§5 — P4: Catalog unit-family repairs ✓ landed 2026-07-11

Executed as supervised sn edit operations against the production graph (lead-authorized window), each repair re-validated and re-reviewed through the full P2 gate stack; accepted-name renames stamped superseded_from_stage=accepted so P1 emits deprecation stubs at the next export. The review §7 table was 4 days stale — every row was reconciled against the live graph first (13 momentum rows had already been superseded by June pipeline work; 3 of the 4 energy-flux wall entries had been re-united to W·m⁻² by the DD interim fix).

Per-family outcome

Family	Decision	Result
Momentum → momentum_source	d-momentum-rename	18 accepted names, 17 edits (5 subtree roots + 12 projections). 14 successors accepted through the gates (scores 0.875–1.0); the deferred diamagnetic cascade child recovered by explicit self-scope rename. Residual check: *0 accepted bare `_momentum` at kg·m⁻¹·s⁻²**.
Energy-flux dimension flips	d-energy-flux-reserve	2 still-applicable recombination entries (m⁻²·s⁻¹) folded into the particle-flux family as `incident_neutral{,_state}_particle_flux_at_wall_due_to_recombination` — both accepted. The energy_flux family is now uniformly W·m⁻² across all accepted rows.
Power-density outliers → energy_flux	d-power-density (lead-refined)	The locked `_power_flux_` target was grammar-inexpressible (a flux of power is dimensionally nonsensical — W·m⁻² is the flux of energy). Lead decision: fold the 6 W·m⁻² entries into `energy_flux`; peaks use the canonical transformation `maximum_of_energy_flux_at_<locus>` (the infix `energy_peak_flux` spelling was explicitly disallowed). 5 renames accepted/collapsed (`energy_flux_at_divertor_target`, `maximum_of_energy_flux_at_{first_wall,divertor_target}`; `energy_flux_at_first_wall` collapsed into the existing `energy_flux_at_wall` by the reviewer). power_density (W·m⁻³) family untouched (71 rows).
One-line self-contradictions	—	`toroidal_angle_of_active_limiter_point`: recomposed from its true DD source (a dimensionless flux-limiter multiplier) rather than blessing the hallucinated `rad` anchor — the DD has no φ for active_limiter_point. Hz→s⁻¹ (fusion neutron rate) and electron_source_rate deferred (see residuals).
Source/flux validator rule	—	Live at ISN `4d43923`: flags families where `_X` and `_X_flux`/`*_X_source` share identical units under near-identical names (warning severity) + 6 tests.

Deprecation-stub spot-check

Every renamed accepted old name shows name_stage=superseded, superseded_from_stage=accepted in the graph (the 802 pre-existing superseded nodes carry sfs=None, so the new stamps are a clean signal) — confirming P1 will emit their deprecation stubs at the next export.

Residuals (documented, safe)

3 neutral_state momentum successors exhausted (parallel/poloidal/toroidal_neutral_state_momentum_source): near-threshold quorum (0.756–0.825 vs 0.85), not a naming defect — the pipeline's own refine agreed no better name exists, and the state-resolved DD sources are genuinely distinct from the accepted state-summed siblings. Their old names are superseded; P1's skip-logic (successor not published ⇒ stub skipped) prevents any dangling pointer, so this is safe. Not force-accepted — that would bypass the P2 gate. Left for a future escalated-review or lead physics-judgement pass.
maximum_of_energy_flux_at_limiter drafted: the refine loop reworded along_limiter → at_limiter (canonical locus preference); the successor drains through a normal review rotation.
parallel_net_ion_momentum_source: refine reworded from parallel_ion_momentum_source ("net" is redundant — true of every family member); the clean re-rename is blocked by the superseded-intermediate tombstone id (queued: tombstone-id supersede mechanism — now blocking two honest repairs).
Pair supersede (power_density_at_wall → existing energy_flux_at_wall): no sn edit rename-onto-existing-id mechanism exists; staged as a targeted write (p4_pair_supersede.py, both confirmed accepted W·m⁻²) run under the supervised window.
Hz→s⁻¹ + electron_source_rate: DD-unit-driven (unit_overrides territory) and a mixed-source-set unit issue respectively — cross-scope, queued.

Mixed-unit sweep

Regenerated from the live graph: 2312 accepted names, 310 grammar-groups, 36 mixed-unit — the residual is principled splits (eV vs K temperatures, global s⁻¹ vs volumetric m⁻³·s⁻¹ rates) plus a short list of genuine oddities recorded for later review (neutron-detector unit vectors carrying m; the bare-flux group).