The Axiom Pipeline

From input to auditable evidence

Axiom processes financial decisions through a five-step pipeline. Every step is designed to ensure the outcome is deterministic, documented, and independently verifiable.

  1. 1

    Ingest

    Transaction data and supporting documents flow into Axiom from your ERP or source systems.

  2. 2

    Bind Policy

    The active policy version is locked to the decision, ensuring the exact rules are recorded.

  3. 3

    Decide

    Deterministic logic evaluates the inputs against the bound policy to produce an outcome.

  4. 4

    Emit Bundle

    A complete Decision Evidence Bundle is generated with all inputs, reasoning, and integrity hash.

  5. 5

    Replay

    Any decision can be independently verified by replaying the inputs against the same policy version.

1

Ingest

Axiom receives transaction data and supporting evidence from your existing systems through adapter-based integration. Your ERP remains the system of record — Axiom operates alongside it.

  • Transaction metadata (amounts, dates, categories)
  • Supporting documents (receipts, invoices, contracts)
  • Contextual data (submitter role, department, approvers)
Data Flow
SourceERP / Source System
MethodAdapter-based API
FormatStructured + Documents
2

Bind Policy

Before evaluation begins, the currently active policy version is bound to the decision. This creates an immutable link between the decision and the exact rules that will be applied.

  • Policy identified by version number
  • Effective date recorded for audit trail
  • Historical decisions always use their bound version
Policy Binding
Versionv2.3.1
Effective2024-01-15
StatusLocked
3

Decide

The bound policy rules are applied to the ingested data deterministically. This is not probabilistic AI — the same inputs with the same policy version always produce the same outcome.

  • Deterministic evaluation (no randomness)
  • Rule-by-rule evaluation logged
  • Human-readable explanation generated
Decision Logic
IF amount ≤ $5,000
AND role = "Manager"
AND category ∈ [Travel]
AND receipt.valid = true
THEN APPROVE
4

Emit Bundle

A complete Decision Evidence Bundle is generated containing everything needed to understand and verify the decision: policy version, inputs, outcome, explanation, and integrity hash.

  • Complete audit trail in one artifact
  • Cryptographic integrity hash prevents tampering
  • PII redaction available (see Trust)
Evidence Bundle
Policy version + effective date
Input facts + source evidence
Decision outcome
Human-readable explanation
Cryptographic integrity hash
5

Replay

Any historical decision can be independently verified through Deterministic Replay. Feed in the same inputs with the same policy version, and you'll get the exact same outcome — guaranteed.

  • 100% reproducible outcomes
  • Auditor can verify independently
  • Debug and understand past decisions
Deterministic Replay
Original
APPROVED
Replay 1
APPROVED
Replay 2
APPROVED

Same inputs + same policy = same outcome. Always.

Operating principles

Disciplined. Evidence-led. Finance-controlled.

Axiom records the basis for every outcome and keeps uncertainty visible. It provides decision evidence; your team retains authority over policy and action.

No outcome without a basis

Every outcome is traceable to a policy version, the facts evaluated, and the source evidence available at decision time.

Explicit about uncertainty

When evidence is incomplete, Axiom records what is known, what is missing, and why the decision was routed for review.

Your team decides

Axiom applies the policies you approve and preserves the evidence. It does not author policy or replace human authority.

Why Deterministic Replay matters

In traditional AI-based automation, you can't reliably reproduce a past decision. Models may be updated, weights may shift, and even the same inputs can produce different outputs. This makes audit defense difficult — you're asking auditors to trust a black box.

Deterministic Replay changes this. Because Axiom uses versioned policy logic rather than probabilistic models, every decision is reproducible. Take any historical decision, feed in the recorded inputs with the bound policy version, and you'll get the exact same outcome.

This enables:

  • Independent verification — Auditors don't have to take your word for it. They can replay decisions themselves.
  • Debugging clarity — When something goes wrong, you can trace exactly which rule fired and why.
  • Compliance confidence — Prove that your automation behaves consistently and predictably.

Policy versioning with effective dates

Business rules change. Approval limits increase, compliance requirements evolve, and organizational structures shift. Axiom handles this through policy versioning with effective dates.

When you update a policy, the new version takes effect from a specified date forward. Historical decisions remain bound to the policy version that was active when they were processed. This means you can always understand a past decision in its original context.

Example: Approval limit change

Jan 1
Policy v2.2 — Manager approval limit: $5,000
Jan 15
Expense #1234 ($4,500) processed → APPROVED(bound to v2.2)
Feb 1
Policy v2.3 — Manager approval limit: $3,000
Feb 15
Expense #1235 ($4,500) processed → ESCALATE(bound to v2.3)

Replaying Expense #1234 uses v2.2 (its bound version), not current v2.3.

See the workflow in action

Book a demo to walk through how Axiom Governance would work with your specific financial workflows.