Assumptions

Adopted policies, on the record.

Every open question the library answered by adopting a policy is logged, with the diff signature that would expose the policy if it is wrong.

The discipline: an assumption is removed only when resolved with evidence, and a genuinely scholarly ambiguity graduates to the residue instead of being silently decided. The canonical, current record is ASSUMPTIONS.md in the repository. The load-bearing entries:

The standing assumptions

A-001: the ṭarīq is al-Shāṭibiyyah

Madd muttaṣil and munfaṣil at four counts and the rest of the Shāṭibiyyah parameter set. If the reference muṣḥaf followed a different ṭarīq convention, it would surface as systematic madd-length differences; none were observed.

A-002: the canonical text is Tanzil Uthmani, current snapshot

The reference dataset's indices are pinned to a 2017 Tanzil snapshot that differs slightly from the current one; an explicit per-verse index-mapping layer bridges them. Any mapping error would appear as spurious index-shifted diffs, which distinguishes it from a rule error.

A-004: package publication is user-gated

Publication to npm requires the author's credentials and review; the library does not self-publish.

A-006: sakt is riwāyah data

The four Ḥafṣ sakt sites are carried as transmitted data and applied by annotateVerse; they are not derivable from orthography and no derivation is pretended.

Flagged sites rather than silent choices

Where the transmission records two admissible ways (the ʿayn of the fawātiḥ at four or six counts; the rāʾ of فِرْقٍ and the stop on مِصْرَ and similar khilāf sites), the annotation is emitted with confidence: "flagged" and both ways are noted in the metadata.

Resolved entries

Entries are closed only with evidence, and the closure is recorded. Two examples: the former A-008 (a policy for a madd before a sākin hamzah) became unnecessary when the 2:72 letter-model defect was found and fixed (R-006); and the basmalah-prefix scope question was closed as a harness classification rather than a rule difference, as described in the method.