Rigorous AI-assisted academic paper authoring for every research format — from sharpened question to venue-ready PDF.
Precise questions, locked methodology, honest reporting against yourself, proper citations — researchers learn these habits over years. Everyone else reinvents the mistakes.
LLMs default to confident prose. Without guardrails, exploratory findings get reported as confirmatory — and causal language replaces correlation.
NeurIPS, ACL, IEEE, USPTO, NSF — every venue has different templates, page limits, and compliance checklists. Each one is a manual burden.
Scientific rigor is a set of habits. This bundle encodes the habits — so a patent attorney, a policy analyst, or a junior researcher drafting their first workshop paper all get the same scaffolding and the same honest critic.
Six disciplined modes walk you from raw intuition to venue-ready output — with pre-registration lockdown and honest-pivot enforcement baked in.
Sharpen a rough claim into a falsifiable research question with explicit predictions, named mechanisms, and disconfirmation criteria.
Lock your methodology before seeing results. Hash-sealed pre-registration with power analysis, named tests, and alpha levels.
Run the analysis, gather prior art, or pull evidence — depending on what you're making. Execution log tracks every step.
Honest peer review names specific limitations. Draft produces venue-structured output. Publish blocks until critique has run and honest-pivots are acknowledged.
Full IMRAD structure with CONSORT/STROBE compliance, pre-registration, and honest-pivot enforcement.
Dataset releases, method comparisons, and performance evaluations with protocol definition and fairness checks.
Funding applications for NSF, NIH, DARPA. Feasibility scoring, well-powered research plans, preliminary data emphasis.
USPTO-style invention disclosure with prior-art section and claim chart for filing attorneys.
Systematic review with PRISMA compliance, multi-database search logging, and evidence synthesis.
Technical thesis defense for decision-makers with comparative analysis and practical case studies.
Evidence-based briefs for policymakers with counter-argument coverage and implementation feasibility.
Reproducibility research measuring deviations against original claims. Honest-pivot measures drift, not fresh hypotheses.
You — any skill level, any venue, any document type
Routes intent to the right mode/agent, enforces honest-pivot
NeurIPS, ICML, ACL, IEEE, ACM, arXiv, USPTO, NSF, policy memo
Hypothesis designer, methodologist, statistician, preregistration reviewer — sharpen the question and lock the method.
Paper architect, technical writer, citation manager, figure designer, venue formatter — structure and produce the artifact.
Honest critic, ML paper reviewer, literature scout, idea generator — review, challenge, and strengthen the work.
Hash-sealed study plans before you see results. The preregistration-reviewer checks predictions are specific and directional, tests are named, alpha levels set, MDEs calculated. Issues pass/fail/warn per section.
If your data contradicts the pre-registration, exploratory findings are automatically labeled as such. /publish blocks until pivots are acknowledged. No silent overclaiming.
Multi-agent figure generation with 8 quality veto rules: vector format, readable text (≥8pt), colorblind-safe palette, error bars, white background, proper aspect ratio, no misleading axis truncation, unobstructed legends.
Mitigates reflexivity hazards by routing critique through a different LLM vendor than the one that drafted. No model grading its own homework.
The honest-critic issues BLOCK/WARN/NOTE findings with severity levels. BLOCK items must be resolved before /draft. It argues against the work: overclaiming, methodology gaps, alternative explanations, limitation specificity.
Experiment audit, power analysis, provenance check, stage analyzer, block-hypothesis, resume/repair, PaperBanana — each with CLI and tests.
~4,946 lines of tests across 7 Python tool modules. ~13,048 lines of Python source. Real assertions, real coverage.
Honest-pivot, exploratory-labeling, PaperBanana, figure-generation, LaTeX authoring, conference-styling, cross-vendor-judge, stop-slop, and more.
Repository: michaeljabbour/amplifier-bundle-research (v0.8.5, MIT License)
agents/ directory listingrecipes/ directory listing (8 primary paper types + orchestration, idea-generation, cache-verify, smoke test, and others)context/conference-formats/ and context/venue-formats/: NeurIPS, ICML, ACL, IEEE, ACM, arXiv, USPTO, NSF grant, policy memomodules/: experiment-audit, block-hypothesis, power, provenance-check, resume, stage-analyzer, PaperBananaData collection: git log, git shortlog -sne, directory listings, file line counts via wc -l, README.md, bundle.md frontmatter, and team-knowledge YAML capability manifests.
Lineage: Thin wrappers over K-Dense scientific-agent-skills, orchestrated with Superpowers-style mode workflow, informed by Denario multi-agent topology. Gene-transfers from AI-Scientist (idea-generator, ML-paper-reviewer, literature-scout) and amplifier-bundle-scientificpaper (paper-architect).
Deck generated: May 2026. All figures are repo-verifiable counts, not estimates.
Install with one command. Every paper type. Every venue. Pre-registration, honest critique, and PaperBanana figures — built in.