count, drawn from the live publication record.
Ethen is an AI-native technology organization building toward a research institution. Two formal research programs are ratified; neither has reported a measured result yet.
Research universe
Select a field to see how it connects, and how far Ethen's work in it has actually gone.
Agent reliability, verification, routing, evaluation, memory and model behaviour. The Applied program and most of the publication record live here.
Connects to
Research fields
Research programs
A field becomes a program only through an admission gate: a falsifiable question, rights-cleared data, a named independent reviewer and budget headroom. Five simultaneous institutes were rejected on purpose.
Ratified program, running
P1 Applied
When can machine work be trusted, what does one verified outcome cost, and do verifiers fail in measurable ways?
P2 Frontier
Can an agent, the Lean kernel and an independent human formalize known results at a measured, falling cost per verified proof?
Methods or infrastructure under construction
M Shared methods
Does structured, verified knowledge improve future work on tasks it has never seen?
Next in line, behind an admission gate
P3 Frontier
Anomaly detection and archive mining on public astronomical data.
Preparing or partnering; not admitted
Applied / Frontier
Preparing protocols and partners. Not admitted as programs.
Watched; deferred to a later stage
Frontier
Watched and deferred to a later stage of the institution.
No program has produced a measured result yet. When one does, it will appear in the record with its evidence status.
How Ethen researches
AI agents scale the work: reading, drafting, running attempts, checking. They are a multiplier, not the point. The institution exists for discovery and for capability that holds up under independent review.
A falsifiable question with stated kill criteria.
What is already known, cited only from verified sources.
A prediction specific enough to be wrong.
Confirmatory protocols are hash-pinned before data collection starts.
Attempts captured with their rights, cost and verifier outcome.
Synthetic results are labelled synthetic, never physical evidence.
Licences checked at capture; failures kept as corpora.
Sealed sets, paired designs, stated intervals.
No producer is the sole acceptor of its own result.
Replication by a different operator or pipeline before a claim hardens.
Claims, evidence and decisions kept in institutional memory with their lineage.
Methods public, corpora sealed; corrections public within 30 days.
Verified findings become verifiers, policies, datasets or methods.
Integrity rules
Evidence standards
Every publication declares how strong its evidence is. Publication type and evidence status are separate: a benchmark design is a specification, not a benchmark result. Select a rung to filter the archive.
Measured / Empirical
Ethen ran an experiment, benchmark, or measurement and reports the result.
Publication types on this rung: Evaluation Report, Benchmark Report, Experiment Report
System Evidence
Evidence tied to a specific implemented or pinned Ethen system or build. It does not generalize beyond that build.
Publication types on this rung: System Card, Evaluation Report, Benchmark Report, Experiment Report
Research Synthesis
Analysis of existing evidence and literature. No new Ethen measurements are implied.
Publication types on this rung: Survey
Protocol / Planned Experiment
The method is defined. The experiment has not been run, so no results are reported.
Publication types on this rung: Benchmark Design, Research Protocol
Benchmark Design
A benchmark specification. It has not necessarily been run, and no scores are implied.
Publication types on this rung: Benchmark Design
Proposal / Hypothesis
A proposed direction, system, or hypothesis that remains untested.
Publication types on this rung: Research Proposal
Research to capability
Not all research is commercialized. Some knowledge, datasets, models and infrastructure stay internal where that is the right call, and some research exists only to find out what is true.
Where a finding can go
Methods, accepted results and reproduction slices.
Sealed evaluation corpora, failure detail, strategic work.
Where protection serves the mission.
Where others can use it better.
Where a capability deserves its own organization.
ConceptualHow a finding can travel. Most stop earlier, and that is by design.
Research infrastructure
Each system carries its real status. Most are being designed or built; the labels say which.
Model-certification, completion-gate and offline-enumeration harnesses exist; the four-arm experiment harness has not run.
Pre-registration with hash-pinned protocols; the machine store is planned.
One ledger for every attempt, verification, intervention and cost. Not built yet.
A rights-aware dataset compiler that fails closed. Rights at capture is not built.
Memory records with provenance, contest and supersession exist in one product; the shared memory is being built.
Pinned inputs and independent re-runs as the rule for every public claim.
Simulated enterprises and robot simulators, each opening at its gate.
Rented today; owned compute only past a measured break-even.
Agents already draft, check and run much of the work under human review.
No laboratory exists. A long-horizon direction, gated on funded protocols.
Publications
System card for one offline enumeration of a pinned Ethen build: 197 conditions, 194 passed. Not a paper and not a security disclosure.
Evidence status: System EvidenceOffline enumeration · one pinned build
A research agenda for AI agents that learn only from experience that is verified, rights-cleared and shown to transfer across tasks, tools and model generations.
Evidence status: Research Synthesisresearch synthesis; no new measured Ethen results
Logs, traces, trajectories, outcomes and corrections are not the same asset. A research note on what turns agent telemetry into verified experience.
