Loading...
Loading...
That is the gap. Every instrument in the PME toolkit measures something adjacent to command judgment — traits, reflexes, recall, retrospective prose — and none of them produce a record of reasoning committed before the consequence was known. DecisionProof is an assessment instrument built to close it, with an evidence chain your accreditors can verify without us.
“How does this officer decide when the information is incomplete and the consequences are unknown?” Set each current instrument against what the record actually has to contain.
Expensive, slow, subjectively scored. Produces no structured data, cannot be replayed, cannot be audited. The record is the instructor’s memory.
Every exposure, commit and revision is a sequenced event. The session can be replayed deterministically months later.
Measure stable traits, not situational decisions. An officer’s profile does not predict what they do at 0300 with half the picture.
Assessment is scoped to observable behaviour inside one session. Personality attribution is blocked at the language layer.
Measure reflex and procedural knowledge. A high score means the officer executed the drill, not that they framed the problem correctly.
The learner must commit political object, competing frames, assumptions and a mechanism of effect before any course of action is available.
Written after the fact. The learner already knows the outcome, so the essay documents a rationalisation, not a decision.
Five rejection rules hold the hindsight fence automatically. Reasoning is sealed before consequence is revealed — no honour-system required.
A cohort can pass every one of those instruments and still leave you with no defensible answer about judgment under ambiguity.
See the instrument →It is an assessment system that uses structured decision scenarios as its instrument. Six rules make the record admissible.
Reasoning is committed before the consequence is revealed and preserved immutably. Five temporal rejection rules enforce it automatically.
CLOSES: Post-hoc rationalisation in written AARs
Before any course of action: political object, operational problem, preferred and competing frames, critical assumptions, disconfirming indicators. Frame first, act second.
CLOSES: Action scored without problem definition
Every decision carries an explicit causal chain — mechanism, expected response, success and failure indicators, material risks, and the conditions under which the officer would change approach.
CLOSES: Plausible choices with no stated logic
The system proposes observations from the evidence record; faculty accepts, edits, rejects, or marks insufficient evidence on each. No AI-generated grade ever reaches a learner.
CLOSES: Black-box scoring you cannot defend
Historical outcome is revealed only after self-assessment and AAR release, and labelled context, not an answer key. Agreeing with the historical commander changes no rating.
CLOSES: Answer-key thinking in wargames
A structurally similar case in a different domain — era, geography and actors change, constructs hold. Tests whether the principle generalised or the domain was memorised.
CLOSES: Domain-specific heuristics mistaken for judgment
Every transition is timestamped into the evidence chain, so the sequence itself is part of the proof. The learner cannot reach the outcome before their reasoning is committed — that is what makes the hindsight fence enforceable rather than aspirational.
Acknowledge the unclassified training boundary; review the period source packet. No progression without acknowledgment.
Commit problem understanding: political object, principal actors, constraints, preferred vs. competing frame, critical assumptions, disconfirming indicators, confidence.
Commit the causal explanation: desired condition, target actor, proposed action, mechanism, expected response, indicators, material risks, change conditions.
Commit the structured decision: action, intended effect, rationale, evidence cited, alternatives with rejection reasons, assumptions relied upon, confidence, risk.
Ten-field self-assessment written before any outcome is visible: what was intended, what happened, which assumption mattered, what transfers.
Faculty dispositions every candidate observation. Released to the learner only once all claims are resolved.
Historical context shown after self-assessment and AAR release. Labelled context, not an answer key.
The same decision process in a new domain — the test of whether a generalisable principle was extracted.
Candidate observations are generated from the evidence record — never from a model’s opinion of the learner. Faculty accepts, edits, rejects, or marks insufficient evidence on every single claim before the AAR unlocks.
Sessions awaiting review, filtered by status, cohort or scenario — straight to the claims needing disposition.
War frame, theory of action, decision journal and self-AAR preserved exactly as written.
Accept, edit, reject or mark insufficient evidence. Forbidden language blocked inline.
The AAR stays sealed until every claim is dispositioned and faculty explicitly releases it.
Six dimensions — framing, theory coherence, evidence-assumption management, adaptation, historical reasoning, transfer. Outcome is not a dimension.
Sealed package with full event chain, dispositions and verification checksums, ready for accreditation review.
Every event — source exposure, decision commit, frame revision, faculty disposition — is appended to a cryptographically chained, tamper-evident ledger. Completed sessions are replayed for state consistency, verified for evidence-link validity, and sealed with a root hash. The export package stands on its own in front of an accreditation team.
Full technical annex →We will not claim predictive validity we have not earned. The honest position, stated plainly, because you will be asked it by your assessment officer.
Accreditation mapping (JPME learning-area outcomes, institutional assessment criteria) is prepared per institution during onboarding — citations confirmed against your accrediting body, not asserted here.
Immediate deployment, auto-updates, row-level security, no infrastructure. Encrypted at rest and in transit.
Air-gapped deployable container. NIPR/SIPR compatible, full data sovereignty, supports classified scenario content.
Cloud instructor dashboard with local play nodes; assessment data syncs when a connection is available.
Section 508 compliant · WCAG 2.1 AA · browser-based, no install · 31 unclassified synthetic scenarios across 5 eras, plus Scenario Studio authoring.
We provide the platform, all 31 scenarios, the instructor dashboard, faculty AAR tools, audit export, technical support, and a co-authored validation study. You provide learners, faculty review time, and honest feedback on the instrument.
Request a Pilot Programkumar@quantumlearningmachines.com · accepting 3 institutions for Fall 2026.