junte/Model
1
0
Fork 0
mirror of https://github.com/Hestia-Homes/Model.git synced 2026-06-30 13:10:47 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Record 17.0 band-4/5 synthesis transfer gaps at the seam (ADR-0028)

Browse source 
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This commit is contained in:
Jun-te Kim 2026-06-12 12:50:29 +00:00
parent 6c03f3323c
commit 1ff50374e7
1 changed files with 8 additions and 1 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

9
datatypes/epc/domain/mapper.py
View file

@ -3241,7 +3241,14 @@ def _synthesise_17_0_sap_windows(schema: RdSapSchema17_0) -> List[SapWindow]:
core (968/1000 band-1 with no measured band-1 windows; only 10 rich certs).
17.0 glazed_type codes (observed 1-4, 7) are a subset of 20.0.0's verified
1-8 space: route integer codes through the verified cascade; the "ND" string
falls back to the DG-modal default. Own seam so 17.0 can diverge."""
falls back to the DG-modal default. Own seam so 17.0 can diverge.
Transfer note (ADR-0028 validate-don't-re-fit): 17.0's Validation Cohort
splits two ways — band-4 (n=6) observes 0.170 vs predicted 0.223 (under),
band-5 (n=4) observes 0.150 vs predicted 0.092 (over). Both cohorts are tiny
and at the glazed extremes; band-1 (96.8%) is unfittable, so inherit stands.
Revisit only if synthesis fidelity (scoring) needs it. See
`scripts/eon/profile_corpus.py`."""
mgt = schema.multiple_glazing_type
glazing_type = (
_api_cascade_glazing_type(mgt)