← Mimir docs

Mimir — the back-test, on one page

13 sample products + the sync/async case, run through the architecture, forced 18 requirements. Every product forces at least one.

● forced this requirement ○ confirms / exercises it ∗ = spans all products (not a single forcer) hover any header or cell for detail

The 13 products (+ the execution case)

The 18 requirements

Verdict: the engine + packs split held for all 13 products — each is "collect → understand → check → cost → decide," composed from the same services with a different pack. The back-test added one capability service (Estimate), a second orchestration mode (Agent Runtime), four shared/cross-cutting components (Model Gateway, Reference Data, Entity/Link Graph, Capture-Quality), a dual sync/async execution model (R17), and a Compliance plane (R18). The one existential risk is R4 — domain vocabulary leaking into the engine — guarded by the generic Signal model (R2) and the pack boundary.

Mimir · the Modules intelligence engine — source: docs/modules/03-products-and-usecases.md