Evidence status: Research Synthesisresearch synthesis; no new measured Ethen results
A benchmark design for AI agents that take action: executable environments, state-based verification, cost reporting, permissions, recovery and reproducibility.
Evidence status: Benchmark Designbenchmark design; not yet run
A research protocol on learning from experience for AI agents: do verified, failure-rich data beat raw traces at matched volume, on held-out work?
Evidence status: Protocol / Planned Experimentresearch protocol; not yet run; no measured gain curves exist
A research proposal for sovereign AI: how enterprise AI could improve across private deployments without raw data leaving the customer boundary.
Evidence status: Proposal / Hypothesisfrontier research proposal; not built; not production-ready; requires privacy, security and legal review
The complete published archive.
A research agenda for AI agents that learn only from experience that is verified, rights-cleared and shown to transfer across tasks, tools and model generations.
Evidence status: Research Synthesisresearch synthesis; no new measured Ethen results
Logs, traces, trajectories, outcomes and corrections are not the same asset. A research note on what turns agent telemetry into verified experience.
Evidence status: Research Synthesisresearch synthesis; no new measured Ethen results
A technical report proposing the Work Receipt: one signed record of authority, actions, effects, verification, cost and rights for every unit of autonomous AI work.
Evidence status: Proposal / Hypothesisarchitecture proposal; no measured Ethen results
A research note on mandates: one object that compiles human intent into AI agent authorization, budgets, approvals, delegation limits, expiry and revocation.
Evidence status: Proposal / Hypothesisarchitecture proposal; cites one published measured Ethen system card (AgentTrustBench R08)
A research proposal for commitment graphs: tracking intent, obligations, preconditions, effects and evidence so AI agents stop declaring unfinished work complete.
Evidence status: Proposal / Hypothesisresearch proposal; untested
A research proposal on AI agent error recovery: branch each failure in a sandbox into candidate recoveries and learn when to retry, reconcile, escalate or stop.
Evidence status: Proposal / Hypothesisresearch proposal; untested
A methods paper on reward integrity for AI agents: verifier false accepts and rejects, abstention, grader drift, expert disagreement and reward hacking.
Evidence status: Research Synthesisresearch synthesis; no new measured Ethen results
A position paper reframing AI model routing as an execution-configuration decision across model, context, tools, verification, recovery, cost and risk.
Evidence status: Research Synthesisresearch synthesis; no new measured Ethen results
A survey of learned LLM routing: what RouteLLM, RouterBench and LLMRouterBench show, why strong rules are the right baseline, and how to test non-inferiority.
Evidence status: Research Synthesisliterature survey; no Ethen measurements
A research proposal for counterfactual evaluation of AI agents: replaying completed tasks under alternative models, tools, context and recovery strategies.
Evidence status: Proposal / Hypothesisresearch proposal; untested
A research note on model change assurance: replaying an organization's own historical tasks to find regressions before an LLM upgrade reaches real work.
Evidence status: Research Synthesisresearch synthesis; no new measured Ethen results
A position paper on the AI data moat: why volume is not defensibility, and eight axes, from rights to outcome density and transfer, that decide what compounds.
Evidence status: Research Synthesisresearch synthesis; no new measured Ethen results
A research note on the Outcome Warehouse: storing verified, versioned AI outcome data with corrections, cost, lineage, rights and delayed business results.
Evidence status: Research Synthesisresearch synthesis; no new measured Ethen results
A research proposal for Skill IR: AI agent skills as portable, testable capability contracts with permissions, verifiers and compatibility records.
Evidence status: Proposal / Hypothesisresearch proposal; untested
A methods paper on measuring capability transfer: whether AI agent skills keep working across model, tool and task changes, including negative transfer.
Evidence status: Proposal / Hypothesismethods proposal; no measurements
A research proposal for context compaction that keeps obligations, permissions, deadlines and evidence out of lossy summaries while cutting an agent's token cost.
Evidence status: Proposal / Hypothesisresearch proposal; untested
A research note on AI training data rights as infrastructure: purpose grants, consent, residency, expiry, revocation, descendants and training eligibility.
Evidence status: Proposal / Hypothesisarchitecture proposal; requires legal review; no measured results
A research proposal for a rights-aware dataset compiler: dataset builds that check purpose, lineage, consent and contamination at build time and fail closed.
Evidence status: Proposal / Hypothesisresearch proposal; untested; requires legal review
A research note defining cost per verified outcome (CPVO) and comparing it with cost per token, request, seat and task as a unit for agentic AI economics.
Evidence status: Research Synthesisresearch synthesis; no new measured Ethen results
A position paper stress-testing AI evaluation against 10× better models and 10× cheaper inference, and arguing that verification and assurance gain value.
Evidence status: Research Synthesisresearch synthesis; no new measured Ethen results
A benchmark design for AI agents that take action: executable environments, state-based verification, cost reporting, permissions, recovery and reproducibility.
