Chapter 6.2 follows chapter 6.1, Glossary of campaign terms, and precedes chapter 6.3, The batch ledger, within part 6, the reference part.
6.2.1 Pooling and invalidation terms
Pooling, combining cells in one analysis, is limited by a boundary, a recorded turning point after which cells cannot be pooled with cells before it. A per-cell invalidation, a validity decision attached to an individual cell, removes that cell without creating a campaign boundary. A waiver, an explicit permission to depart from a standing rule, does not by itself move a boundary. A run set, a named group of cells collected under one operating configuration, is governed by the boundary that applies when it begins.
6.2.2 Inclusion conditions of the campaign record
The inclusion rule used by the analysis plan and this page has four conditions: valid_for_primary_analysis == true, status == complete, the run set is inside the applicable boundary, and no operator invalidation marks the cell. The two per-cell invalidations described below therefore change cell eligibility without changing the pooling schedule.
source: operations/results-integrity/B0-scope-filter-finding.md section 7
6.2.3 Boundary schedule of the campaign record
Measured cells enter the first boundary. The first run set governed is 006. source: B0-scope-filter-finding.md section 7 table row 1 The records that establish this boundary are the chronology of design amendments, the ordered record of design changes, the container census in Appendix G, and the ruling recorded in Appendix A. Pooling cells before this boundary with later cells invalidates every analysis because the earlier cells are shakedown pilots excluded outright.
The single neutral prompt is placed on every build at the next boundary. The first run set governed is 011. source: B0-scope-filter-finding.md section 7 table row 2 The establishing record is the prompt amendment recorded in Appendix A and the corresponding statement in Appendix G. Pooling across this boundary invalidates documented-versus-stripped comparisons that combine different prompt regimes.
The task suite becomes additive and the removal suite is retired at the next boundary. The first run set governed is 035. source: B0-scope-filter-finding.md section 7 table row 3 The establishing records are the chronology of design amendments, the retired-suite archive note, the run index, and Appendix G. Pooling pass-rate or cost figures across the two suites is invalid. The model changes at the same boundary, so its effect cannot be separated from the suite change.
The campaign moves to local disk at the next boundary. The first run set governed is 055. source: B0-scope-filter-finding.md section 7 table row 4 The establishing records are the instrument defect list and the tag registry. Pooling wall-clock or staging-time comparisons across the filesystem change is invalid. The record reports no change to task briefs, checks, or cell definition, so correctness outcomes are not invalidated by this boundary.
Artifact hints and retry notices become recorded factors, and the model becomes an analysis stratum, at the next boundary. The first run set governed is 057. source: B0-scope-filter-finding.md section 7 table row 5 The establishing records are the amendment chronology, Appendix E, and Appendix G. Pooling documented-arm cells across hint or feedback settings without the setting as a factor is invalid. Pooling across models without the model as a factor is also invalid.
Documentation upkeep becomes a required step and the handoff note is carried at the final boundary. The first run set governed is 073. source: B0-scope-filter-finding.md section 7 table row 6 The establishing records are Appendix E, Appendix G, Appendix I, and the tag registry. Pooling documented-arm cells under the earlier and later prompt versions without prompt version as a factor is invalid. The upkeep-token figure is meaningful only from this boundary onward.
6.2.4 Boundary diagram of the campaign record
The campaign has six turning points, each with a first governed run set and records that establish the boundary. The two per-cell invalidations remain separate because they do not create boundaries. source: operations/results-integrity/B0-scope-filter-finding.md section 7
Campaign turning points, governed run sets, records, and non-boundary invalidations.
6.2.5 Per-cell invalidations of the campaign record
The shell-availability ruling invalidates cells measured while the container denied the agent a shell. It is carried in the cell matrix through shell_available and valid_for_primary_analysis, including cells in separated run sets. It is a per-cell invalidation, not a boundary. The shell ruling estimates roughly 134 affected cells.
source: Appendix I line 525
The retirement of the single-model transcript rule is also a per-cell invalidation. It re-admits eligible cells where the rule no longer applies, so it does not create a boundary. These events include the source records in Appendix I and the decision sheet. The isolation contract is likewise recorded as a scheduling and bookkeeping change that does not alter task briefs, hidden checks, or the cell-level measurement definition.
The invalidation migrated from page ten, section fifteen, as a non-boundary. It remains separate from the boundary schedule and from any waiver.
6.2.6 Unresolved boundary candidate of the campaign record
The chain-ordering rule remains an unresolved candidate between run sets 047 and 055. If it changed how chained tasks were staged, it would govern sequence analyses across that interval. The available grep evidence does not settle the issue, so the candidate is not entered as a boundary. source: B0-scope-filter-finding.md section 7 candidate note
6.2.7 Registry placement of the campaign record
The proposed normative home for the schedule is a top-level boundaries array in 5. Experiment/0. Plan/claim_registry.json. The array is absent from the current registry, whose keys are schema_version, registry_id, questions, and claims. Each boundary entry should carry an identifier, its first governed run set, its date, its establishing amendment or decision, the pooling consequence, and source references in the registry’s existing shape. Claims should name their governing boundary.
source: 5. Experiment/0. Plan/claim_registry.json
The operator-facing tags key in 5. Run Sets/run_tags.json should remain sparse. An optional boundary identifier may be attached to boundary tags, while an untagged run set remains ordinary. The tag registry should not become the normative schedule because it is rewritten by the browser and launcher.
source: 5. Run Sets/run_tags.json
6.2.8 Appendix G citation control
Appendix G citations in the source memo are stale where they rely on line numbers. The census and count-change passages must be located again by their quoted phrases before this page carries those citations. This preserves the source meaning without relying on line positions that now return other text.