Generating a full scientific paper from a prompt has become almost trivial for large language models, but until now the price was reliability: experimental results shaped by chance, phantom citations, and a veneer of coherence that collapsed under scrutiny. Prompt-to-Paper, a multi-agent system just presented by a research team, attacks this problem at the root with three innovations that shift automated scientific writing from linguistic play to computational transparency.

The first pillar is a deterministic Retrieval-Augmented Generation pipeline: every claim is tied to a corpus of 60-100 real papers, selected through section-aware relevance scoring and snowball citation expansion. No hidden probabilities, no hallucinations — the reference literature remains traceable and verifiable. The second element is an autonomous coding agent that runs real computational biology experiments, replacing synthetic outputs with genuine numerical results. The third piece is an eight-dimensional automated quality scorer, benchmarked against statistics from published papers and augmented with explicit hallucination penalties. The quality score drives a continuous improvement loop: a context-rich reviser routes each iteration to one of three research actions, and every ten steps a deep research cycle re-runs experiments and regenerates the manuscript from the strongest outputs.

Validation on five bioinformatics case studies is clear: all resulting PDFs are submission-formatted and contain zero out-of-range citations. The improvement loop raised average manuscript quality by +17.96 points on a 0-100 scale, with a maximum boost of +26.04. A human reviewer scored the manuscripts at an average of 7.0 out of 10. And all this costs about 0.31 USD per paper.

Numbers like these signal a structural shift for scientific publishing. This is not the usual co-pilot packaging sentences; here, the system delivers the entire chain — from running the experiment to bibliographic justification — at a cost low enough to become a commodity. The most revealing detail for those thinking in terms of local deployment is the deterministic nature of the pipeline. If hosted on on-premise hardware, Prompt-to-Paper (or a similar framework) could guarantee complete reproducibility: same question, same corpus, same experimental code, same evaluation. In labs handling sensitive data or proprietary algorithms, infrastructure sovereignty would become the key to ensuring scientific integrity without depending on external APIs that drift over time.

Of course, the risk of saturation remains open: if a credible paper costs less than a cup of coffee, journals may be flooded by electronically impeccable preprints with no real value. Yet the framework inverts the hallucination dynamic: instead of fabricating facts, it runs computations and anchors them in the literature. In a landscape where trust in machine-generated content is the scarcest resource, that is perhaps the most significant innovation.