Observed directly
Comes from a benchmark run, public dataset, system log, source record, or validated field signal. Strongest tier, but still includes scope and freshness caveats.
Methods Library
The rules behind every public claim.
Methods is the public standard for how ALCUB3 classifies evidence, versions benchmarks, reports caveats, and decides whether a claim is ready to appear on a product, research, learning, or impact surface.
Primary job
Claim discipline
Make it clear what is measured, estimated, modeled, or planned.
Used by
Research + Impact
Platform and mission tracks use the same evidence language.
Status
Live method note
Updated as new benchmark and publication objects mature.
Evidence tiers
Every claim carries its evidence class.
The tier is part of the claim. If the tier is missing, the claim is not ready for public use.
Calculated from known inputs
Uses transparent formulas or weights over known data. Useful for scorecards and comparisons when the assumptions are named.
Projected by a model
Comes from a model, simulation, or inference layer. Useful in aggregate, handled carefully per instance, and always labeled with limits.
Planned, not shipped
Names a committed direction or planned capability without presenting it as present-tense evidence.
Do not publish
Marketing language that cannot trace to a source, evaluation, field artifact, or explicit assumption does not move forward.
Claims age
Method pages carry review expectations so stale claims are either refreshed, downgraded, or removed.
Method standards
The publication bar is operational.
| Standard | What it means | Where it applies |
|---|---|---|
| Traceability | Every serious claim links to a source, benchmark, dataset, method note, or explicit assumption. | Research, product pages, Impact, Learning. |
| Versioning | Benchmark or scoring changes identify what changed and what remains comparable. | Benchmarks, reports, methods, product claims. |
| Caveats | Limits appear near the claim, not buried in an appendix. | Impact, Higher-trust planning, platform research. |
| Ownership | Every method belongs to a domain lane and has a review owner. | Research pipeline and website updates. |
Track bridges
Methods split into platform and mission tracks.
Agent runtime evidence
Memory, execution, tool use, orchestration, approval paths, trust boundaries, and deployment quality.
Open platform methods →Public-interest evidence
Water intelligence scoring, field signals, source quality, observatory claims, and program-level caveats.
Open impact methods →Provenance discipline
Method standards depend on source standards: what can support the claim, what cannot, and how often the source gets refreshed.
View sources →