Evidence-ledger draft
Library of generated evidence-ledger research papers.
Automated research factory
Pick a direction (or paste your own), the harness fetches arXiv, triages papers, builds an evidence ledger, full-text verifies every quote against the PDF, runs a 5-round audit, and drafts the paper. Every claim links back to a sourced paper ID and a verified quote.
Papers
11
in this release
Submission ready
7
passed 5-round audit
Supported claims
47
linked to source IDs
Full-text quotes
51
verified against PDF
Avg novelty
0.97
0–1, vs prior productions
Avg correctness
0.72
0–1, quote ↔ PDF ↔ claim
Avg demo proof
0.77
0–1, claims independently re-verified
Avg research value
0.43
0–1, gap × contradiction × surprise × recency
Publishable
2
product-health clean
Avg first TTV
0s
time to first evidence ledger
Avg draft TTV
2s
time to paper draft
Grouped by discipline / category. Cards sorted by research-value score (highest first), then readiness, then recency.
1 paper · 1 ready for submission
ad-hoc
Systems in Open Problems in Pre- And Post-Ln Architectures: An Evidence-Ledger Investigation increasingly promise longer-horizon, higher-autonomy workflows, but their outputs are difficult to trust when claims are not explicitly tied to evi…
1 paper · 0 ready for submission
auto-deep-llms-layer
Modern deep LLMs accumulate dozens of transformer layers, yet evidence repeatedly shows large fractions of layers are skippable, prunable, or contribute marginally — but the conditions, depth-vs-quality tradeoffs, and architecture-specific…
1 paper · 1 ready for submission
ad-hoc
Systems in Liquid Neural Networks and Continuous-Time Models increasingly promise longer-horizon, higher-autonomy workflows, but their outputs are difficult to trust when claims are not explicitly tied to evidence. This draft synthesizes ta…
1 paper · 0 ready for submission
ad-hoc
Systems in Autonomous Research Harnesses increasingly promise longer-horizon, higher-autonomy workflows, but their outputs are difficult to trust when claims are not explicitly tied to evidence. This draft synthesizes taxonomy-scoped eviden…
1 paper · 0 ready for submission
auto-especially-rope
RoPE has become the default positional encoding for modern LLMs, yet recent papers report that explicit rotary signals interact unpredictably with implicit positional biases from attention, normalization, and massive activations — producing…
1 paper · 1 ready for submission
custom-transformer-attention-variants
Attention has fragmented into MHA/MQA/GQA, sparse windowed DSA/SWA/CSA, linear/state-space Mamba, Delta, KDA, and hybrid attention; claims about quality parity, throughput, and long-context behavior are scattered and not directly comparable…
1 paper · 1 ready for submission
custom-transformer-ffn-and-moe
FFN dominates parameter count in transformer LLMs, and MoE/DeepSeek MoE/shared-expert variants now claim large efficiency wins, but sparse vs dense comparisons mix training-cost, serving-cost, and quality claims in inconsistent ways. This d…
1 paper · 1 ready for submission
custom-transformer-normalization
Transformer normalization has fragmented into LayerNorm, RMSNorm, Pre-Norm, Post-Norm, and QK-Norm variants, with overlapping stability and training-speed claims that are not directly comparable. This draft synthesizes taxonomy-scoped evide…
1 paper · 1 ready for submission
custom-transformer-position-encoding
Modern LLMs claim long-context capability, but position encoding choices sinusoidal, RoPE, YaRN, NoPE are reported with mixed evidence and incompatible benchmarks, making it hard to compare claims about extrapolation and stability. This dra…
1 paper · 1 ready for submission
custom-transformer-residual-connections
Recent LLM labs DeepSeek, Kimi, ByteDance report modifications to residual connections in or around the FFN, but the design space is small and the evidence is scattered across system papers and ablations. This draft synthesizes taxonomy-sco…
1 paper · 0 ready for submission
ad-hoc
Systems in Medical AI and Clinical Decision Support increasingly promise longer-horizon, higher-autonomy workflows, but their outputs are difficult to trust when claims are not explicitly tied to evidence. This draft synthesizes taxonomy-sc…
Open problems extracted directly from the cached corpus — what the field itself says is unsolved. Highest scientific value lives here.
Future work could explore incor- porating common noise-handling strategies into the agent’s prompt to guide its data analysis.
Generated 2026-06-05T10:03:39+00:00. Duplicate production is blocked by direction, normalized title, and evidence-paper fingerprint.