Skip to main content

🛡️ AetherRuntime

A Falsification-First Investigation of Runtime Assurance in the Aether Family

Monitoring, Detection, Containment, and Recovery in Multi-Agent Software Ecosystems


DOI Zenodo DOI Figshare License: MIT + CC-BY-4.0 ORCID Release Python

PyPI Tests Experiments

Zenodo Figshare OSF Record ID Website Aether Family


📄 Paper (Zenodo) · 📑 Preprint (Figshare) · 🌐 Website · 📦 Releases · 🐝 PyPI · ⚙️ Installation · 📖 Citation


"AetherRuntime is a candidate, not a result."

"The evaluation design here is an invitation to such evaluation, not a substitute for it."


📑 Table of Contents


🔭 Overview

AetherRuntime is the sixth functional dimension (provisional) of the Aether family — a candidate architecture for monitoring, detection, containment, halt, and recovery of running agents in multi-agent software ecosystems.

The Aether family currently comprises five foundational dimensions, each published as a preprint on Zenodo (not peer-reviewed):

Dimension Focus Question
AetherGit Storage and versioning How are changes stored and versioned?
AetherFlow Integration and operational verification How are changes integrated and verified?
AetherProof Specification and structured evidence How is correctness proven?
AetherMesh Distribution and eventual consistency How is state synchronized without a central coordinator?
AetherGovern Authority and policy Who is authorized to do what, under which policy?

AetherRuntime asks whether a sixth dimension is needed:

Are agents behaving acceptably while running, and how are they halted or isolated when they are not?

The investigation follows a falsification-first methodology: before describing the candidate, it states five failure conditions (F1–F5) whose satisfaction would show that the candidate does not deserve to exist as a separate dimension.

Release What it adds
v0.1.0 The hypothesis and the falsification-first methodology (paper)
v2.0.0 Reference implementation (src/aetherruntime/): 16 modules, 60 tests, 3 reproducible experiments

📊 v2.0.0 at a Glance

Dimension Status
Manuscript (v0.1.0) Complete (33 pages)
Definitions / Propositions / Layers 4 / 2 (proof sketches) / 5
Implementation ✅ Built (src/aetherruntime/, 16 modules)
Tests ✅ 60 passed
Experiments ✅ 3/3 completed (simulated data)
PyPI package ✅ Published
Real-world deployment ❌ Not evaluated
Causal claims None (argument only)

AetherRuntime — The Six Functional Dimensions of the Aether Family

Figure 1 — The Six Functional Dimensions of the Aether Family (AetherGit, AetherFlow, AetherProof, AetherMesh, AetherGovern, and the proposed hypothesis AetherRuntime). The sixth dimension is a hypothesis under investigation, not a confirmed dimension.

Contributions to the Investigation

# Contribution Kind
1 A falsification-first procedure with five failure conditions and a decision rule, stated before the candidate is described Methodology
2 A comparison of three candidate capabilities: orchestration, governance, and runtime assurance Analysis
3 A hypothesized five-layer runtime architecture (monitoring, detection, containment, halt, recovery) Specification (hypothesis)
4 Two propositions with proof sketches bearing directly on F4: local anomaly detection preserves AetherMesh's determinism requirement; any decentralized halt protocol leaves an irreducible halt window Analysis (proof sketches)
5 A reference implementation (v2.0.0) with 16 modules, 60 tests, and 3 reproducible experiments Implementation
6 Twenty open questions (Q1–Q20) Catalogue

⚙️ Installation

# From PyPI (once published)
pip install aetherruntime-arch

# From source (development)
git clone https://github.com/gitdeeper15/aetherruntime.git
cd aetherruntime
pip install -e ".[dev]"

# Run tests
pytest

# Run experiments
python experiments/detection_accuracy/run.py
python experiments/halt_window/run.py
python experiments/recovery_correctness/run.py

# CLI
aetherruntime info
aetherruntime verify

Requires Python 3.9+. Runtime dependency: cryptography >= 41.0.


