01 // case study results

What the verification system produced

94,777 words synthesized

17 structured chapters

1,669 papers ingested

95% doi verification rate

The domain specialist reviewed the output and rated it "excellent text." The point of this case is not autonomous writing. It is that a large, literature-heavy draft can be checked, structured, and surfaced for expert signoff without hiding the weak spots.

02 // the pipeline

Researcher directs. The verification system does the mechanical work.

scope + structure

The researcher defines the topic, scope, and chapter structure. The engine designs search queries, ingests papers from PubMed and Semantic Scholar, and builds the evidence corpus. 1,669 papers, 97% with DOIs.

multi-model synthesis

Each chapter receives 25 relevant papers and passes through the full verification pipeline: producer, DOI gate, CTO audit, reviewer audit, feedback moderator, revision pass. No model reviews its own output.

validation + delivery

The researcher validates the output against domain knowledge. Optional bilingual delivery (this case: EN → RU with per-chapter linguistic audit). The engine compiles; the specialist approves.

$ axion-academic --domain "physical rehabilitation medicine"
                   --topic "physical factors in surgical wound healing"
                   --scope "metabolic syndrome patients"
                   --chapters 17 --rigor standard

[structure]  4 parts, 17 chapters, 53 queries designed
[corpus]    1,669 papers ingested · 97% with DOIs
[synthesize] chapter 1/17... done (4,195 words, 10m)
[synthesize] chapter 2/17... done (4,850 words, 10m)
[synthesize] ...
[synthesize] chapter 17/17... done (4,075 words, 15m)
[verify]    DOIs: 300+ verified (95% avg)
[localize]  EN → RU: 17/17 chapters... done

status: ready for specialist review
total: 94,777 words · ~380 pages · 6 hours pipeline time

03 // multi-model verification

Verification is the product

Three different AI models in independent verification configuration. The producer never sees its own audit. The auditors never coordinate. Conflicting feedback is synthesized by a separate moderator. This is the same pipeline that powers axion-research and our compliance-facing review work.

the verification pipeline

Producer: synthesis model — synthesizes from corpus

DOI Gate: reference verifier — every citation verified before proceeding

CTO Audit: adversarial model — attacks claims, flags gaps

Reviewer: audit model — structural and compliance audit

Moderator: synthesis model — resolves conflicting feedback

Specialist: Human domain expert validates final output

what verification catches

Fabricated DOIs — replaced with verified references

Mechanism gaps — biophysical claims without citations

Evidence hierarchy — in vitro cited where RCT data exists

Clinical inaccuracy — dosing, contraindications, protocols

Structural drift — chapters deviating from researcher's outline

Statistical claims — prevalence data without source

04 // per-chapter data

17 chapters across four parts

Per-chapter synthesis data
Ch	Topic	Words	Time	DOI Rate
1	Introduction — The Clinical Problem	4,195	18m	92%
2	Normal Wound Healing	4,850	20m	—
3	The Metabolic Wall — Pathophysiology	6,541	24m	100%
4	Classification and Assessment	4,709	19m	100%
5	Photobiomodulation (LLLT)	6,494	26m	100%
6	Magnetotherapy (PEMF)	5,572	22m	100%
7	Therapeutic Ultrasound	5,705	23m	100%
8	Electrical Stimulation	6,531	25m	94%
9	Oxygen Therapies (HBO + Ozone)	6,391	24m	86%
10	Emerging Modalities	6,177	23m	100%
11	Combination Protocols	6,560	27m	100%
12	Perioperative Physical Medicine	5,379	21m	100%
13	Monitoring and Assessment	5,803	22m	100%
14	Safety and Contraindications	5,114	20m	100%
15	Evidence Synthesis	6,057	24m	95%
16	Future Directions	4,624	18m	67%
17	Conclusions	4,075	16m	67%

Chapters 16-17 show lower DOI rates because future directions and conclusions rely on synthesis rather than direct citation. The gate flags this but accepts it within the expected range for those chapter types.

05 // the economics

Six months of mechanical work in six hours

6h total pipeline time

17 chapters synthesized

95% doi verification rate

2 languages delivered

This doesn't replace the researcher. The domain specialist defined the scope (17 chapters, 4 parts, specific modalities), provided the clinical context, and validated the output. What the engine replaced was the mechanical work: reading 1,669 papers, extracting evidence, structuring chapters, verifying citations, and formatting references. That work takes a human 3-8 months. The engine completed it in 6 hours.

06 // the research domain

Physical rehabilitation medicine

The evidence synthesis covers physical therapeutic modalities for surgical wound healing in metabolic syndrome patients — a domain at the intersection of biophysics, clinical medicine, and rehabilitation science.

photobiomodulation low-level laser therapy

magnetotherapy pulsed electromagnetic fields

therapeutic ultrasound tissue repair acceleration

electrical stimulation bioelectricity + wound repair

oxygen therapies hyperbaric + ozone

metabolic syndrome impaired healing phenotype

The engine adapts to the domain it reads. Physical rehabilitation, psychedelic pharmacology, oncology immunotherapy — the same verification pipeline can operate across corpora, but the human specialist still owns the final judgment.

what you gain

Zero hallucinated citations. Every reference verified against source corpus.

what it costs

Slower than ChatGPT. Synthesis sessions take 4-8 hours of compute. Human review required for final output.

07 // next step

Start from the entry page that matches your review pressure.

Use research for submission and literature-review pressure, or clinical for doctor and biomedical-team use cases. This page is supporting proof, not the main route.

[ submit claim ]

A systematic evidence verification proof case.