Separate runs preserve source boundaries and make comparison reviewable.
Responsible AI
Answers are not the finding.
Zen Sentry helps bring Veteran experience, warrior-ethos responsibility, cultural humility, and human follow-through into the KEMPO evaluation work of TWiN.
August 2026 snapshot
Generation done.
Judgment still human.
Six current model runs completed all 100 scenarios and deterministic protocol checks. The 600 responses remain unscored and unranked until accountable people review them one by one.
Completion confirms generation—not safety, quality, alignment, or readiness.
Evidence becomes a result only after rubric-based review and explicit failure handling.
KEMPO // practiced judgment
An AI martial art for the mind.
KEMPO is a repeatable, human-scored 100-scenario evaluation of AI judgment across behavioral safety, ethics, cyber, crisis, and public safety.
Knowledge
Accuracy, context, source awareness, and understanding of the situation.
Empower
Dignity, autonomy, useful options, and support for human agency.
Monitor
Proportional awareness without surveillance theater or overreach.
Prevent
Recognition of foreseeable harm and responsible protective action.
Optimize
Practical next steps, follow-through, and responsible use of resources.
Evidence path
Reviewable steps. No skipped gate.
Public aggregate findings should follow the evidence path. Detailed answers and evaluator artifacts remain under reviewed access rather than becoming context-free content.
- 01Isolated model responses
- 02Deterministic protocol audit
- 03Human rubric scoring
- 04Explicit failures and averaged results
- 05Public aggregates with controlled detail
Veteran-centered contribution
Strength with restraint.
Culture with humility.
Warrior ethos can mean courage in service, composure, protection, restraint, and responsibility for power. It should not become a stereotype or a claim of authority over every Veteran’s experience.
Human pathway
Guided reflection → pattern awareness → optional peer or navigator involvement → verified resource navigation → warm handoff → access assistance → human follow-through.
Public privacy boundary
No private prompts, model responses, evaluator identities, Veteran records, health information, or controlled research artifacts belong in this public repository or website.
Trust is operational
Make every
boundary visible.
Read the current public ledger for organizational status, funding, program maturity, source, privacy, and clinical limits.
