5 research spaces.
Every one carries cited evidence.
5 spaces · 4 domains · 43 targets · 94 capabilities · 433 claims · 143 sources · last capture 2026-09-15 · 43 rechecks due
Sorted by evidence, not by name. Sourced is the share of target × capability cells carrying a cited claim: it says how much has been researched, not how good anything is.
Researched5Cited claims collected. Sourced is the share of target × capability cells carrying a claim; it is a measure of research coverage, not of quality. | ||||||
| Model runtimes AI infrastructure | 4 | 14 | 42 | 75% | 10 | |
| Collaborative docs & knowledge Productivity & collaboration | 9 | 17 | 84 | 55% | 31 | |
| Vector databases | 5 | 17 | 26 | 31% | 10 | |
| Databases & storage Data & storage | 10 | 17 | 48 | 28% | 20 | |
| Agent platforms AI & agent platforms | 20 | 46 | 259 | 28% | 41 | |
A space moves from proposed to seeded to researched as evidence is collected in it, never by relabelling. Proposing a space asserts nothing about it.
Useful knowledge shows its work.
Testedknowhow connects software to the claims we can support, the environments they apply to, and the questions we still need to answer.
Start with sources
Prefer upstream documentation and pinned repository revisions. A vendor claim is recorded as documented, without turning it into a test result.
Keep the conditions
A finding needs a version, environment, scope, and check date. Missing documentation remains unknown. Changes flag a recheck; they do not prove a claim false.
Make it repeatable
An execution result needs actual commands, artifacts, failures and outcomes. Independent reproduction is a separate attestation tied to the exact revision.
Keep knowledge portable
The JSON dossier is authoritative. Markdown, comparisons, and the capability graph are derived views. Export a package and use it with your own agent.