docs(register): reframe C-37/C-38 as near-term preconditions for probabilistic reconciliation

reports/technical_risk_register.md

@@ -433,6 +433,8 @@ See also the migration plan (reconciliation slices 2-4) and views-reporting issu
 
 **Update 2026-06-24 (expert-code-review, Kleppmann lens):** the per-draw index-pairing is only *valid* if the production cm and pgm forecasts are the **same joint posterior draws**. If they come from independent models (separate posteriors), pairing draw *s* of the grid with draw *s* of the country is arbitrary and the reconciled uncertainty is meaningless — and the parity fixture cannot detect this, because it manufactures aligned draws. **At S7 (#39) wiring, verify the sample-alignment assumption against the real pipeline as a hard precondition** (or escalate the C-37 upgrade). This is a correctness precondition distinct from the "is the method principled" question.
 
+**Update 2026-06-24 (reframe — now near-term, not deferred):** this is no longer a someday concern. FAO (`rusty_bucket`) sidesteps reconciliation entirely (pure-grid ensemble aggregated *up* — sums by construction), but the **next UN-agency deliverable** is an FAO-like grid ensemble that **does** reconcile against a CM model, with **full pooled draws (~1024)** → **probabilistic** reconciliation. The reconciler is already probabilistic-ready (fully vectorized over samples), so the open question is purely C-37's: does that deliverable reconcile grid draws to a country **point total** (well-defined, no alignment needed) or to country **draws** (needs a defined draw-alignment — and today CM models are point-only / independently trained, so no aligned draws exist)? **This decision gates the probabilistic-reconciliation-on-`PredictionFrameEnsembleManager` work** (pipeline-core#200, under epic #193); resolve it once the UN models / CM target are defined.
+
 ---
 
 ### C-38: Reconciliation grouping is O(groups × N) and materializes the whole grid frame — won't scale to global volume
@@ -451,6 +453,8 @@ Tier 2: structural fragility under the realistic change of wiring to global, wit
 
 **Update 2026-06-24 — compute RESOLVED.** The per-group `np.nonzero(inverse == gi)` was replaced with **group-by-sort** (`argsort(inverse)` + contiguous slices from `np.unique` counts, O(N log N), one index array). Parity stays **bit-exact** (`tests/test_reconciliation_grouping.py`, `test_reconciliation_e2e_parity.py` → 0.0) and a scale guard (`tests/test_reconciliation_scale.py`) protects against regression. **Residual (still open):** the module holds the whole pgm frame in memory at once; at global volume the **caller must chunk by time** (reconciliation is independent across months) — the chunk-by-time contract is documented in the CIC (`docs/CICs/ReconciliationModule.md` §5), to be **verified on a global-volume dry-run at S7 (#39)**. This entry stays open until that verification.
 
+**Update 2026-06-24 (reframe — residual now near-term):** the memory residual is no longer "verify someday." The upcoming UN-agency deliverable reconciles **frames with ~1024 pooled draws** — squarely in the >100 GB-at-global regime. When pipeline-core's `PredictionFrameEnsembleManager` wires probabilistic reconciliation (pipeline-core#200, under epic #193), the **caller must chunk by time** (reconciliation is independent across months) and **measure peak memory on a global-volume dry-run** as part of that work. The reconciler code itself is unchanged (compute already O(N log N)); this is a consumer-side obligation.
+
 ---
 
 ### C-40: FAO delivery logic fused to pipeline-core via double inheritance + interleaved infrastructure