Atlas Components · Manufacturing

MAIA Intelligence · Platform overview · Apr 30 – May 6

···
KERNEL
Signals → Eval → Reflect → Patch → A/B → Promote
CYCLES · 30D
184
loops completed
PROMOTIONS · 30D
41
patches that won
ROLLBACKS · 30D
7
auto-reverted in window
ACTIVE EXPERIMENTS
4
in eval / A-B / shadow
EVAL HARNESS · CASES
1,420
frozen + auto-extending
The Kernel Loop
~4-hour mean cycle time · 6 stages · gated promotion
SIGNALS
production decisions, rejections, escalations
EVAL
frozen suite + new failure cases
REFLECT
agent self-critique on misses
PROPOSE
patch: prompt · skill · LoRA · policy
A / B TEST
shadow → 1% → 5% → 25%
PROMOTE
win? promote. lose? auto-rollback.
The promise: when a customer accepts a proposal, that's a positive example. When they reject one, that's a hard negative. When they edit one, that's a directional gradient. The kernel never stops learning — and never stops being safe to do so.
Recent promotions
last 5 days
Fatigue Guardian · rotation policy
Caught 18 more fatigue breaches in shadow eval; no false positives on graveyard shifts.
v23 → v24·2h ago
+2.4 F1
Compliance Sentinel · cert renewal queue
Renewal proposals 6% more likely to be accepted; ranking now considers manager workload.
v17 → v18·9h ago
+0.06 PR
Production Watcher · downtime classifier
Mean time-to-resolution down 14% on stamping line; new sub-cause taxonomy from reflection.
v41 → v42·1d ago
−14% TTR
Briefing Composer · daily plant brief
Operators read 9% more briefings to completion after format A/B.
v8 → v9·1d ago
+9% read
Cost Watcher · OT rebalance
Catches 3 more rebalance opportunities per week without breaching fatigue floor.
v12 → v13·2d ago
+$11k/wk
Active experiments
4 in flight
ABFatigue Guardian · LoRA-12-fatigue-v3+2.1 F1
traffic: 5%vs v24promote in 4h
ABShift Market · DSPy auto-prompt v3−0.2 PR
traffic: 1%vs v15rollback in 1h
PROPOSEBriefing Composer · format C
traffic: 0%vs v9shadow eval
REFLECTCompliance Sentinel · taxonomy v2
traffic: 0%vs v18patch drafting
Auto-rollbacks
regression-gated
Maintenance Optimizer · predictive PMv22 → v21
Regressed on stamping line; eval F1 dropped 0.04 vs baseline.
5h ago
Safety rails
policy contracts
EVAL GATE THRESHOLD
≥ baseline + 1σ
AUTO-ROLLBACK WINDOW
60 min
MAX TRAFFIC ON UNVERIFIED
5%
SHADOW EVAL CYCLE
every 4h
FROZEN-EVAL SUITE SIZE
1,420 cases
PRIVACY BOUND
rule-only · per-tenant LoRA
No ungated change has reached production in the last 184 cycles.