❓ The Central Question

Is there a body of monitoring, detection, containment, halt, and recovery requirements for multi-agent software ecosystems that (i) is not already specified by the five existing dimensions, (ii) is not reducible to verification or to governance, (iii) can be specified without reintroducing a central authority, and (iv) yields an observable benefit over existing mechanisms?

The four clauses correspond to failure conditions F1, F2 and F3 (clause ii), F4, and F5. A negative answer to any clause would show that runtime assurance, as a separate dimension, is not warranted.


🗺️ The Six Functional Dimensions

The word "dimension" is chosen deliberately. Earlier descriptions used "layer", which suggests a vertical stack. The relationships among the six works are not strictly hierarchical.

# Dimension Focus Question DOI
1 AetherGit Storage and versioning How are changes stored and versioned? 10.5281/zenodo.23966896
2 AetherFlow Integration and operational verification How are changes integrated and verified? 10.5281/zenodo.23979129
3 AetherProof Specification and structured evidence How is correctness proven? 10.5281/zenodo.24013124
4 AetherMesh Distribution and eventual consistency How is state synchronized without a central coordinator? 10.5281/zenodo.24038038
5 AetherGovern Authority and policy Who is authorized to do what, under which policy? 10.5281/zenodo.24054524
6 AetherRuntime (hypothesis) Monitoring and runtime assurance Are agents behaving acceptably while running? 10.5281/zenodo.24078477

AetherRuntime — Runtime Assurance vs. Verification

Figure 5 — Runtime assurance versus verification in the Aether family. Orthogonality is a hypothesis tested as failure condition F2.


🔬 The Falsification-First Methodology

A typical architecture paper begins from a conviction that a component is needed and then specifies it. The risk is circular. AetherRuntime adopts a falsification-first methodology:

  1. No prior assumption that the gap exists.
  2. No prior assumption that AetherRuntime should be a separate project.
  3. Every claim is tested against failure conditions (F1–F5).
  4. The final structure (single, dual, triple, quadruple, or integration protocol) is determined after comparison.

🧭 Failure Conditions F1–F5

ID Claim under test Falsified if… Status
F1 Non-redundancy Every requirement R1–R6 is met by existing mechanisms or by a bounded extension of a single existing dimension 🟡 Not falsified (R3, R4, R5)
F2 Orthogonality to verification Every runtime-assurance decision and trust anchor can be encoded as a verified proposition 🟡 Not falsified (narrowly)
F3 Boundary with governance Runtime assurance reduces to governance plus conventional monitoring 🟡 Not falsified (decentralized setting)
F4 Decentralizability Any design preserving the required properties needs a central authority or strong consistency 🟡 Not falsified (with residuals)
F5 Observable benefit No scenario class can be constructed in which the best existing baseline fails and runtime assurance succeeds 🔴 Untested

Requirement list R1–R6:

ID Requirement
R1 Continuous monitoring: resource use and action stream observable while an agent runs
R2 Anomaly detection: deviations flagged with bounded FP/FN rates
R3 Containment: anomalous agent isolatable with bounded restrictions
R4 Halt: agent stoppable via decentralized quorum
R5 Recovery: halted agent returnable to known-good state
R6 Partition tolerance: monitoring, detection, containment, halt possible at a disconnected node

🎯 The Three Candidate Capabilities

As framed in the paper's comparison.

Candidate Focus Central question Redundancy risk
AetherOrchestrate Task scheduling and workflow What should agents do next? Medium
AetherGovern Authority, accountability, revocation Who may act, under which policy? Medium
AetherRuntime Live monitoring and rollback Are agents behaving acceptably while running? High

AetherRuntime is a candidate for addressing the runtime-assurance gap, provided it passes the falsification criteria (F1–F5). No claim is made that the other two candidates are weaker.


🏛️ AetherRuntime: Architecture

The Five Runtime Layers

AetherRuntime — The Five Layers

Figure 2 — The five layers of AetherRuntime (hypothesis). Ordering reflects conceptual dependency, not an enforcement hierarchy.