Evidence status: Benchmark Designbenchmark design; not yet run
A benchmark design for evaluating AI agents under failure: stale state, tool errors, unknown effects, interrupted actions, partial completion and safe stopping.
Evidence status: Benchmark Designbenchmark design; not yet run
A benchmark design measuring whether AI agent capabilities survive changes of model, provider, tool interface, task family and harness, including negative transfer.
Evidence status: Benchmark Designbenchmark design; not yet run
A benchmark design measuring what AI agents must remember: obligations, constraints, permission changes, superseded facts and revoked data, across compaction.
Evidence status: Benchmark Designbenchmark design; not yet run
A benchmark design for AI agent authorization: delegation, approval binding, revocation latency, budgets and prompt-injection resistance, tested on system and agent.
Evidence status: Benchmark Designbenchmark design; cites one published measured Ethen system card (R08)
A research protocol for an LLM routing experiment: strong-rule baselines, propensity-logged data, factorial cache controls, non-inferiority tests and decision rules.
Evidence status: Protocol / Planned Experimentresearch protocol; not yet run
A research protocol for LLM verifier reliability: stratified gold sets, false accept and reject rates, calibration, drift and exploit audits, and grader cards.
Evidence status: Protocol / Planned Experimentresearch protocol; not yet run
A research proposal for enterprise agent simulation: an executable synthetic company with CRM, support, documents, identity, approvals, finance and email.
Evidence status: Proposal / Hypothesisresearch proposal; not built
A research note on idempotency for AI agents: unknown effects after timeouts, reconcile-before-retry, idempotency keys, compensation and exactly-once limits.
Evidence status: Research Synthesisresearch synthesis; no new measured Ethen results
A research paper proposing a multi-axis AI agent failure taxonomy, the failure genome, built on recent work on failure modes, attribution and critical steps.
Evidence status: Research Synthesisliterature survey and taxonomy proposal; no Ethen failure dataset measured
A research proposal for process memory for AI agents: mining completed, verified work into per-organization process models that guide plans and flag anomalies.
Evidence status: Proposal / Hypothesisresearch proposal; untested
A research proposal for private AI evaluation: replaying an enterprise's own historical agent tasks inside its boundary, with only bounded aggregates leaving.
Evidence status: Proposal / Hypothesisresearch proposal; not built; requires security and legal review
A research protocol testing whether model change assurance predicts live outcomes: paired replay, regression certificates, randomized rollout and false reassurance.
Evidence status: Protocol / Planned Experimentresearch protocol; not yet run; no model change assurance results exist
Why AI routers need propensity logging from the first decision: candidate sets, selection probabilities and verified outcomes make off-policy evaluation possible.
Evidence status: Research Synthesistechnical research note; external literature and Ethen design decisions; no Ethen measurements
A research protocol for AI skill portability: a pre-registered event study of whether agent skills transfer, degrade or reverse after a frontier-model upgrade.
Evidence status: Protocol / Planned Experimentresearch protocol; not yet run; no skill-survival measurements exist
A research protocol for the AI agent context budget: dose-response curves for five context strategies, with obligation retention and revoked-data exposure.
Evidence status: Protocol / Planned Experimentresearch protocol; not yet run; no measured context-budget curves exist
A research protocol for recovery transfer: does agent recovery knowledge carry to unseen APIs, tool versions and failure mechanisms, including when to stop?
Evidence status: Protocol / Planned Experimentresearch protocol; not yet run; no recovery-transfer evidence exists
A research protocol on learning from experience for AI agents: do verified, failure-rich data beat raw traces at matched volume, on held-out work?
Evidence status: Protocol / Planned Experimentresearch protocol; not yet run; no measured gain curves exist
Private AI improvement without raw data export: tenant-local learning, federated evaluation, secure aggregation and differential privacy, each with its threat model.
Evidence status: Proposal / Hypothesisresearch survey and proposal; no Ethen implementation or measurement; requires privacy, security and legal review
A research proposal for sovereign AI: how enterprise AI could improve across private deployments without raw data leaving the customer boundary.
Evidence status: Proposal / Hypothesisfrontier research proposal; not built; not production-ready; requires privacy, security and legal review
System card for one offline enumeration of a pinned Ethen build: 197 conditions, 194 passed. Not a paper and not a security disclosure.
Evidence status: System EvidenceOffline enumeration · one pinned build
One system
The loop runs in every direction. Product use raises research questions, evaluations reopen old answers, and some research ends in knowledge that never becomes a product.
Discovers and validatesYou are here
The research organization studying intelligence, computing and science, with an explicit evidence status on everything it publishes.
Turns research into intelligence
Ethen's model and intelligence-development program: evaluation, routing, data and adaptation, building toward increasingly proprietary capability.
Explore FarosPuts intelligence to work
The product people use to think, research, write and build with AI.
Explore EthenCloses the loop
Outcomes, failures and evaluations become the questions the next round of research starts from.