AIEP LRM — Cumulative Evidence-Bound Reasoning
AIEP LRM is a private reference reasoning architecture built on the Architected Instruction & Evidence Protocol.
It is designed to support cumulative reasoning, where later inference steps can commit to and reuse prior evidence-bound reasoning state instead of repeatedly reasoning from scratch.
Current status: private prototype / controlled research layer.
What AIEP LRM addresses
Standard AI inference is stateless at the reasoning layer. Each query begins from scratch. Evidence retrieved in a prior cycle is not committed, not reusable, and not auditable across calls.
This is not a retrieval problem. It is a reasoning architecture problem.
AIEP LRM investigates a different model: one where each reasoning step commits its evidence inputs, its intermediate state, and its output to the AIEP evidence substrate. Later steps can verify, extend, or explicitly reject prior committed state — without repeating the retrieval and qualification work that produced it.
Reference implementation: Piea
AIEP LRM is used inside Piea as a controlled reference implementation for cumulative, evidence-bound reasoning.
AIEP LRM is used inside Piea as a controlled reference implementation for cumulative, evidence-bound reasoning. AIEP Token and AIEP Live provide public-facing usage and activity surfaces that illustrate how AIEP reasoning, validation, and evidence-bound state can be measured without disclosing the private LRM implementation.
Piea is the only current controlled preview of AIEP LRM in operation. Detailed LRM mechanics are not publicly disclosed through Piea or any other channel.
Public observation surfaces
AIEP Token and AIEP Live provide observable signals of AIEP LRM activity without exposing the private implementation:
| Surface | Purpose | Link |
|---|---|---|
| AIEP Token | Illustrates measurable reasoning, validation, and evidence-bound activity | aiep.dev/aiep-token |
| AIEP Live | Public dashboard showing AIEP ecosystem usage and protocol signals | aiep.dev/aiep-live |
These surfaces show that AIEP LRM is active and in use. They do not expose token accounting logic, state reuse algorithms, or internal Piea traces.
Relationship to the public AIEP protocol
AIEP LRM builds on the public AIEP protocol layers:
| Layer | Status |
|---|---|
| Evidence schema and canonical record types | Public — AIEP Schema |
| Content hashing and append-only ledger | Public — AIEP Genome SDK |
| Evidence validation and negative proof | Public — AIEP Validator |
| Mirror pattern for governed reasoning artefacts | Public — AIEP Mirror |
| Cumulative reasoning mechanics | Private |
| Memory reuse and state compression algorithms | Private |
| Scoring and arbitration loops | Private |
| Token-reduction implementation | Private |
| Benchmark harness | Private |
The public protocol exposes the evidence, validation, schema, and replay layers. The detailed LRM architecture, optimisation methods, and implementation remain private.
Disclosure boundary
AIEP LRM operates under a controlled disclosure boundary. The following are not publicly disclosed:
- Cumulative reasoning mechanics
- Memory reuse algorithms
- State compression methods
- Token-reduction implementation
- Private benchmark harness
- Internal orchestration details from Piea or Forecast
The machine-readable disclosure record is available at /.well-known/aiep/lrm.json.
Enquiries
Serious partnership or evaluation enquiries — including NDA-gated technical briefings — should be directed to [email protected].