Record the glazing reconcile collapsing to a MIXED sentinel (ADR-0042 amendment)

Per-window assignment from a landlord aggregate is strictly worse than the cert's
existing per-window sap_windows, so a mixed override always defers — the proportion
comparison only fed a 'flag when far' branch not worth a new schema field. Realized
decision: a GlazingType.MIXED sentinel (LLM classifies mixes to it) resolves to no
overlay, preserving the cert glazing; uniform assertions still apply.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

docs/adr/0042-glazing-override-reconciles-against-cert-composition.md

@@ -29,3 +29,14 @@ The landlord glazing override **reconciles against the cert's per-window composi
 - Genuine landlord-vs-cert glazing disagreements are **surfaced** rather than silently trusted (old behaviour: flatten to double) or silently overwritten.
 - **Alternatives rejected.** *Dominant type wins* — still clobbers per-window variation. *Carry a full landlord composition to re-derive per-window types* — per-window assignment from an aggregate is unknowable; we decline to fabricate it. *Deterministic regex of "X% double, Y% single"* — brittle against varied input, which is the LLM's job.
 - Pairs with the per-window fidelity the calculator already relies on: the reconciliation is only correct on a faithful Effective EPC whose `sap_windows` round-trip (cf. ADR-0040).
+
+## Amendment (#1376 implementation): the reconcile collapses to a `MIXED` sentinel
+
+Implementing this surfaced that the full "carry a proportion → compare to the cert → flag when far" design buys less than its cost. The cert **already** carries the best-possible per-window mapping (`sap_windows`, each with its own `glazing_type`); a landlord aggregate ("40% double, 60% single") says *how much* is double but **not which windows**, so applying it would overwrite real per-window data with an arbitrary guess. Per-window assignment from an aggregate is therefore not just unknowable but strictly worse than the cert — so a mixed override **always defers**, regardless of the exact proportion. The proportion comparison only ever fed the "flag when far" branch, and carrying a proportion needs a new schema field (an FE/Drizzle dependency).
+
+**Realized decision.** Add a `GlazingType.MIXED` sentinel. The LLM classifies a genuine aggregate-mix to `MIXED` (the LLM handles the varied phrasings — no regex); `MIXED` is absent from the overlay's `_GLAZING_CODES`, so it resolves to **no overlay** and `_fold_glazing` never runs — the cert's per-window glazing is preserved. A **uniform** assertion (one type ≳ 90%) still resolves to that type and is applied to every window (unambiguous — no guess). Existing mixed-as-double rows are reclassified to `MIXED` (a deterministic parse of the `%X / %Y` format is acceptable for fixing *known* rows). The "uniform vs mixed" threshold (~90% one type) is tunable, pinned against the real distribution.
+
+**Consequences of the amendment.**
+- The defer behaviour needs **no apply-seam logic and no proportion field** — it falls out of "no overlay". The slice is: add the `MIXED` member, give the classifier the option, reclassify the ~319 rows.
+- `MIXED` is a **new FE-owned pgEnum value** (one member) — added by the FE owner (Dan); `property_overrides` (TEXT) reclassify runs immediately, the cache pgEnum write defers until the migration (the Class-A/B pattern).
+- The **"flag when landlord disagrees with the cert" is dropped for now** — since a mix always defers, the flag was informational only. `MIXED` is the hook if a future slice wants to surface those disagreements.