Evidence policy
Every relationship in OptAtlas is a claim.
Every claim has
- source(s)
- an evidence grade (A / B / C / D / X)
- an equivalence level (E0–E4)
- a last-verified date
- a caveat, where applicable
Evidence grades — source reliability
Evidence and equivalence are independent axes. A strong claim (E0) can rest on weak evidence (C); when it does, it is shown that way.
- APeer-reviewed / official
- BSurvey / tutorial / community
- CVendor / case-reported
- DExpert inference
- XUnverified / draft
Equivalence levels — mathematical similarity
- E0_exact — E0 · near-identical formulation
- E1_variant — E1 · same base, different side constraints
- E2_method_shared — E2 · shared methodology
- E3_analogy — E3 · structural analogy
- E4_inference — E4 · inferred, needs verification
Forbidden vs preferred phrasing
| Avoid | Prefer |
|---|---|
| “This problem is the same as …” | “This is an E1 variant of …” |
| “Solved by reinforcement learning” | “Reported with RL-based approaches under benchmark conditions” |
| “Used in shipyards” | “Vendor-reported / case-reported usage” |
| “SOTA” | “Best reported result under this benchmark / protocol” |
This policy is what keeps OptAtlas from reading like a blog: relationships are recorded with their verification state, not asserted as fact.