docs(spec): P36 build outcomes (refs #588-#597)

This commit is contained in:
kert
2026-08-13 16:28:51 -04:00
parent a444d75f39
commit bfe95d63a4

View File

@@ -180,3 +180,49 @@ Dependency order: 1–3 independent; 4 needs 2; 5 needs 3+4+7; 6 needs 2;
- MIPS scoring/measure modeling beyond the QP/Advanced-APM parameters above.
- Cloud LLM usage — the llm module stays local-only per project policy.
- Modeling every NPRM addendum; only Addendum B (RVUs) is tabled.
## P36 build outcomes (2026-08-13)
Issues #588–#597, milestone P36 (17 commits, 4a9be6f..a444d75, branch
`p36-cy2026-rule`). Deviations and measurements vs the design:
- **Comment farm was already complete** — the design's "only ~2,905 of
14,092 extracted" understated prior work: all 2,905 attachment-bearing
comments had extractions; the remaining ~11.2k are body-only (abstract
fallback covers them), and the whole corpus (164k comments, 726,620
chunks incl. 62,486 for CMS-2025-0304) was already embedded post-P33.
#596 reduced to rule-text indexing.
- **Rules collection**: 5,769 chunks (2,254→2,208 NPRM CMS-1832-P,
3,611→3,561 final CMS-1832-F) after an inline-HTML-strip fix round and
force reindex; final-rule bib item gained `cms-rule:CMS-1832-F`.
- **pfs.rvu_proposed**: 14,169 rows (NPRM Addendum B), in aco.duckdb +
replica + DuckLake; `cms.ingest_log` provenance (18 rows/2 runs) with
sha256, FR citation, pincite. Column names deviate from `pfs.rvu`
(`modifier`/`mod`, `nonfac_pe_rvu`/`non_fac_pe_rvu`) — follow-up #607.
- **CY2026 NPRM parameters** (all source-verified at 90 FR 32352 et seq.):
proposed CF 33.4209 non-QP / 33.5875 QP, anesthesia 20.5728/20.6754,
BN adjustor 1.0055, telehealth originating-site fee $31.85, comment
close 2025-09-12.
- **qpp module**: QP thresholds 75%/50% (partial 50%/35% — patient-count
value is a disclosed inference), revenue nominal-risk standard 8%
(414.1415(c)(3)(i)(A), established 83 FR / 2018-24170); the
benchmark-based standard (c)(3)(i)(B) is stated nowhere in the corpus
as a bare percentage → `None` with notes. APM incentive sunset and the
first-year QP/non-QP CF split (payment year 2026) encoded with
citations at 90 FR 49928/49979–49980.
- **Citation correction found by final review**: `RULES[2026]`'s
final-rule citation had been "90 FR 101174" — that is the *OPPS* final
rule's citation; the PFS final rule is **90 FR 49266–50481** (CFs at
49961). Corrected in registry + notebook (a444d75). Design note: the
spec itself restated the wrong citation as current state; future rule
cycles should re-verify pre-existing registry citations against newly
captured text.
- **Interstitials landed on the branch**: #598 (test poisoned global
polars.DataFrame → duckdb replacement-scan cache corruption; full
suite 14,023 green post-fix), #599 (stale .git/hooks copies removed),
CVE bumps aiohttp 3.14.3 / pyasn1 0.6.4 / cryptography 50.0.0
(#581–#587 closed with per-CVE exploitability rationale).
- **Notebook**: `notebooks/cy2026_pfs_proposed_rule.py`, 5 sections, all
values registry/lake-sourced (thresholds interpolated after review
fix), 18/18 headless integration green. CMS Table 92 (specialty
impact) is a scanned graphic → clearly-labeled CPT-range proxy.