bp Sphere Decision Memory

Decision Memory

Policy absorbs prior human decisions — effort reduces because fewer people rediscover the same answers.

Administrator · Enterprise · LIVE

Sources: bp Sphere execution ledger, Evidence Vault · Refreshed: 19 Jul 2026 19:56 UTC

Decision Memory Proof

Refreshed: 2026-07-19 19:56:31 UTC

SOURCE

bp Sphere execution ledger (iaf_agent_executions)

Scope: Org-wide learning statistics

POLICY

Learning requires approval before policy change

EVIDENCE

Evidence packs linked to prior decisions

RECORDS

Decision execution IDs (scoped when provided)

ACTION

Review decision memory or open decision detail

We never say "the model learns" - we say "policy evolves based on human decisions." Every human override, every AMBER resolution, every exception handled strengthens governance. Learning is explicit, approved, and reversible.

5826289

DECISIONS ANALYZED →

161

PATTERNS IDENTIFIED →

Impact of Learning (Last 90 Days)

Organizational context — how learning reduces friction across all decisions

15.0% -> 0.0%

Override Rate Reduced

+12%

Auto-Posted Journals

582628

Escalations Avoided

14 days

Pattern -> Policy

Human Override Analysis

Overrides are valuable signals, not failures

Override Rate	0.0%
Total Overrides (30 days)	0

Learning Sources

Where learning comes from

0% Human overrides

35% Agent drift detection

20% Policy review input

Learning Boundaries

Click to expand

Explicit limits on what bp Sphere can learn

X Outside approved policy scope

X From one-off overrides (pattern required)

X Without human review and approval

X Cross-domain without explicit authorization

X Beyond configured autonomy ceiling

Learning is constrained by design. bp Sphere cannot learn its way out of governance.

All learning artifacts are versioned, reversible, and auditable.

Policy changes can be rolled back. Every recommendation includes the evidence that generated it.