The drift-report-actionable work (branch worktree-drift-report-actionable, see tools/README.md's "Known limitation" note added in commit 8531c4c) aggregates systemicGaps/conventionStrength in tools/Build-PfbApiDriftReport.ps1 over high-confidence endpoints only -- deliberate, since a partial-confidence endpoint's gap list can contain false positives (see the false-positive procedure in tools/README.md).
Consequence: 54 distinct wire field names appear only on partial-confidence rows and are entirely absent from both systemicGaps and conventionStrength (e.g. base_dn, bind_password, domain, nameservers, qos_policy, storage_class, owner, ...). Names that do appear can also undercount -- context_names is 253 in systemicGaps but 289 across all confidence levels.
No per-endpoint row is affected; this is purely a gap in the aggregated triage view.
Proposed fix: aggregate across all confidence levels, keeping the current high-confidence-only count as endpointCount and adding a separate partialEndpointCount, rather than dropping names or silently re-baselining endpointCount.
Not fixed in the original branch since it would move the context_names/allow_errors/etc. figures that branch measured and pinned throughout.
The drift-report-actionable work (branch
worktree-drift-report-actionable, see tools/README.md's "Known limitation" note added in commit 8531c4c) aggregatessystemicGaps/conventionStrengthintools/Build-PfbApiDriftReport.ps1over high-confidence endpoints only -- deliberate, since a partial-confidence endpoint's gap list can contain false positives (see the false-positive procedure in tools/README.md).Consequence: 54 distinct wire field names appear only on partial-confidence rows and are entirely absent from both
systemicGapsandconventionStrength(e.g.base_dn,bind_password,domain,nameservers,qos_policy,storage_class,owner, ...). Names that do appear can also undercount --context_namesis 253 insystemicGapsbut 289 across all confidence levels.No per-endpoint row is affected; this is purely a gap in the aggregated triage view.
Proposed fix: aggregate across all confidence levels, keeping the current high-confidence-only count as
endpointCountand adding a separatepartialEndpointCount, rather than dropping names or silently re-baseliningendpointCount.Not fixed in the original branch since it would move the
context_names/allow_errors/etc. figures that branch measured and pinned throughout.