MovingOrders
← Trust Center
AI GOVERNANCE

Responsible AI

AI carries 90% of the build weight on MovingOrders. That level of autonomy requires a governance framework that is auditable, bounded, and transparent. Every AI decision on this platform operates under the Algorithm of Trust™.

Algorithm of Trust™ — Five Pillars (Required on All Agents)

A
Auditable
Every agent action writes to an immutable audit table. 7-year retention. Evidence exported to S3 before it counts as evidence.
R
Reversible
Every agent action is reversible within a defined window. No irreversible agent action without dual consent at Tier 4.
C
Compliant
Pre-execution compliance gate runs before every agent action — not as post-hoc moderation. OPSEC-FILTER synchronous on every text write.
E
Explainable
Plain-language reasoning chain exposed on every agent decision. No black-box outcomes for housing, crisis, or financial actions.
B
Bounded
Each agent operates only within its defined scope. Registry-enforced. No agent may exceed its registered capability set.

AISCO Autonomy Tiers

Every agent is assigned an autonomy tier. Tier 1 is the mandatory minimum — no agent ships below it. Tier 4 (fully autonomous, no human override) requires dual CEO + CFO consent and is subject to a standing FLAG review.

Tier 1 (80/20)
MANDATORY MINIMUM
AI handles 80% — human in the loop for 20%
Tier 2 (90/10)
DEFAULT TARGET
AI handles 90% — human override on 10% of decisions
Tier 3 (95/5)
EARNED
Requires CCS ≥ 85 and zero SENTINEL flags for 90 days
Tier 4 (100/0)
DANGEROUS OVERRIDE
Dual CEO + CFO consent only. Standing FLAG-04 on OPSEC-FILTER and CRISIS.

OPSEC Filter — RELOCATE-19 (Always On)

RELOCATE-19 OPSEC-FILTER runs synchronously before every POST or PATCH that writes text to the database. If the filter returns BLOCK, the API returns 422 and no data is persisted. This applies to original content and to all translations — a translation is content and filtered identically. The filter cannot be bypassed by any user, lister, or admin action at runtime.

Crisis Protocol — RELOCATE-20 (Always Escalates)

RELOCATE-20 CRISIS uses claude-opus-4-6 (the highest-capability model) for all crisis and DV safety interactions. Emergency contact notification is delayed exactly 60 seconds via RabbitMQ TTL — never via application sleep timers. Crisis listing boost is session-scoped only and never modifies base_score in the database.

21 RELOCATE Agents — Full Registry

All agents operate in full production mode. Admin dashboard kill switches are active on every agent.

AgentFunctionModel
RELOCATE-01 SCOUTPCS probability engineclaude-sonnet-4-6
RELOCATE-02 MATCHListing match algorithmclaude-haiku-4-5-20251001
RELOCATE-03 LEASEAI lease reviewclaude-opus-4-6
RELOCATE-04 VERIFYDEERS / identity verificationclaude-haiku-4-5-20251001
RELOCATE-05 SCOREHousing Passport scoringclaude-sonnet-4-6
RELOCATE-06 SCOREUHS score engineclaude-sonnet-4-6
RELOCATE-07 OCONUSOCONUS listings / OHAclaude-opus-4-6
RELOCATE-08 TIMELINEPCS phase timelineclaude-sonnet-4-6
RELOCATE-09 EXITDual Exit Engine optimizerclaude-sonnet-4-6
RELOCATE-10 MATCH-AGENTRealtor marketplace matchclaude-haiku-4-5-20251001
RELOCATE-11 LENDERVA loan pre-qualclaude-sonnet-4-6
RELOCATE-12 TAXWISEMilitary housing tax intelclaude-sonnet-4-6
RELOCATE-13 WHEELSVehicle / rental carclaude-haiku-4-5-20251001
RELOCATE-14 GUARDGuard/Reserve activationclaude-sonnet-4-6
RELOCATE-15 GOVMOVEFederal civilian relocationclaude-sonnet-4-6
RELOCATE-16 VOUCHERPCS travel claim / DD 1351-2claude-sonnet-4-6 (CONUS) / claude-opus-4-6 (OCONUS)
RELOCATE-17 PAYMASTERPayment railsclaude-haiku-4-5-20251001
RELOCATE-18 CONCIERGEVoice-first PCS assistantclaude-sonnet-4-6
RELOCATE-19 OPSEC-FILTEROPSEC content filter — synchronous on every writeclaude-haiku-4-5-20251001
RELOCATE-20 CRISISCrisis / DV safety protocolclaude-opus-4-6
RELOCATE-21 SURPLUS-INTELMO SURPLUS™ marketplace intelligenceclaude-sonnet-4-6