An answer is only worth as much as the thing behind it
Ask a question in plain English and get an answer assembled from this site’s own structured data, with a link to every source. It handles protocols, cost, what is actually known about a compound, documented interactions, and how long anyone has really run something.
Research engine
Ask the catalogue, not a chatbot
Answers are assembled from this site’s structured data and carry links to every source. Ask the same question twice and you get the same words, because the words came from the data rather than from a model’s memory.
This engine answers from the site's own structured data rather than from a model's memory, so everything it says carries a link to where it came from. It can: · find protocols by goal, compound, or a time limit · say what a protocol costs at each vendor, including water and syringes, per day and in total · summarise what is actually known about a compound, and what is not · report documented interaction records, and say plainly when none exists · compare how long people run something against how long anyone actually has · list what is graded under a research topic, strongest evidence first It will not tell you what to take, and it will not produce a number it cannot cite.
Why this is not a chatbot
A language model asked how much BPC-157 people use will answer confidently and plausibly whether or not it knows, and that answer looks exactly like one that came from a citation. In a category where most figures have never been through a trial, that is not a small problem. It is the whole problem.
So the architecture is inverted from the usual one. The engine does not know things. It calls tools, and the tools read the same structured records every page on this site reads. Where a model is attached, it is allowed to choose which tool to call and to phrase what comes back. It is never allowed to supply the answer. The failure mode is therefore “I could not find that”, which you can recover from, rather than a fabricated number with a fabricated identifier attached, which you cannot.
Two consequences you can check yourself. Every answer names the tools that produced it. And asking the same question twice returns the same words, because the words came from the data.
What it will not do
- It will not tell you what to take. It reports what was studied, in whom, at what quantity, and what happened. The gap between that and a recommendation is the entire point of the site.
- It will not turn an absent interaction record into a safety claim. “No documented interaction” means nobody has looked, for most compounds here. A test in the build asserts this specific sentence can never come out the other way.
- It will not produce a figure it cannot cite. A tool that cannot cite its answer returns no answer.
Free allowance, and what membership adds