I
Run it, don't cite it.
A claim becomes a spec, and a spec gets run. Better to simulate something and be wrong than to cite it and never check.
Oriphim
Sight is a gift.
The problem
A claim is not a result.
Most computational work in science and engineering has never been checked. Not because researchers are careless, but because checking it properly costs months that nobody has. So results get published, cited, and built on, and the question of whether the mathematics was actually solved correctly goes unasked.
Oriphim asks it. You describe a physical system, or hand over a paper. Oriphim determines what system is being posed, derives the governing equations, builds the model, runs it, and subjects the result to every check reality permits. Then it assembles the argument: why these equations, what regime they hold in, what was assumed, what was neglected, and what the numbers do and do not license you to conclude.
Evidence
Two tiers of evidence, never conflated.
Verification asks whether the mathematics was solved correctly - error budgets, convergence under refinement, conservation, integrator agreement, and whether the conclusion survives perturbation of every free parameter. It needs no reference data, so it's available for any system you can describe.
Validation asks how far the model sits from reality, with uncertainty carried on both sides. It requires reality to compare against, and it's earned one domain at a time.
Oriphim states which tier a report belongs to, on the report. It does not claim a model is correct. It claims the model was solved to a stated accuracy, or that it is wrong by a quantified amount under stated conditions. Both are defensible. Neither is “good agreement.”
Principles
What we hold to.
II
Local-first, always.
Your data and models run on your own hardware by default. Cloud is something you opt into, never something you discover later.
III
Critique is our job, not yours.
A plausible result that is silently wrong is the failure mode worth engineering against. Finding it is the product, not an advanced feature for users expert enough to catch it themselves.
IV
Judgment stays with a person.
Oriphim produces the model, the run, the checks, and the argument in hours instead of months. A named person reviews it, corrects it, and owns it. Every correction is captured.
V
Shipped and roadmap are labeled separately.
Everywhere on this site, including here.
Where this is right now.
Oriphim is built by one person, aimed at a mission that spans fields. The gap is deliberate: the engine gets built first, the team comes after. Everything on this site is either running today or explicitly marked roadmap. There is nothing in between.