Fix equivalence harness miscalibration from #777 (DL-16)

[ktehranchi]

Changes proposed in this Pull Request

Repairs two defects in the Tier-C equivalence harness that landed on develop with #777 and would make the next run return an untrustworthy verdict. Surfaced by the adversarial review of #778, but not caused by it — both are present on develop today.

Three files: one config line, one waiver, one ledger amendment. No pipeline code changes.

B1 — the usa leg was comparing two different demand pipelines

config.equivalence.yaml:77 pins bus_allocation: breakthrough so the western leg stays apples-to-apples with the anchor, which has no census-population method (git show e7f8bd70:workflow/scripts/build_base_network.py | grep load_weight → empty). config.equivalence-usa.yaml carried no such pin, and build.py copies the candidate's config into the anchor worktree — so both sides ran unpinned at #777's new population default.

Effect on any EQ_INTERCONNECT=usa run: the candidate weights demand by 2020 census county population while the anchor weights by Breakthrough Pd. That diverges Load_t.p_set, Bus.LAF_state, the kmeans geometry (cluster_network.py:68,82 weights on load_weight) and the objective — by the full reallocation. Every finding would read as a migration regression and not one of them would be. DL-13's recorded usa leg stopped reproducing the moment #777 merged.

Fix: add the same pin and comment already present in config.equivalence.yaml:75-77.

B2 — prong 1 reported pass: false for a non-migration reason

load_weight is candidate-only and survives every compared stage: cols2drop in aggregate_to_substations.py:165-188 omits it, cluster_network.py:269 sums it into the clustered frame, and clean_bus_data in add_electricity.py:1058-1067 drops only load_dissag/LAF/LAF_state. So compare.py:174-184 raises a column_set finding at five stages and compare.py:546-549 flips the whole run to failing.

Fix: one waiver, following the Generator/ba_eia precedent at waivers.yaml:59-64.

Why the waiver is sound

Only because of the B1 pin. With breakthrough pinned, load_weight carries the anchor's own legacy Pd weighting, so the column's presence is numerically inert. The two fixes are a package — the waiver alone would suppress a real signal.

Scoping was verified against compare.py's actual is_waived (lines 80-87), not by inspection:

the 5 findings B2 predicts:
  assembled_substation_network     waived=True
  clustered_network                waived=True
  extra_components                 waived=True
  prepared_network                 waived=True
  sectored_network                 waived=True

control - must NOT be over-waived:
  Bus.load_weight  / value        waived=False
  Load.p_set       / value        waived=False
  Bus.v_nom        / column_set   waived=False

load_weight value differences remain live, and demand content stays guarded by the unwaived Load / Load_t.p_set comparisons.

Ledger

Recorded as DL-16 in docs/superpowers/specs/2026-08-07-deltas-ledger.md, marked PENDING COUNTERSIGNATURE per the existing convention.

Checklist

• I tested my contribution locally and it seems to work fine.
• Code and workflow changes are sufficiently documented.
• Changed dependencies are added to envs/environment.yaml.
• Changes in configuration options are added in all of config.default.yaml.
• Changes in configuration options are also documented in doc/configtables/*.csv.

Verification performed: waivers.yaml parses (29 entries); the new waiver was exercised against compare.py's verbatim is_waived for both the intended matches and the controls above. No harness run — the Tier-C re-baseline that #778 requires is still outstanding, and this PR does not discharge it. B1 in particular is a config change whose effect can only be confirmed by an actual EQ_INTERCONNECT=usa run.