Your library. Your byline. Your citation trail.
Titrate is being built so that a position cannot exist without the citations attached to it — and so that a disagreement in the literature stays a disagreement.
Titrate is pre-launch. What is described here is what it is being built to do, with Jeff Casebolt, PhD.
You have already started adding the asterisk yourself.
Hallucinate is now an ordinary word among practitioners, not a machine-learning term — people describe tools inventing DOIs, making up quotes, and confidently reporting studies that do not exist. The response is not to stop using the tools. It is to check everything by hand afterwards and to mentally attach a quiet may-not-be-correct to anything a tool hands back. In 2026 fabricated references turned up in peer-reviewed literature at a scale that drew national coverage — one scan of a major conference's accepted papers confirmed a hundred hallucinated citations across fifty-one of them. If your name goes on the claim, that manual check is not optional, and no accuracy claim removes it.
What that means structurally
01
A position carries its citations as part of the record — each claim points at the article it came from, with the formatted reference alongside it.
02
A position belongs to a topic, so it is written against a body of reading rather than a single paper.
03
Nothing generates a conclusion for you. There is no chat window, and no confident paragraph appears that you did not write.
A real topic stack
Seven papers. Not one of them supports the claim.
A stack is not a folder. It is a question, with everything read against it assembled underneath. This one is worth looking at because of how it comes out.
The central question
Does prescribing training load from real-time bar-velocity feedback produce better strength, power or speed adaptations than traditional percentage-of-1RM prescription?
| Paper | Year | Design | N | Bears on the question |
|---|---|---|---|---|
| Banyard, Nosaka & Haff | 2017 | Reliability study | 17 | contradicts |
| Guerriero et al. | 2018 | Narrative review | — | mixed |
| Dorrell, Smith & Gee | 2020 | RCT (6-week intervention) | 16 | mixed |
| Liao et al. | 2021 | Systematic review / meta-analysis | 6 studies / 124 | contradicts |
| Włodarczyk et al. | 2021 | Systematic review | — | mixed |
| Orange et al. | 2022 | Systematic review / meta-analysis | 4 trials / 88 | contradicts |
| Hickmott et al. | 2022 | Systematic review / meta-analysis | 15 studies | mixed |
0 supporting · 3 contradicting · 4 mixed
Why this is the stack worth looking at
Zero of seven papers support the claim outright, and the reliability study undercuts the method from underneath: an athlete's actual 1RM reproduces across sessions almost perfectly, but the bar velocity recorded at that 1RM does not — a coefficient of variation of 22.5%. One of the mechanisms the method rests on rests on a number that moves. Meanwhile four papers still find velocity-loss thresholds genuinely useful for managing fatigue. Both of those things are true, and neither survives being averaged.
Ask a tool that summarises and you will get a paragraph, because producing a paragraph is what it does. It cannot hand back a zero. A stack can, and this one does.
Every paper listed here is real and independently re-verified on 2 August 2026 against PubMed, PMC, the DOI resolver, Unpaywall and an open citation-graph API. The screen is not. Titrate is pre-launch, and the topic-stack view is designed, not built — this is the specification, shown with real material rather than mocked up. Labelling how a paper bears on the question is a design commitment, not a shipped feature, and it is not what makes Titrate different.
The whole seed corpus
Every paper behind this site, in one place you can sort.
The stack above is one of six. Below is all 38 papers across all of them — filter by how a paper bears on its question, sort by year or sample size, and open any row for the journal, the design, the authors’ own framing of the finding, and a link that resolves.
38
papers, each with a resolvable identifier
6
genuinely contested questions
2006–2025
publication span
23
distinct study designs
All 38 papers were independently verified against PubMed, PMC, the DOI resolver and Unpaywall on 2 August 2026; 38 carry a DOI and 37 a PMID. This table is generated from that record rather than retyped, and regenerating it re-checks every count against the corpus index (last generated 2026-08-25). The papers are real and the identifiers resolve. Titrate is pre-launch: this is the material it is being built to organise, not a picture of the product organising it.
When the evidence disagrees with itself
Supporting and contradicting work is kept where you can see both, and Titrate is being built so a position never quietly resolves a disagreement the literature has not resolved. This is a design commitment for a pre-launch product, not a shipped feature — and it is not what makes Titrate different. What makes it different is whose name ends up on the output.
What you can actually open, stated before the pitch
One question, on PubMed
Papers indexed on the question
269100%
Free to read somewhere
9736%
In the redistributable open-access subset
3613%
Take one contested question — does passing objective return-to-sport criteria reduce second-ACL-injury risk. PubMed indexes 269 papers on it. Ninety-seven are free to read somewhere; thirty-six are in the redistributable open-access subset. Retrieved 2 August 2026, reproducible in a minute. So 64% of the literature on that question can only ever reach you as a citation — and no product in this category changes that, because none of them redistributes paywalled full text and none of them tells you the proportion before you search. Titrate will not either. What it will do is say so: a paper record is a PDF where a free copy exists and a citation where one does not, a field it could not retrieve says unknown, and when it cannot find you a copy it says we did not find a copy you can open rather than unavailable — because the first is a fact about the search and the second is a claim about the world that we would be wrong about often enough for you to notice. Three of Jeff Casebolt's own five publications are paywalled. He cannot hand you those either.
The queries these numbers came fromQuestions
I pay for a research review specifically because there is no AI in it. Why would I use software?
That is a reasonable position and it is close to the one this is built on. Nothing in Titrate reads a paper for you, summarises it for you, or writes your position. There is no chat window and no generated paragraph anywhere in it — the judgment is human, it is just yours instead of an editorial team's. What the software does is the clerical half: hold the papers, keep a question attached to them, and keep every claim's citation on the record itself rather than in a second app you have to remember to update.
Will it hallucinate citations?
It does not generate citations at all. Citations come from the articles you put in your own library, and a claim points at one of them. If a paper is not in your library, there is nothing for a claim to point at.
Other tools already label supporting and contradicting evidence. How is this different?
They do, and they do it well — at the level of a search result. Titrate is not trying to win that comparison. The difference is what you walk away with: a position in your own library, publishable under your name.
Can I check the sources myself?
That is the intended workflow. Every claim keeps its link to the article and its formatted reference, so verifying is one click rather than a request for trust.
If it finds papers for me, where do the PDFs come from?
From places allowed to give them to you: the open-access copy on PubMed Central, a publisher's own free version, an author's accepted manuscript in a repository. Never Sci-Hub, never a shadow library, never a scraped copy of a paywalled article. Where none of those exists you get the citation and the DOI and you go through your own institutional access, which is the same route every legitimate tool in this category takes. If your first thought on reading that Titrate can find papers was that it had to involve something dubious, that was the right instinct, and this is the answer to it.
Does Titrate search better than PubMed?
No, and it is not being built to. PubMed already does five of the six things the spec asks for — subject headings, author field, journal field, date ranges, a result-count control — and it is free; Semantic Scholar does author disambiguation better than any of them and is also free. Titrate has no index of its own and would be querying the same public corpora those tools use. The reason to bring a topic search inside Titrate is not that the search is better. It is that the result lands somewhere: in a library that is yours, attached to a question, ready to carry your name out the other side.
Titrate is pre-launch.
Join the practitioner waitlistOne email when the first practitioner cohort opens. Nothing else.
