Reachability & federation.

CRC does not treat every graph edge as equal and it does not reduce graph risk to a connection count. It defines what authority and consequence may become reachable when a governed operation traverses beyond an isolation surface.

The rule. CRC defines the admissible exposure envelope. Query, traversal, reasoning, and federation engines optimize inside that envelope; they do not enlarge it for performance.

Three different questions.

Surface sizing asks how much authority and value should sit inside one local isolation boundary. Reachability asks what becomes accessible when traversal crosses relationships. Federation asks what must move, coordinate, or become mutually visible when execution spans multiple governed surfaces.

CRC is an envelope, not a single risk score.

1. Local surface sizing.

The existing isolation-surface model remains intact:

Recommended local surface size
S* = √[(c_r + c_c f) / (p v̄)]

S* answers a partitioning question. It should not be overloaded with graph or distributed-query variables that measure different phenomena.

2. Consequence-weighted traversal exposure.

Raw edge count is useful as a resource guard, but it is not the risk model. A descriptive, non-traversable edge is not equivalent to an authority-bearing edge into a production or regulated domain. CRC therefore models realized exposure over the relationships actually exercised by a plan.

Traversal exposure — conceptual form
X(P) = Σ w_e, for e in R(P)
w_e = q_e · V_e · A_e · B_e
SymbolMeaning
PThe candidate execution plan.
R(P)External or boundary-significant relationships exercised by the plan.
q_eEffective traversability of edge e under the relevant principal and operation.
V_eValue or consequence exposed beyond the relationship.
A_eAuthority amplification associated with crossing the relationship.
B_eBoundary significance or domain sensitivity factor.

The exact weighting model is a standardization and qualification task; the important invariant is already clear: effective reachable authority and consequence matter more than connection count alone.

3. The traversal envelope.

A CRC profile can place hard bounds on dimensions that remain directly measurable during execution:

Traversal constraints
X(P) ≤ X_max
D(P) ≤ D_max
E(P) ≤ E_max
N(P) ≤ N_max
edge-type(P) ⊆ T_E
X_maxMaximum consequence-weighted reachable exposure.
D_maxMaximum traversal depth.
E_maxMaximum examined or traversed edges.
N_maxMaximum reachable/result nodes.
T_EPermitted external edge classes or relationship types.

4. Federation is a separate exposure class.

A query can remain within its traversal budget and still create unnecessary cross-boundary movement. CRC therefore treats federation as a vector of constraints rather than forcing unlike costs into one arbitrary scalar.

Federation envelope
F(P) = (S_f, C_f, B_f, V_f, A_f, U_f)
S_fParticipating governed surfaces.
C_fCross-surface exchanges or boundary crossings.
B_fBytes, records, or result material crossing boundaries.
V_fConsequence-weighted information exported from a surface.
A_fAuthority differential crossed during federation.
U_fUncertainty introduced by snapshot age, incomplete remote knowledge, or coordination state.

Optimization comes after admissibility.

A CRC-aware planner first removes every plan that violates identity, authority, edge-class, reachability, or federation constraints. Only the remaining plans are candidates for cost optimization.

Constrained plan selection
P* = arg min C_execution(P)
subject to P ⊨ CRC ∧ Authority ∧ Provenance

This ordering matters. CRC should not use a weighted objective in which enough CPU or latency savings can compensate mathematically for impermissible exposure. A forbidden traversal is not a high-cost plan; it is not a plan.

Federation creates an optimization opportunity.

When multiple surfaces are needed, the preferred plan can minimize exposure as well as conventional compute cost. Predicate pushdown, surface-local filtering, remote aggregation, narrow projection, and returning attributed summaries instead of raw graph fragments can reduce boundary crossings and data movement while preserving the query's semantics.

Move the question to the data when possible. Do not move the graph to the question by default.

Permitted exposure vs. exercised exposure.

A principal may be authorized for a broad envelope without every query consuming it. CRC should therefore distinguish the maximum permitted envelope from the exposure actually exercised by the selected execution plan. That enables attestation of both.

Execution attestation
permitted: (X_max, D_max, E_max, N_max, F_max)
actual: (X(P), D(P), E(P), N(P), F(P))

Profiles, not arbitrary knobs.

A serious standard cannot simply tell an administrator to invent limits. CRC profiles should ultimately derive recommended envelopes from measurable domain characteristics such as authority concentration, data consequence, external connectivity, autonomy level, reversibility, regulatory exposure, and recovery cost. High-exposure domains can then use conservative traversal and federation envelopes without pretending that one universal edge count is safe everywhere.


Independent federation qualification.

CRC should not validate its federation model only against a CRC-aware implementation. The v1.2 paper therefore defines an external reference program using established RDF/SPARQL federation engines as controls.

ReferenceQualification role
Eclipse RDF4J FedXPrimary transparent-federation reference. FedX federates multiple SPARQL endpoints, performs source selection, and exposes an optimizer/evaluation strategy suitable for measuring selected sources, joins, remote work, and source-selection behavior.
Apache Jena ARQ / FusekiExplainable baseline. Explicit SPARQL federation and SERVICE-based workloads provide a simple control in which remote boundary crossings are directly visible in the query.
GraphDBProduction-oriented comparison for remote repositories and SPARQL federation. It is a comparison engine, not the source of CRC semantics.

The same logical fixture should be executed with different plan shapes while measuring surfaces available, permitted, planned, and actually contacted; remote requests; cross-surface joins; bytes and bindings moved; graph edges examined; boundary edges crossed; maximum depth; latency; and the proposed CRC exposure quantities.

Equal answers do not imply equal exposure. That proposition is testable.

Implementation example: AletheiaGraph + Marshaller.

AletheiaGraph is a General Reasoning implementation candidate for CRC-aware execution, not evidence that CRC is correct. Its graph-native planner, bounded traversal, reasoning, proof/provenance, and authority model provide a place to enforce CRC admissibility before physical optimization. Marshaller is the intended cross-surface execution layer, where predicate pushdown, surface-local filtering, narrow projection, and attributed aggregation can reduce boundary movement.

Reference execution order
query + principal
→ CRC / authority admissibility
→ permitted candidate plans
→ exposure minimization
→ execution-cost optimization
→ execution + attestation

The purpose of the independent FedX/ARQ/GraphDB qualification program is precisely to prevent CRC from becoming a standard that merely describes General Reasoning's own engine.

Standardization status. The surface-sizing model is retained as its own component. The reachability and federation quantities on this page define the proposed CRC expansion and the qualification variables that follow from it. The exact consequence-weighting factors and domain-profile calibration should be frozen only after empirical qualification.
Isolation Surface Size → AletheiaGraph →

General Reasoning, Inc. · Birmingham, Alabama · MIT License · 2026
Enterprise inquiries: inquiries@genreason.com