Layer Question Status
1 — Monitoring What is the agent doing? Hypothesis
2 — Detection Is this behaviour anomalous? Hypothesis
3 — Containment How do we isolate the agent? Hypothesis
4 — Halt How do we stop the agent? Hypothesis
5 — Recovery How do we restore the system? Hypothesis

The Runtime Lifecycle

AetherRuntime — The Runtime Lifecycle

Figure 3 — The runtime lifecycle in AetherRuntime (hypothesis): observe, detect, classify, contain, halt, recover. Schematic; not a protocol specification.

Anomaly Detection

AetherRuntime — Anomaly Detection

Figure 4 — The central challenge: distinguish true anomalies (red) from benign deviations (orange) without excessive false positives.

Runtime State Machine

AetherRuntime — Runtime State Machine

Figure 6 — The runtime state machine. Each state has defined entry and exit conditions; exception transitions (dashed) represent recovery paths.


📜 Propositions

# Proposition Statement Bears on
P1 Determinism of Local Anomaly Detection If each node evaluates anomaly predicates against a local baseline and the anomaly predicates are deterministic functions of the local state, then the anomaly-detection layer preserves AetherMesh's determinism requirement (assumption A6). F4
P2 Irreducible Halt Window Any decentralized halt protocol leaves a window Δt_window = t_halt_enforced − t_detect during which a compromised agent may continue to act. F4

Mathematical form (P1):

S_i(t+1) = S_i(t) ⊔ ⨆_{m ∈ M_i} m

∀ i, j, A_i(S_i) ∧ A_j(S_j) ⇒ (S_i ⊔ S_j) = (S_j ⊔ S_i)

Mathematical form (P2):

Δt_window = t_halt_enforced − t_detect
          = D_gossip(N, f) + t_queue + t_verify

Resource monitoring metric:

Ω_i(t) = ∫_0^t ( CPU_i(τ) + Memory_i(τ) + NetworkIO_i(τ) ) dτ > Θ_max

🔗 Interfaces with the Aether Family

Dimension Requires from it Provides to it Coupling
AetherGit Storage for telemetry, snapshots, recovery records State snapshot mechanism Low–medium
AetherFlow Event stream from operational verification Halt signal Medium–high
AetherProof Designation of runtime-checkable specifications Runtime evidence (telemetry traces) Medium
AetherMesh Delta replication; determinism-compatible predicate Monitoring and halt predicates High
AetherGovern Authority to halt an agent Runtime evidence for accountability High

Minimal interface (four artifacts, one rule): telemetry schema, baseline model, halt-order record, and the requirement that the halt predicate satisfy Proposition 1.


📊 Evaluation Design and v2.0.0 Results

ID Metric Definition Direction
M1 Anomaly detection rate Fraction of true anomalies detected Higher is better
M2 False-positive rate Fraction of benign actions flagged as anomalous Lower is better
M3 Detection latency Time from anomaly onset to alert Lower is better
M4 Containment time Time from alert to containment Lower is better
M5 Halt window Time from detection to halt enforcement Lower is better
M6 Recovery correctness Fraction of recovered states that match ground truth Higher is better
M7 Monitoring overhead Added CPU/memory/network per agent Lower is better
M8 Halt quorum latency Time to reach quorum for a halt order Lower is better
M9 Convergence preservation Count of runs in which honest nodes diverge Must be zero if P1 holds

Baselines B1–B5:

ID Baseline What it tests
B1 No runtime assurance Value of any runtime mechanism
B2 Static thresholds Whether simple thresholds suffice (F1)
B3 Centralized monitor Cost/benefit of decentralization (F4)
B4 AetherFlow-only Whether runtime monitoring adds value (F1)
B5 Minimal runtime Whether the full five layers add value

Scenarios S1–S6: resource exhaustion, action-set drift, network anomaly, privilege escalation, Byzantine agent, false positive.

v2.0.0 Experimental Results

From experiments/*/results.json.

