Changelog

All notable changes to this project are documented here. The format follows Keep a Changelog, and the project uses Semantic Versioning.

[0.3.1] - 2026-11-04

Calibration tables regenerated and re-verified; four stale figures corrected.

Added

• scripts/check_repo.py and scripts/check_calibration.py, plus a validate GitHub Actions workflow: manifest shape, skill frontmatter, relative links and the inventory example on every push, and 190 assertions that data/*.csv still matches the published tables.
• references/calibration-method.md: corpus provenance, counts, what is generated versus descriptive, the verification gate, and the limits to state when quoting a figure.
• data/calibration-distributions.csv, calibration-summary.csv, calibration-merged-ab.csv and calibration-cross-criterion.csv: the same numbers in machine-readable form.

Fixed

• plugin.json and four Markdown files no longer carry a UTF-8 BOM. Strict JSON parsers rejected the manifest; the others were cosmetic but inconsistent.
• The Biology Evaluation distribution now shows the one report that scored 0, so the column sums to 220 rather than 219.
• The Chemistry and Physics per-grade profile tables and merged A/B lines now include the single grade-2 report in each subject; the merged counts are 271 and 212 (not 270 and 211), and the A/B subset is 148 session-tagged reports (41 / 72 / 35).
• The grade-label deviation sentence now reads 20 of 220 (9.1%), which is what the corpus produces against the published boundary map.
• plugin.json no longer declares a skills array. Agent Plugins 1.0.0 has a closed manifest schema and discovers skills from the fixed skills/ location, so the field was non-conformant and ignored by clients.
• Chemistry conventions no longer permit a standard deviation at n = 3; the shared n-rule (SD from ≥5 replicates, range below that) now applies, as it already did in data-and-analysis-guide.md and statistics-decision-tree.md.
• Chemistry uncertainty terminology corrected: the (uncertainty / value) × 100 formula is percentage uncertainty, not percentage error; percentage error and percentage difference are now defined separately, matching session-subject-reports.md.
• Digital-instrument uncertainty corrected to ±1 in the last digit (least count) in both physics-conventions.md and equipment-uncertainties.md; the previous "± half the last digit" rule contradicted its own worked example.
• Dot-to-dot guidance reconciled: joining points is acceptable for showing raw data, but a trend claim requires a best-fit line. data-and-analysis-guide.md previously said dot-to-dot was acceptable while anti-patterns.md capped it at 3–4.
• The missing-trial anti-pattern now points at strand 3 (methodology description), matching calibration-feedback-library.md and the teacher-check notes in ia-criteria.md.
• lab-inventory.example.yaml now complies with its own rule that count is a number or a stock string; "made fresh" and "on demand" moved into notes.
• Grade-boundary arithmetic in grade-boundaries.md corrected: 18/24 plus 60–65% on the papers reaches an overall grade 6, not a grade 7 (which needs roughly 75–80%).
• README no longer advertises an IB_DOCS_DIR variable that no skill implements.
• The assessed-sample count in marking-calibration.md now matches anti-patterns.md and the two sample-pattern files (7 + 7 chemistry/physics commentaries, not 6 + 5).

Added

• This changelog.

[0.3.0] — 2026-08-28

Initial public version: ib-sciences-praxis-coach (five modes, four-criteria IA model, subject conventions, calibration tables, assessment samples) and the optional equipment-inventory skill.