Fix registry fallback notes to match actual trigger conditions

Broaden sieve fallback descriptions from "singular basis matrix" /
"exception-triggered" to "when no sieve degree K succeeds" — covers
both the n_basis >= n_group early-exit and singular system cases.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: igerber

docs/methodology/REGISTRY.md

@@ -618,8 +618,8 @@ where `q_{g,e} = pi_g / sum_{g' in G_{trt,e}} pi_{g'}`.
 - **Single pre-treatment period (g=2)**: `V*_{gt}(X)` is 1x1, efficient weights are trivially 1, estimator collapses to standard DiD with single baseline
 - **Rank deficiency in `V*_{gt}(X)` or `Omega*_{gt}(X)`**: Inverse does not exist if outcome changes are linearly dependent conditional on covariates. Detect via matrix condition number; fall back to pseudoinverse or standard estimator
 - **Near-zero propensity scores**: Ratio `p_g(X)/p_{g'}(X)` explodes. Overlap assumption (O) rules this out in population; implement trimming or warn on finite-sample instability
-- **Note:** On sieve ratio estimation failure (singular basis matrix at all K values), the estimator falls back to a constant ratio of 1 for all units with a UserWarning. The outcome regression adjustment remains active, so the generated outcomes (Eq 4.4) still incorporate covariate information via the m_hat terms. The DR property ensures consistency as long as the outcome regression is correctly specified.
-- **Note:** On sieve inverse propensity estimation failure (algorithm step 4), the estimator falls back to unconditional n/n_group scaling with a UserWarning, which reduces to the unconditional Omega* approximation for the affected group. Both sieve fallbacks are exception-triggered and logged.
+- **Note:** When no sieve degree K succeeds for ratio estimation (basis dimension exceeds comparison group size, or all linear systems are singular), the estimator falls back to a constant ratio of 1 for all units with a UserWarning. The outcome regression adjustment remains active, so the generated outcomes (Eq 4.4) still incorporate covariate information via the m_hat terms. The DR property ensures consistency as long as the outcome regression is correctly specified.
+- **Note:** When no sieve degree K succeeds for inverse propensity estimation (algorithm step 4), the estimator falls back to unconditional n/n_group scaling with a UserWarning, which reduces to the unconditional Omega* approximation for the affected group.
 - **All units eventually treated**: Last cohort serves as "never-treated" by dropping last time period (Phase 1: raises ValueError; last-cohort-as-control fallback planned for Phase 2)
 - **Negative weights**: Explicitly stated as harmless for bias and beneficial for precision; arise from efficiency optimization under overidentification (Section 5.2)
 - **PT-Post regime (just-identified)**: Under PT-Post, EDiD automatically reduces to standard single-baseline estimator (Corollary 3.2). No downside to using EDiD -- it subsumes standard estimators