Self-Refine: Iterative Refinement with Self-Feedback
arxiv.org/abs/2303.17651- Supports explicit iterative feedback loops.
- Informs design iteration and candidate refinement.
- Does not justify trusting unsupported self-critique by itself.
Research bibliography
Sources behind the Jaysearch ERA loop.
These sources support the research, design-review, candidate-evaluation, provenance, and recommendation loop. They justify design pressure, not every exact implementation choice or scoring weight.
Agent feedback and critique
Iterative review should be explicit and tool-grounded
Reflexion: Language Agents with Verbal Reinforcement Learning
arxiv.org/abs/2303.11366- Supports feedback plus retained context for agent improvement.
- Informs memory-aware review and iteration loops.
- Does not replace external validation or governance gates.
CRITIC: Tool-Interactive Critiquing
arxiv.org/abs/2305.11738- Supports tool-grounded critique over unsupported introspection.
- Informs design-review automation and evidence-backed review.
- Motivates critique through artifacts, packets, validators, and tests.
Software evaluation
Candidate quality should be task-grounded
SWE-bench: Can Language Models Resolve Real-World GitHub Issues?
arxiv.org/abs/2310.06770- Supports task-grounded software evaluation over plausibility-only scoring.
- Informs evaluation and recommendation packets.
- Motivates requiring evaluation records before recommendation.
AlphaEvolve: A coding agent for scientific and algorithmic discovery
arxiv.org/abs/2506.13131- Supports evolutionary generate/evaluate/select loops for code candidates.
- Motivates separating implementation attempts from selected artifacts.
- Does not solve decomposition, governance, or provenance by itself.
ERA and computational discovery
Empirical coding loops need searchable candidates and measurable outcomes
Google Research: ERA from Nature publication to Computational Discovery
research.google/blog/empirical-research-assistance-era- Supports the design relevance of an empirical coding loop that searches literature, writes code, explores solutions, combines techniques, and evaluates results.
- Informs Jaysearch's separation between research intake, candidate generation, evaluation, and selected solution artifacts.
- Does not imply Jaysearch has equivalent scientific benchmark performance or Gemini-backed capabilities.
An AI system to help scientists write expert-level empirical software
nature.com/articles/s41586-026-10658-6- Supports the ERA framing of expert-level empirical software generation driven by a quality metric.
- Informs the generate/evaluate/select pattern and the value of tree-search-like exploration over candidate implementations.
- Motivates explicit evidence and metric capture for implementation attempts.
Google Research ERA code and experiments
github.com/google-research/era/tree/main/era_applications- Provides a concrete public artifact surface for ERA applications and experiments.
- Motivates keeping code, experiments, manuscripts, and evaluation evidence linkable from the bibliography.
- Supports Jaysearch's preference for source-backed artifact references over prose-only claims.
Governance foundations
Rules, incentives, and stopping conditions need structure
Nash: Equilibrium Points in N-Person Games
pubmed.ncbi.nlm.nih.gov/16588946- Informs the platform's game framing.
- Does not determine platform payoffs or scoring weights directly.
Maskin: Mechanism Design
nobelprize.org/maskin/lecture- Supports explicit rule and incentive design.
- Informs anti-cheat and governance framing.
Myerson: Perspectives on Mechanism Design
nobelprize.org/myerson/lecture- Supports mechanism-design framing.
- Informs separation between outcomes, rules, and allowed moves.
Simon: A Behavioral Model of Rational Choice
academic.oup.com/qje/article/69/1/99/1919737- Supports bounded-rationality assumptions.
- Informs bounded search budgets and explicit stopping rules.
Provenance, state, and inspection
Traceability and review are first-class constraints
W3C PROV-Overview
w3.org/TR/prov-overview- Supports provenance-first packet and artifact references.
- Informs traceable DAG and handoff contracts.
W3C PROV-DM: The PROV Data Model
w3.org/prov-dm- Supports structured provenance modeling.
- Informs source refs, artifact refs, and graph relationships.
Harel: Statecharts
sciencedirect.com/science/article/pii/0167642387900359- Supports explicit state and transition modeling.
- Informs governance transition legality and orchestration state.
Fagan and Myers: Software inspections
research.ibm.com/software-inspections- Support structured inspection and review discipline.
- Inform keeping review, validation, and execution evidence separate.
The bibliography supports design choices but does not make the platform's exact scoring weights, DAG boundaries, or tool contracts mathematically inevitable. Those remain Jaysearch design decisions and should be labeled as design inferences when they extend beyond the cited sources.