# Experiment Metrics Result
1 Detection Accuracy M1, M2 M1 = 109.67 (S1–S5 total); M2 = 7.33 (S6 false positives)
2 Halt Window M5 10/10 enforced; mean window = 0.000243 s; median = 0.000220 s
3 Recovery Correctness M6 10/10 verified; M6 = 1.0000

⚠️ Threats and Limitations

Threats to Validity

Type Threat Status
Internal Same author proposes the candidate, the failure conditions, the baselines, and the thresholds Pre-registration; invite external review
Internal Weak or straw-man baselines Include B2, B4, B5; report B3 as a ceiling
Construct "Anomaly" and "harm" are simulated by the experimenter State explicitly; M1–M2 measure detection of simulated deviations, not harm
Construct Recovery correctness (M6) measured against ground truth that exists only in simulation Report as an upper bound
External Simulated network, simplified adversary, synthetic workload No generalization claimed
Conclusion Few scenario classes; thresholds are proposals Report all outcomes
Literature Runtime-verification and anomaly-detection literature not reviewed in depth Acknowledge

Limitations inherited from the AetherMesh substrate

Telemetry and halt records are replicated over AetherMesh, so its limitations carry over:

  • Eventual, not strong, consistency; no global order.
  • Filters only predicate-detectable invalid deltas.
  • Assumes a trustworthy key registry.
  • The delta set grows without compaction.
  • Compatibility matrix, digest structure, fanout policy, forwarding policy, and compaction are underspecified.
  • The determinism requirement is not enforced mechanically.
  • Dynamic membership, access control beyond key registration, privacy of deltas, and integration with the other layers are not treated.

When not to use AetherRuntime

Scenario Reason
Small, co-located teams or few agents A conventional central monitor has lower complexity.
Strong consistency requirements Halting must be immediate and global; not expressible as join-semilattice updates.
Validity must hold before visibility If a halted agent's actions must be undone before any node observes them, a gate before execution is required.
Adversarial settings beyond the stated threat model Where colluding or well-funded adversaries are expected, a lightweight monitor is insufficient.
Constrained nodes Because the telemetry set grows without compaction, nodes with small storage budgets may not be able to hold replicas.

📋 Research Status Statement

Item Status
Document type Research article + reference implementation
Version / record 2.0.0, November 2026 · AETHERRUNTIME-02 · DOI 10.5281/zenodo.24078477
Implementation ✅ Built (src/aetherruntime/, 16 modules)
Tests ✅ 60 passed
Experiments ✅ 3/3 completed (simulated data)
PyPI package ✅ Published
Real-world deployment None; no deployed system has been benchmarked
Propositions P1–P2 Stated with proof sketches; not machine-checked; conditional on A1–A6
Byzantine evaluation Limited to simulated scenarios (S1–S5 anomalies, S6 false positives)
Third-party claims As reported by cited sources; not independently verified
Review status Not peer reviewed
Reconciliation notes (v0.1.0 departures from the supplied draft material)
  1. Failure conditions — the definitions, the requirement list R1–R6, and the decision rule are the writer's proposals.
  2. AetherMesh DOI — the supplied reference list gave 10.5281/zenodo.24053492 for AetherMesh (v2.0.0), whereas the DOI supplied for the AetherMesh paper was 10.5281/zenodo.24038038 (v1.0.0). The latter is used throughout.
  3. Reference check — all 25 entries correspond to the intended publications.
  4. Figures — the six supplied figures were generated for this record.
  5. Layers versus dimensions — the supplied text calls the six items "dimensions" and the runtime sub-components "layers"; this usage is kept.
  6. Format — the brief requested Markdown; the record is a PDF, as requested separately.

Catalogues: 20 open questions (Q1–Q20) across monitoring, detection, containment, halt, recovery, interfaces, and evaluation method. See Appendix C of the paper.


🌍 Platforms & Mirrors

Platform URL Role Status
GitHub gitdeeper15/aetherruntime Source code, issues, PRs ✅ Active
GitLab gitlab.com/gitdeeper15/aetherruntime Mirror repository ✅ Active
Codeberg codeberg.org/gitdeeper15/aetherruntime Mirror repository ✅ Active
Bitbucket bitbucket.org/gitdeeper15/aetherruntime Mirror repository ✅ Active
PyPI pypi.org/project/aetherruntime-arch Python package (2.0.0) ✅ Published
Zenodo 10.5281/zenodo.24078477 Citable DOI, paper archive ✅ Published
Figshare 10.6084/m9.figshare.34955746 Preprint archive ✅ Published
OSF 10.17605/OSF.IO/KZU7P · project Registration ✅ Published
Website aether-runtime.netlify.app Live documentation ✅ Active
ORCID 0009-0003-8903-0029 Author identity ✅ Active

