PostgreSQL Unicode Sigma Evidence

Issue #127 tracked a portable dtcs:lower mismatch on PostgreSQL 16 arm64 when the compiler lowered each code point without retaining the surrounding sigma context.

The PostgreSQL compiler now derives final-sigma context from Unicode data known at compile time. It does not use PostgreSQL's locale-sensitive UPPER or LOWER functions to classify neighboring characters. The runtime still uses PostgreSQL for execution, so the same compiled plan is exercised across collations and architectures.

Arm64 verification

The issue reproduction was verified on 2026-09-12 with the official postgres:16 arm64 image:

Property	Observed value
PostgreSQL	16.14 on aarch64-unknown-linux-gnu
Architecture	arm64 host and PostgreSQL server
Database encoding	UTF8
Database collation	en_US.utf8
Database ctype	en_US.utf8
Locale provider	c

The focused PostgreSQL release-blocker tests passed with:

ETLANTIC_SQL_URL='postgresql+psycopg://postgres:postgres@127.0.0.1:55432/etlantic' \
  uv run pytest -q tests/portable_conformance/test_review_blockers.py -m sql

The result was 10 passed, 1 skipped, 6 deselected. The full SQL and portable SQL suites also passed with 51 passed, 1 skipped, 165 deselected.

The regression corpus covers non-final and final sigma, word boundaries, case-ignorable punctuation, and combining marks. Examples include:

Input	Expected dtcs:lower
A-Σ	a-σ
AΣ-B	aς-b
AΣ:B	aσ:b
A'Σ	a'ς
AΣ'B	aσ'b

The dedicated CI evidence step prints both the runner and PostgreSQL server architectures, plus the server version, encoding, locale, and locale provider. The assertions keep the proof bound to PostgreSQL 16 and a valid UTF-8 database environment.