Technology
From state to permission.
Tajalli converts source evidence, governing constraints and operational authority into deterministic decisions before systems commit or act.
From probabilistic intelligence to deterministic operational control.
Source evidence
01Telemetry, records, declarations and system responses — as received.
Canonical state
02Evidence normalized into one explicit, versioned representation.
Admissibility / authority
03Constraints and authority evaluated against that state.
Deterministic decision
04One decision for one state and one policy version.
Validity boundary
05The conditions under which the decision stops holding.
Execution / commitment
06Only what was decided is permitted to proceed.
Replay
07State, policy and lineage reproduce the decision exactly.
01Architecture
Source-bound state
Every decision starts from evidence that can be named: what was received, from where, and when. Tajalli normalizes that evidence into a canonical state before anything is evaluated.
Inference is never silently promoted to fact. Where evidence is missing, the state records the gap and the decision reflects it.
- Explicit provenance per input
- Normalization before evaluation
- Gaps represented, not filled
02Architecture
Deterministic kernel
The kernel evaluates admissibility and authority as a deterministic function of state and policy. Equivalent admissible state under the same policy version yields the same decision on every run.
Probabilistic components can feed the kernel. They do not sit inside it.
- No sampling in the decision path
- Equivalent inputs, equivalent outputs
- Bounded, inspectable outcomes
03Architecture
Boundary activation
A decision is only valid while the evidence and governing state that produced it still hold. Each decision carries its validity boundary explicitly.
When a relevant boundary is crossed, the decision is withdrawn and the new state is recertified. Decisions do not quietly age into falsehood.
- Validity conditions attached to every decision
- Boundary-triggered withdrawal
- Deterministic recertification
04Architecture
Policy versioning
Rules, thresholds and authority models are versioned artifacts. Every decision references the exact policy version under which it was made.
Policy changes are deliberate releases, not runtime drift.
- Immutable policy versions
- Decision-to-policy binding
- Reviewable change history
05Architecture
Replay and audit
Each decision carries a lineage: the state it was derived from, the evidence behind that state and the policy that governed it. Decisions are identified by a cryptographic hash of that content.
An auditor does not need to trust a log entry. The decision can be reconstructed.
- Content-addressed decision identity
- Immutable lineage
- Replay from recorded inputs
06Architecture
Integration architecture
Tajalli is placed between proposing systems and executing systems. It does not replace forecasting, optimization, planning or the AI models a customer already operates.
Client-specific adapters translate local formats into canonical state. Adapters change per environment. The deterministic core does not.
- Adapter layer per environment
- Stable deterministic core
- No replacement of the customer stack
07Adoption path
Shadow → advisory → gated operation.
Control is earned in stages. Each stage produces evidence that justifies the next, and no stage requires the customer to remove a system they already trust.
- Stage 1
Shadow
Tajalli observes proposals and computes decisions in parallel. Nothing is enforced. Decisions are compared against what the existing stack did.
- Stage 2
Advisory
Decisions are surfaced to operators and systems as recommendations, with lineage. Humans remain the gate.
- Stage 3
Gated operation
The decision becomes the gate. Commitments and actions proceed only within what Tajalli has certified or authorized.
Disclosure
The architecture is public. The kernel is not.
This page describes capabilities and interfaces. State construction, certification methods and policy synthesis are proprietary and are discussed under NDA as part of a technical evaluation.
Enterprise enquiries
Put a deterministic boundary before consequence.
Discuss an enterprise evaluation, shadow deployment or technical integration with Tajalli.