docs: drop draft-pdf CI workflow; scope survey-composition claims (review round 2)

- Remove .github/workflows/draft-pdf.yml: a durable CI job for the paper
  is overkill (user call). It served its purpose as a one-time compile
  check - the PDF built successfully and was visually verified; JOSS's
  editorialbot compiles the paper on demand during review.
- paper.md: survey variance no longer claimed to compose with "every
  estimator" / "uniformly" - now states per-estimator support documented
  in a compatibility matrix with unsupported combinations failing closed
- llms.txt / llms-full.txt / choosing_estimator.rst: "All estimators" ->
  "Most estimators" with SyntheticControl called out and a pointer to the
  Survey Design Support matrix (aligns with existing README wording)

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

Author: igerber

.github/workflows/draft-pdf.yml

@@ -2075,7 +2075,7 @@ clear_cache()
 
 ## Survey Support
 
-All estimators accept an optional `survey_design` parameter in `fit()`. Pass a `SurveyDesign` object to get design-based variance estimation.
+Most estimators accept an optional `survey_design` parameter in `fit()` (`SyntheticControl` rejects it as not yet supported); depth of support varies by estimator - see the compatibility matrix in `docs/choosing_estimator.rst` (Survey Design Support). Pass a `SurveyDesign` object to get design-based variance estimation.
 
 ```python
 from diff_diff import SurveyDesign, CallawaySantAnna

diff_diff/guides/llms-full.txt

@@ -104,7 +104,7 @@ Full practitioner guide: call `diff_diff.get_llm_guide("practitioner")`
 
 ## Survey Support
 
-All estimators accept an optional `survey_design` parameter. Pass a `SurveyDesign` object to get design-based variance estimation:
+Most estimators accept an optional `survey_design` parameter (`SyntheticControl` does not yet support it); coverage and weight types vary by estimator - see the [Survey Design Support matrix](https://diff-diff.readthedocs.io/en/stable/choosing_estimator.html#survey-design-support). Pass a `SurveyDesign` object to get design-based variance estimation:
 
 - **Design elements**: strata, PSU, FPC, weight types (pweight/fweight/aweight), lonely PSU handling, nest
 - **Variance methods**: Taylor Series Linearization (TSL), replicate weights (BRR/Fay/JK1/JKn/SDR), survey-aware bootstrap

diff_diff/guides/llms.txt

@@ -771,10 +771,10 @@ If you're unsure which estimator to use:
 Survey Design Support
 ---------------------
 
-All estimators accept an optional ``survey_design`` parameter in ``fit()``
-(``SyntheticControl`` does not yet support it and raises ``NotImplementedError``).
+Most estimators support an optional ``survey_design`` parameter in ``fit()``
+(``SyntheticControl`` accepts the parameter but raises ``NotImplementedError``).
 Pass a :class:`~diff_diff.SurveyDesign` object to get design-based variance
-estimation. The depth of support varies by estimator:
+estimation. The depth of support varies by estimator and variance method:
 
 .. note::
 

docs/choosing_estimator.rst

@@ -71,7 +71,8 @@ staggered-adoption support. General-purpose causal inference toolkits such as `D
 `diff-diff` was built as a new library, rather than as contributions to these packages,
 because its central contribution is cross-cutting: one estimator contract, one shared
 inference core, and an influence-function architecture that composes design-based survey
-variance with every estimator in the taxonomy. To our knowledge, no existing DiD software
+variance across the estimator taxonomy, with per-estimator support documented in a
+compatibility matrix and unsupported combinations failing closed. To our knowledge, no existing DiD software
 in any language provides design-based variance estimation for complex survey data, and no
 Python package covers the modern estimator taxonomy end-to-end; `diff-diff` provides
 both, validated against the R reference implementations where they exist.
@@ -130,8 +131,10 @@ is not identified, rather than silently reporting partial results.
 Two design choices carry the survey capability and the deployment story. First, estimators
 compute influence functions for their target parameters, so design-based variance - Taylor
 series linearization over strata and clusters, replicate weights, survey-aware
-bootstrap - composes uniformly with estimators as different as Callaway-Sant'Anna and
-synthetic DiD instead of requiring per-estimator derivations. Second, the runtime
+bootstrap - composes through one shared mechanism with estimators as different as
+Callaway-Sant'Anna and synthetic DiD; supported design-estimator combinations are
+documented in a per-estimator matrix, and unsupported ones are rejected explicitly rather
+than silently approximated. Second, the runtime
 dependency footprint is minimal by policy - numpy, pandas, and scipy only - keeping the
 library easy to install in restricted industry environments; high-dimensional fixed
 effects are handled by within-transformation rather than by delegating to a heavier