Read surface: windowed event counts (per-actor, per-kind) + section co-usage aggregates

[ohdearquant]

Problem

Two consumer designs need aggregate read surfaces that do not exist. Events are written (memory recall/remember, kg create/link, brain feedback) but there is no sanctioned aggregate read: today the only options are per-record verbs or direct SQL (banned surface).

Ask A — windowed event counts

A read verb returning event COUNTS grouped by actor and kind over a time window, e.g.:

brain.events(actor?, kind?, since, until?) -> {counts: {kind: n, ...}, by_profile?: {...}}

• Kinds: memory.recall, memory.remember, kg create/link, brain feedback (feedback splittable by served_by_profile_id).
• Consumer: fleet flywheel-metrics reporting (recall-before rate, persist-after rate, feedback coverage per profile).

Ask B — section co-usage aggregates

A read surface answering: which atom-section types are served together and co-earn positive section_signals feedback, over a window.

• Consumer: domain auto-generation from usage (domains as patterns of co-used, co-liked atom sections).
• Likely home: knowledge pack (section serve/feedback events already flow through compose + knowledge.feedback).

Notes

• Both are read-only Assertive verbs; no new write paths.
• Verb surface verified 2026-07-08 (75 verbs): no existing verb covers either ask. brain.profile returns snapshot/state summary, not windowed event aggregates; knowledge.feedback is write-only.
• Consumers can carry a pending-surface placeholder; not launch-blocking.

[ohdearquant]

Consolidation requirement (from the resource-management SPEC directive, same day): the resource-management subsystem's observability section lands resource events in the same event plane this issue's Ask A reads. There must be ONE windowed per-actor/per-kind event-read surface serving both the flywheel metrics consumer and resource-management observability — not sibling verbs. Design the Ask A verb shape with both consumers at the table before impl.