B.1.3:4.4 - Core rules for epistemic aggregation (design‑time synthesis)

When computing Γ_epist^synth(D_know):

1. Provenance preservation. The provenance/evidence graph is unioned with de‑duplication; every claim in the aggregate remains traceable to its sources and methods. No source, method, or dataset that supports a retained claim may be dropped.

2. SCR construction. Build a U.SCR that lists all symbol carriers (texts, code, figures, datasets) that materially participate in the aggregate. Provenance nodes must be mappable to SCR entries.

3. Object alignment. Identify the result’s one exact EntityOfConcern. Reuse the same already identified entity when the inputs concern it. A governed least common ancestor in a domain taxonomy may support that identification, but the calculation does not create the entity. If the claim requires a collection, relation occurrence, or other joint subject, identify that entity under its direct pattern and show that its identity rule obtains. A list, dependency graph, shared label, or mapping cannot create a joint subject; if none is governed, stop with the missing composition governor instead of inventing a generic composite entity. Record the semantic mappings and their CL evidence summaries without silently merging homonyms.

4. Recover the support relation before combining. For the exact claim and scope, distinguish:

• Indispensable premises: the conclusion requires each named premise. A missing or defeated premise blocks that inference, not every narrower conclusion.
• Alternative sufficient arguments: each actually sufficient argument can support the conclusion; expose shared premises, datasets, assumptions, and failure causes. Different argument names do not prove independence.
• Complementary evidence: a source may constrain a rival explanation, magnitude, uncertainty, or applicability without being a necessary premise. A limited additional study need not lower the existing support.
• Different scope slices: retain their populations, outcomes, conditions, and time extents separately unless an explicit transport or combination rule supports the joint claim.
• Counterevidence: retain credible results that conflict with the proposed conclusion. Weak support, absence of decisive support, an uninformative study, and evidence against a claim have different consequences.

Deduplicate actual evidence, not just citations. Account for overlapping data and shared biases before treating agreement as additional corroboration. State what each live limitation changes in the resulting claim.

5. Compose under the receiving model, or synthesize without a score. Keep F ordinal under C.2.3; any minimum for essential formal constituents concerns that formality claim, not R. Form G through the applicable C.2.2/A.2.6 scope rules; adding a study does not by itself extend applicability. For R, name the target quantity, compatible scales, dependencies, and warranted operation under B.3/C.2.2. This can justify a bottleneck minimum, a sufficient-argument choice, a probabilistic calculation, or a statistical synthesis; none is the universal default.

For a relied-on mapping, retain its CL summary and the actual limitation. A numerical loss function needs a receiving model that establishes its meaning, units, calibration or derivation, and assumptions. An ordinal CL rank, a monotone penalty table, or clipping to [0,1] supplies none of these. If no such calculation is justified, retain the separate support and mapping limitations in a reasoned synthesis; no penalty table or new study is required merely to return that result. A receiving assurance threshold applies only to the quantity and use for which it was justified.

6. Conflict detection and disposition. Detect contradictions, including overlapping-scope p and ¬p. Resolve a scope or interpretation difference only when the source facts establish it; do not explain away a credible contrary result by an invented subgroup story. Otherwise narrow, qualify, or withhold the affected conclusion and retain explicit conflict edges. A numerical synthesis may represent disagreement only under its justified model, not conceal it. Open B.2 only if exact construction facts leave a separate whole-reidentification question after the existing-whole explanation check.

7. Handling axiomatic and world-facing support. Retain each episteme’s declared mode and actual support:

• For an axiomatic input, empirical R may be N/A. Keep the proof, its conclusion under the stated axioms, and its formal validity; line=formal is a useful tag, not a conversion rule. Do not set R to F. An ordinal F-derived proxy describes only its declared ordinal meaning. Any value proposed for an R calculation needs a receiving model establishing meaning, scale, conversion, and assumptions; rescaling F into [0,1] is insufficient.
• For a postulative input, retain its actual warrant and empirical or other support as applicable. Apply a B.3.4 currentness or decay policy only to the support whose use consumes that policy; changing the mode creates neither evidence nor a conversion model.
• The aggregate declares its mode. If all its operative inputs are axiomatic, it is axiomatic; if an operative input is postulative, it is postulative. Keep any formal subclaim separately usable. A proof about a model supports a claim about a real system only with the needed model-to-world assumptions; evidence violating those assumptions remains visible.
• Constructive derivations. State the chosen proof basis and the correspondence needed by the conclusion. An induction proof may use its stated non-UF foundation. A UF-based equivalence claim supplies the required equivalence or isomorphism witness and the structure that the receiving theorem needs preserved. A lossy mapping cannot transport a theorem that depends on an erased distinction. If the required proof basis or correspondence is unavailable, leave that inference unresolved while retaining independently supported conclusions. A CL summary describes mapping evidence; it establishes neither equivalence nor theorem transport.

8. Order-aware arguments (optional). If the conclusion depends on premise or derivation order, identify the ordered relations and positions, declare the OrderSpec, and retain the branch, join and independence conditions needed by the argument. Use B.1.4:4 to check the bounded aggregation and its determinacy under those conditions.

9. No costs here. Any compute/collection effort is Γ_work; attach references but do not mix costs into epistemic aggregation.