AletheiaGraph · General Reasoning, Inc.

The epistemic computing engine for autonomous enterprise systems.

AletheiaGraph combines authoritative graph persistence, enterprise query execution, logical reasoning, evidence provenance, governed traversal and truth semantics in one durable engine.

Not merely an engine for finding answers. An engine for establishing why an answer can be believed.

Why another engine?

If the job were only to store enterprise records, we would use PostgreSQL. AletheiaGraph exists because autonomous systems push correctness and governance requirements below the application boundary and through persistence, query planning, graph traversal, reasoning, proof and authority.

The requirement stack
Durable data
Transactions, concurrency, recovery and enterprise CRUD.
↓
Graph relationships
Native vertices, edges and graph-local identity.
↓
Optimized query
Indexes, statistics, cost planning, joins, aggregation and pagination.
↓
Logical inference
Bounded graph-backed Prolog reasoning.
↓
Proof & evidence
Exact committed evidence versions, support topology and validity.
↓
Authority & traversal
Explicit promotion and constrained graph exploration.

One engine, six roles

Enterprise database

Persistent operational state

Transactions, concurrency, recovery, indexes, CRUD, joins, aggregation and pagination.

Graph engine

Relationships are first-class

Vertices, edges and graph-native relationships are part of the authoritative state, not an application-side projection.

Query engine

Semantics-preserving optimization

Native algebra, statistics, cost planning and multiple access paths with authoritative-result equivalence.

Reasoning engine

Bounded logical inference

Recursive Prolog reasoning over the graph with budgets, cycle protection, timeout control and query isolation.

Proof engine

Evidence that survives

Conclusions can retain durable relationships to the exact committed vertex or edge versions that support them.

Governance engine

Inference is not authority

Stored, inferred and proposed state remain distinct; authority transition and traversal are explicit rather than implicit.

These are not six attached products. They operate over the same authoritative graph.

The epistemic model

AletheiaGraph treats epistemic state as part of the engine. It preserves the difference between what was stored, what was inferred, what was proposed and what was explicitly promoted into authoritative state.

Stored fact
→
Evidence / proof
→
Inferred conclusion
Proposed result
→
Governed promotion
→
Authoritative state
AletheiaGraph can preserve a durable proof relationship to exact committed evidence, determine whether that evidence remains current, stale, retracted or missing, and separately govern whether an inferred or proposed conclusion becomes authoritative without rewriting its epistemic history.

Explore proof, reasoning and epistemic state →

Reachability is not authority

A conventional graph engine asks whether one node can reach another. An autonomous enterprise system must also ask whether a principal may cross that relationship, under the relevant authority, for the current operation. AletheiaGraph is being designed so graph exploration can be constrained by governed surfaces rather than treating all topology as equally traversable.

Explore governed traversal →

Why CRC matters to AletheiaGraph

AletheiaGraph is deliberately being built to do more than record retrieval. It can represent first-class relationships, execute graph-native queries, perform bounded logical inference, preserve proof topology, and ultimately coordinate queries across governed graph surfaces. Those capabilities create power that a passive record store does not have: the engine can discover and exercise paths through organizational knowledge and authority.

The more capable the traversal and reasoning engine becomes, the less acceptable ambient reachability becomes.

CRC supplies the constraint model. Isolation surfaces define local boundaries; traversal envelopes constrain depth, edge classes, reachable nodes and consequence-weighted exposure; federation envelopes constrain cross-surface participation, exchanges, data movement and remote-state uncertainty. AletheiaGraph's optimizer is therefore intended to choose the cheapest permissible plan, not merely the cheapest plan.

First: admissibility

Identity, authority, CRC edge classes, reachability and federation limits remove impermissible plans.

Then: exposure

Among permitted plans, prefer plans that reduce unnecessary boundary crossings, graph expansion and transferred material.

Then: execution cost

Indexes, statistics, ordering, joins and other physical choices optimize execution without changing the permitted graph.

This also creates a natural execution record: what CRC envelope was permitted, what plan was selected, what traversal/federation exposure was actually exercised, what evidence supported the result, and what authority allowed the operation.

Correctness before optimization

AletheiaGraph is engineered for semantic correctness, deterministic state transitions and evidentiary transparency. Performance matters, but optimization is constrained by meaning.

Indexes do not become truth

Indexed results are qualified against authoritative graph semantics.

Plans cannot change meaning

Cost-based execution may choose a different physical path, not a different answer.

Inference is attributable

Derived conclusions remain distinguishable from stored facts and can retain exact supporting evidence.

Optimization may change how an answer is obtained. It may not change what the answer means.

Maturity is evidence, not elapsed time

Field age matters because production creates unpredictable combinations no finite test program can guarantee it has anticipated. But age is a proxy for accumulated evidence, not the evidence itself. AletheiaGraph deliberately attacks the failure modes most important to an authoritative agent-first graph and preserves discovered failures as repeatable qualification and regression cases.

370+
enumerated low-level ordered-index tests
21
authoritative write-path qualification stages
16,000
transactional increments in the concurrent lost-update workload
38
qualified proof / provenance / authority gates across F1–F4
100k
record B+ tree scaling qualification

See the qualification testbed →

Why now?

Two years earlier

We likely would have built above PostgreSQL.

The cost of owning durable storage, indexes, planning, recovery and systematic qualification would have made accepting an impedance mismatch the rational choice.

2026

AI changed the economics of scrutiny.

AI-assisted engineering makes continuous implementation review, adversarial test generation, edge-case discovery, comparative design analysis and repeated qualification practical at a level that previously required a much larger specialist team.

What changes now: production exposure remains uniquely valuable, but AI-assisted engineering can increase the rate at which adversarial hypotheses are generated, qualification evidence is produced, and discovered failures are converted into permanent replayable tests. Machine maturity can therefore accumulate through both field exposure and engineered evidence.

One authoritative substrate from transaction to inference

AletheiaGraph
Truth / evidence policy
Confidence, contradiction and trust policy — later Wall F work.
↓
Authority & promotion
Explicit governed transition into authoritative state.
↓
Proof / provenance / validity
Exact committed evidence versions and current/stale/retracted/missing validity.
↓
Logical reasoning
Graph-backed bounded Prolog inference.
↓
Governed graph traversal
Reachability constrained by principal, operation and isolation surface.
↓
Native query planner / executor
Filtering, joins, ordering, aggregation, projection and pagination.
↓
Indexes / statistics
Ordered, composite and traversal-oriented access paths.
↓
Transactional authoritative graph
Durable state, identity, versioning and relationships.

Agent-first requirements created the need. Epistemic computing is the architecture that emerged.

AletheiaGraph can perform the role of an enterprise database, but its purpose extends beyond database management: to provide an authoritative substrate on which autonomous systems can know, reason, explain and act.

What is an epistemic computing engine? →