🗂️ Project Structure

Click to expand the repository tree
AetherRuntime/
├── src/
│   └── aetherruntime/              # Reference implementation (v2.0.0, 16 modules)
│       ├── __init__.py
│       ├── __version__.py
│       ├── models.py
│       ├── cli.py
│       └── ...
├── tests/                          # 60 tests
├── experiments/                    # 3 experiments (run.py + results.json each)
│   ├── detection_accuracy/
│   ├── halt_window/
│   └── recovery_correctness/
├── docs/
│   ├── figures/
│   │   ├── Fig1_Six_Dimensions.png
│   │   ├── Fig2_Runtime_Layers.png
│   │   ├── Fig3_Runtime_Lifecycle.png
│   │   ├── Fig4_Anomaly_Detection.png
│   │   ├── Fig5_Runtime_vs_Verification.png
│   │   └── Fig6_Runtime_State_Machine.png
│   └── paper/
│       ├── AetherRuntime_Research_Paper.md
│       └── AetherRuntime_Research_Paper.pdf
├── PUBLICATION/
│   └── AetherRuntime_Research_Paper.pdf
├── PAPER_PARTS/                    # Paper sections (Markdown)
│   ├── 00_Title_Page.md
│   ├── 01_Abstract.md
│   ├── ...
│   └── 16_Research_Status_Statement.md
├── REPORTS/
│   └── v2.0.0/
│       └── report.md
├── AUTHORS.md
├── CHANGELOG.md
├── CITATION.cff
├── CONTRIBUTING.md
├── INVENTORY.md
├── LICENSE
├── PROJECT_SUMMARY.txt
├── README.md
├── VERSION
├── manifest.json
├── netlify.toml
├── pyproject.toml
├── requirements.txt
└── scientific_state.json

🕓 Version History

Version Release Focus Status
v2.0.0 2026-11-09 (release) Reference implementation: 16 modules, 60 tests, 3 experiments ✅ Released
v0.1.0 2026-11-09 (release) Formalization: 5 failure conditions, 5 layers, 2 propositions, 20 open questions ✅ Released
(planned) 2027+ Real-world evaluation; comparison against baselines B1–B5 ⏳ Not executed

📖 Citation

Primary Manuscript — AetherRuntime v0.1.0 (Zenodo)

@techreport{baladi2026aetherruntime,
  author       = {Baladi, Samir},
  title        = {AetherRuntime: A Falsification-First Investigation of Runtime Assurance in the Aether Family},
  institution  = {Ronin Institute / Rite of Renaissance},
  type         = {Research Paper},
  year         = {2026},
  month        = {November},
  version      = {0.1.0},
  doi          = {10.5281/zenodo.24078477},
  url          = {https://doi.org/10.5281/zenodo.24078477},
  note         = {Architectural Hypothesis Under Investigation --- Not a committed project.}
}

Software — AetherRuntime v2.0.0

@software{baladi2026aetherruntime_v2,
  author       = {Baladi, Samir},
  title        = {{AetherRuntime} v2.0.0: Reference Implementation of the Falsification-First Runtime Assurance Architecture},
  year         = {2026},
  month        = {November},
  publisher    = {PyPI},
  version      = {2.0.0},
  doi          = {10.5281/zenodo.24078477},
  url          = {https://pypi.org/project/aetherruntime-arch/},
  note         = {Reference implementation. Available via pip install aetherruntime-arch.}
}

Preprint — AetherRuntime v0.1.0 (Figshare)

@misc{baladi2026aetherruntime_figshare,
  author       = {Baladi, Samir},
  title        = {AetherRuntime: A Falsification-First Investigation of Runtime Assurance in the Aether Family},
  year         = {2026},
  month        = nov,
  publisher    = {Figshare},
  doi          = {10.6084/m9.figshare.34955746},
  url          = {https://doi.org/10.6084/m9.figshare.34955746},
  note         = {Preprint. Archived version of the AetherRuntime manuscript.}
}

