MyPII 1.0.1

Dokumentasi sahaja; pengecam v2 tidak berubah. / Documentation only; recognizer v2 is unchanged.

Changed

• README follows the Sasa Teguh open-source layout: logo, CI, PyPI, DOI and licence badges, English first and a complete Malay
  section at the end, and a "Work with Sasa Teguh" section.
• README links are absolute, so they work on the PyPI project page; the Zenodo DOI is in the README and CITATION.cff.

MyPII 1.0.0

Keluaran awam pertama; membawa pengecam v2. / First public release; ships recognizer v2.

Added

• Three Presidio recognizers: MY_NRIC (MyKad number with birth date and official JPN place-of-birth code check), MY_SSM
  (SSM registration number: new 12-digit format, old format, LLP) and MY_PHONE (Malaysian phone numbers), with context words for
  Malay, English and Manglish text (ADR-001).
• Recognizer v2: OCR normalisation inside numbers, line-broken numbers, bracketed Sabah and Sarawak area codes, the LLP rule
  (year 2012 or later) and a small local context classifier (logistic regression, numpy only, sha256-checked artifact) (ADR-018).
• all_recognizers(), an English alias of semua_pengecam().
• MyPII-Bench: 11,000 synthetic documents in seven splits, three of them sealed test splits locked by sha256 (ujian, ujian2,
  ujian3), with a dataset card and a canary string (data/README.md).
• Evaluation script with span-level precision, recall and F1, bootstrap intervals, Holm-corrected tests and an iteration log
  (ADR-006, ADR-007, ADR-014).
• Baselines scored by the same evaluation script: stock Presidio, malaysia-pii-guard and zero-shot LLMs, with LLM answers cached in the repo so
  scores reproduce offline (ADR-011, ADR-016, ADR-017).
• Pre-registration (docs/PRA-DAFTAR.md) and the archive of every test run with manifests (hasil/arkib/, docs/PROVENANS.md).
• Human label review protocol and results on a 300-document sample: one human plus one LLM reviewer (ADR-015).
• Comparison reports docs/BANDINGAN.md (dev, v1) and docs/BANDINGAN-V2.md (dev3 and ujian3, v2).

Changed

• The wheel now carries config/ and the classifier artifact inside the package, so pip install mypii works without a clone.
• Evaluation and training tools moved from extras to dependency groups (uv sync --group nilai --group garis --group latih),
  because PyPI rejects direct URL dependencies in package metadata.

Known limitations

• All benchmark text is synthetic and labels are machine-generated; results on real text are not measured.
• On the older-style dev split, v2 scores lower than v1 (F1 0.840 versus 0.888), mostly MY_SSM precision.
• Phone recall is optimistic: synthetic phone labels only use number ranges that libphonenumber accepts (ADR-002).
• The label review has one human reviewer, not two.