G.2:4.3 - Harvester loop (conceptual choreography; pattern-governed)

A conforming G.2 pack publication is built by iterating the following conceptual loop until the declared gates are satisfied:

1. Declare scope and plurality. Identify the exact CG-frame (the declared framing episteme), the initial Tradition set, each intended claim region and EntityOfConcern, the comparison basis, and the receiving use. Record the cited CG-frame and source editions and evidence anchors in the pack pins rather than hiding them in a generic context field. Before counting, fix the HarvestPolicy’s receiving question, counted population, grouping and same-family equivalence, including overlap handling for a combined population.

2. Discover and triage sources (ledger‑first). Populate CorpusLedger via:

   • adding seed sources,
   • expansion via citation chaining and keyword family exploration,
   • pruning using load‑bearing relevance tests tied to the declared CG‑Frame scope.

3. Distill claims per Tradition. For each Tradition, author a Claim Sheet that preserves internal commitments and cites evidence anchors. Do not fuse cross‑Tradition claims at this stage.

4. Inventory operators/objects for downstream authoring. Extract candidate measurement terms and operator stubs for later CHR/CAL authoring (without asserting legality or thresholds locally).

5. Build alignment/divergence surfaces. Where reuse across Tradition is desired, record the obtaining correspondence and its exact basis in BridgeMatrix: F.9 for sense correspondence, C.3.3 for kind correspondence, or the direct rule for a plane relation, as actually used. State preserved distinctions and losses for the receiving question. Consolidation requires explicit alignment proof. Add bundle or gate anchors only for an independently applicable E.18 flow crossing or A.21 gate, under CC‑GCORE‑CROSS‑1.

6. (Alias: G.2‑F) Produce Γ_epist synthesis records when fusion/substitution is asserted. If a G.2 pack publication asserts fusion or substitution across sources or across Tradition records (beyond mere “parallel divergent claims”), it MUST emit GammaEpistSynthId records per G.2:Ext.GammaEpistSynthesis (provenance union + explicit object alignment refs + assurance tuple refs), and it MUST keep penalties routed to R_eff only by delegation (CC‑GCORE‑PEN‑1).

7. Publish teachable micro‑groundings. Attach worked micro-examples to load-bearing claims, each tied to the exact source and edition, claim region, EntityOfConcern, comparison basis, intended use, and evidence carrier or A.10 evidence-provenance path.

8. Apply gates and record repairs. Apply that fixed HarvestPolicy basis and count its distinct units before comparing coverage with FamilyCoverageFloorK (and apply any optional diversity-by-distance gate under its own basis). Missing count semantics returns an unassessable result and the exact missing basis, not an instruction to search more. If a defined gate fails, the pack MUST:

   • record the failure and the repair iteration in FlowRecord and CorpusLedger,
   • pin the updated HarvestPolicyRef / criteria ids (if changed),
   • iterate the loop rather than silently weakening the gate.

9. Emit hand‑off manifests and export views. Produce explicit manifests to:

   • G.3 (CHR authoring),
   • G.4 (CAL authoring),
   • G.5 (registry/dispatch),

   so that downstream work can cite pack components by id rather than re‑authoring them. Each relied-on coverage result carries the same CoverageJudgementRef and its policy basis; a downstream method-selection use cannot treat a combined method/generator count as a method-only count. The pack MUST also export SoTA_Set@CG‑Frame and SoTAPaletteDescription as the default downstream consumption surfaces (ids pinned).