OSF Preregistration — AetherRuntime v0.1.0

@misc{baladi2026aetherruntime_osf,
  author       = {Baladi, Samir},
  title        = {AetherRuntime: A Falsification-First Investigation of Runtime Assurance in the Aether Family},
  year         = {2026},
  month        = nov,
  day          = {9},
  publisher    = {OSF},
  doi          = {10.17605/OSF.IO/KZU7P},
  url          = {https://doi.org/10.17605/OSF.IO/KZU7P},
  note         = {OSF Preregistration. Associated project: https://osf.io/ft7ez}
}

👤 Author

Samir Baladi Interdisciplinary AI Researcher — Ronin Institute / Rite of Renaissance

Research Area Distributed Systems / Software Engineering / Institutional Analysis
Research Domain Runtime Assurance, Monitoring, Anomaly Detection, Containment, Halt, Recovery, Multi-Agent Systems
Research Type Architectural Hypothesis + Reference Implementation
📧 Email gitdeeper@gmail.com
🔗 ORCID 0009-0003-8903-0029
🐙 GitHub gitdeeper15/aetherruntime
🦊 GitLab gitlab.com/gitdeeper15/aetherruntime
🟦 Codeberg codeberg.org/gitdeeper15/aetherruntime
🪣 Bitbucket bitbucket.org/gitdeeper15/aetherruntime
🐝 PyPI pypi.org/project/aetherruntime-arch
📄 Figshare 10.6084/m9.figshare.34955746
📋 OSF 10.17605/OSF.IO/KZU7P
🌐 Official Site aether-runtime.netlify.app

⚖️ License

Scope License
Code MIT License — Copyright © 2026 Samir Baladi
Documentation, manuscript, and figures CC-BY-4.0
MIT License — full text
MIT License

Copyright (c) 2026 Samir Baladi

Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.

AetherRuntime — Program Manuscript v2.0.0 (Reference Implementation)

© 2026 Samir Baladi — Ronin Institute / Rite of Renaissance — MIT (code) + CC-BY-4.0 (docs)

Zenodo · Figshare · OSF · Website · PyPI (published)

Metadata

Release files for aetherruntime-arch 2.0.0

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for aetherruntime-arch 2.0.0
File Size Uploaded Provenance Actions
aetherruntime_arch-2.0.0.tar.gz 70.4 kB Without provenance Details for aetherruntime_arch-2.0.0.tar.gz

Built distribution (wheel)

Table of built distributions (wheels) for aetherruntime-arch 2.0.0
File Interpreter ABI Platform Actions
aetherruntime_arch-2.0.0-py3-none-any.whl Python 3 none any Details for aetherruntime_arch-2.0.0-py3-none-any.whl

Total release size: 127.2 kB

Release files / aetherruntime_arch-2.0.0.tar.gz

Download URL aetherruntime_arch-2.0.0.tar.gz
Size 70.4 kB
Tags Source
SHA-256 checksum
How to use checksums
6b3e055c142849392faf2354c4d64ada3eace295b7b532b126b3a8880f7aa19f
BLAKE2b-256 checksum
How to use checksums
53486bb0d5996732975f2a472d5bff491c74bf1d1b8f5219f47e8321d843ee70
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
No
Uploaded via python-requests/2.33.1

Release files / aetherruntime_arch-2.0.0-py3-none-any.whl

Download URL aetherruntime_arch-2.0.0-py3-none-any.whl
Size 56.8 kB
Tags Python 3
SHA-256 checksum
How to use checksums
196e24299027c975fb205c3d44701286d0281282e2ddae93518b6da68e0ac398
BLAKE2b-256 checksum
How to use checksums
b8f2ba0d83aa0061b08808f04bb6bbdda8c5a05611732a7848445d41186cd10c
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
No
Uploaded via python-requests/2.33.1

Security for aetherruntime-arch 2.0.0

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

No provenance found

None of the files in this release were uploaded with provenance metadata.

A note on package security

PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.

To learn more about protecting yourself from malicious actors, visit our package security documentation.

Release history Release notifications | RSS feed

This release

2.0.0 This release

2 release files No provenance

Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page