Artifacts
Research artifacts published on hbar.science, labeled by epistemic zone, rigor tier, and AI involvement.
Artifacts are pointers and structured metadata — not hosted files. Submit an artifact →
What counts as an artifact?
An artifact is any inspectable object produced in the course of scientific inquiry that enables understanding, critique, or replication.
Artifacts are inputs to knowledge, not just outputs. They make reasoning, evidence, and uncertainty visible rather than compressed into opaque claims.
Hosting scope. hbar.science does not host large, executable, or continuously evolving research assets. Instead, it curates metadata, structure, and durable links to externally hosted artifacts (e.g., code repositories, datasets, notebooks), preserving inspectability while avoiding duplication or version drift.
Artifact types
All artifacts on hbar.science are classified under one of the following canonical types:
- Algorithm — a formally specified procedure, update rule, or method; the intellectual artifact independent of any particular implementation
- Code — executable or inspectable implementations, scripts, or pipelines
- Dataset — raw or generated data used for analysis or evaluation
- Analysis — derived results such as simulations, benchmarks, parameter sweeps, or evaluation outputs
- Protocol — formally specified procedures defining how experiments, simulations, or evaluations are conducted
- Model / Framework — conceptual or mathematical structures with explicit assumptions and scope
- Tool / Application — an interactive interface (web app, CLI, dashboard) that enables use, exploration, or presentation of a scientific object; typically has a deployed URL and source link
- Collection / Bundle — a structured index pointing to multiple related artifacts (e.g., thesis + app + pipelines), used to present a coherent project while keeping sub-artifacts independently inspectable
These categories are intentionally minimal and stable.
Specific implementations (e.g., notebooks, benchmarks, proof sketches) are treated as instances of these types, not as separate categories.
Note: "Paper" and "Essay" are formats, not canonical artifact types. A paper is typically a model or analysis artifact with an associated publication link. An essay is a synthesis artifact in the Synthesis zone.
See also: Method
The Future Evolution of hbar.science
Plausible directions for platform evolution: living artifacts, AI transparency, replication workflows, and lightweight stewardship.
GPU Concurrency Benchmark — vLLM Serving Box (RTX PRO 6000 Blackwell)
Measured the concurrent-request ceiling of a single RTX PRO 6000 Blackwell Server Edition GPU serving Llama 3.3 70B Q4_K_M GGUF under vLLM 0.21.0, to replace the ~25-user-per-box estimate in the sovereign-compute architecture with a measured number. Result: as-deployed ceiling is ~5 in-flight requests, falsifying the architecture-doc hypothesis 5×; the gap is attributable to software configuration (GGUF format, fp16 KV cache, FlashAttention v2, native PyTorch sampler) rather than hardware. The optimization path (AWQ, fp8 KV cache, FlashAttention 3) is projected to raise the ceiling into the 15-25 range.
HOPSO: Harmonic Oscillator–Based Particle Swarm Optimization (Reference Implementation)
Reference Python implementation of the Harmonic Oscillator–Based Particle Swarm Optimization (HOPSO) algorithm, corresponding to the published PLOS ONE article.
!
Compliance: 4/5 (80%)
Missing: Environment hash documented
HOPSO — VQE Optimizer Implementation
Reference implementation of the HOPSO classical optimizer adapted for use in variational quantum eigensolver (VQE) workflows.
!
Compliance: 1/2 (50%)
Missing: Deterministic seed documented
Institutionalized Science: Incentives, Access, and the Limits of the Journal Model
An analysis of structural constraints in contemporary scientific publishing and institutional research.
Sovereign LLM Serving on H200 — a 4-Model Concurrency Curve (70B, gpt-oss-120b, DeepSeek-V3-671B)
Measured the concurrent-request ceiling, time-to-first-token, and throughput of four open models served under vLLM 0.22 on NVIDIA H200 (Hopper) hardware, fp8 weights + fp8 KV cache, to characterize how cheaply a sovereign actor can serve frontier-class models privately. Headline finding: serving concurrency is governed by ACTIVE parameter count, not total size. gpt-oss-120b (MoE, ~5B active) sustains the highest concurrency of all (214x max-concurrency, chat ceiling >300 and long-context >100), beating the dense Llama-3.3-70B despite 120B total params; DeepSeek-V3-671B (MoE, ~37B active) serves privately across one 8xH200 node (~200 concurrent chat/node tuned). The earlier ~5-concurrent ceiling measured on an unoptimized Blackwell stack was a serving-configuration artifact, not a hardware limit -- on mature H200 with an optimized stack, per-GPU concurrency reaches 46-214x depending on the model.
VQC Embedding vs Classical Embedding on QM9 — a Confound-Controlled Negative
Controlled-substitution test of whether a shallow variational quantum circuit (VQC) used as the embedding layer inside a fixed transformer beats a classical linear embedding of equal output dimension on QM9 HOMO-LUMO gap prediction. Only the embedding changes. Result: the VQC (8 qubits, depth 2) is ~13% worse than the linear baseline (0.717 vs 0.634 eV test MAE, >3 sigma); a matched classical bottleneck (Linear 14->8->64) reaches 0.699 eV, showing ~78% of the VQC's deficit is the dimensional bottleneck, not quantum structure. Across a 30-config sweep (qubits 4/6/8/10 x depth 1/2/3 x entanglement linear/circular/all), 0/30 beat the linear baseline and 0/30 beat even the matched classical bottleneck. A clean, reviewer-proof negative at this scale and circuit family.
VQE Regularization Evaluation Pipeline
Computational pipeline for evaluating regularization effects in variational quantum eigensolvers, including parameter sweeps and optimization landscape analysis.
!
Compliance: 1/3 (33%)
Missing: Deterministic seed documented
Missing: AI audit missing/incomplete