Horizon DigitalKaïaui-v2

Improvements

What the autonomous self-improvement loop has shipped. Sourced from [self-improve] commits — fires twice daily at 10:00 + 22:00 AWST.

Needs you — not in Plans

Stale gates, features built but switched off, and actionable health flags.

Plans

Deep-dives for items too big to ship in one self-improve cycle. Answer the open questions → mark ready → self-improve picks it up.

🙋 Operational Actions

Operational work the self-improve cycle prepares for you.

Kaïa recommendations
Peer reviews
Skill trust
Shipped-work re-verification
Adversarial reviews
Automation review queue
Learning flags

Operator-curated corrections + auto-promoted health-check flags.

🚫 Learned Sender Blocklist

Repeat-IGNORE marketing senders learned for one-click suppression.

Version drift

From scripts/version-check.mjs — read-only mirror.

Built but off

Shipped-but-config-gated features — what's built but switched off.

LLM route benchmarks

Scenario regression bench — per-route pass rate from data/bench-reports/, run by the self-improve cycle and on route model changes.

Run a model review

Benchmark one route's incumbent against the candidates you pick — results land as swap recommendations below, nothing changes without approval

Model swap recommendations
Feature usage

From lib/feature-usage weekly rollup — read-only mirror.

AI adoption

Adoption staircase per department — champions, movement, unsticking packs.