Fritts v0.3.0

Release v0.3.0 — 2026-07-01

Fritts v0.3.0 is a research-grade dendrochronology release. Every algorithmic
deviation identified in the v0.2 review cycle has been corrected (79 findings,
100% fix rate), the long-running analyses no longer freeze the UI, the ITRDB
client is hardened against SSRF, and the test suite has grown from ~95 to 230+
tests (234 collected, 213 passing, 21 skipped without DISPLAY, 1 pre-existing
libEGL failure that passes in CI).

Highlights

• Full TSAP-Win Heidelberg .fh reader — FH2.1/FH1.0 magic, KEY=/DATEBEGIN=/
  UNIT=/DENDRO= headers, decadal blocks with year labels, single-column format,
  multi-series files, -9999 missing values, 9999 stop code, UNIT-based divisor.
  Compatible with dplR::read.fh. Legacy HEADER:/DATA: subset auto-fallback.
  20 new tests.

• QThread workers for all long-running analyses — CrossDateWorker,
  DetrendWorker, BuildChronologyWorker, QCWorker move cross-date, detrend,
  chronology build, and COFECHA QC off the GUI thread with modal progress dialogs.
  The UI no longer freezes.

• Standard Wigley/Briffa (1984) EPS with per-tree metadata —
  EPS = (n_trees · r̄_bt) / (1 + (n_trees − 1) · r̄_bt) where r̄_bt is the
  overlap-weighted mean of between-tree pairwise correlations (within-tree pairs
  excluded). Set tree_id on each series' metadata dict. Falls back to all-pairs
  r̄ with a warning when no tree metadata is provided. R-bar matches. Chronology
  metadata now stores eps, rbar, n_trees.

• SSRF-hardened ITRDB client — HTTPS-only, hostname whitelist
  (ncei.noaa.gov + subdomains), redirect-blocking opener, streaming 100 MB cap
  (64 KB chunks), limit clamped to [1, 45], SSL context verification.

• Full undo/redo (including imports) — AddSeriesCommand wraps every
  add_series call site in ui/. Imports (file, image, ITRDB) are now undoable
  via Ctrl+Z. 200-step CommandStack (uses collections.deque).

• Tucson two-pass parser — ITRDB -9999 mid-series values are now correctly
  treated as missing (NaN), not stop codes. Real ITRDB .rwl files no longer
  get silently truncated at the first missing ring.

• Hypothesis fuzz tests for the Tucson parser — 5 property-based tests
  (never-crash, round-trip preservation, random bytes, unicode, embedded nulls)
  with max_examples=200, deadline=1000ms.

• Algorithm correctness fixes — AR(p) lag reversal fixed (CRITICAL), biased
  1/n autocovariance matching dplR::ar.yw, AIC uses n − order, Tukey
  biweight uses raw MAD (removed 1.4826 factor, matches dplR::tbrm),
  _pearson_to_t returns 0.0 for NaN input and -99.0 for perfect negative
  correlation, _wuchswerte preserves NaN.

• Spline detrend period parameter — Butterworth low-pass filter at
  1/period cycles/year when period is provided.

• RCS upgrades — Tukey biweight robust mean per cambial-age cohort (was
  median), pith_offsets parameter, min_depth parameter (default 5).

• Workspace rebinding — set_session() on SeriesListPanel, SeriesView,
  ChronologyBuilderDock properly rebinds panels after Open Workspace.
  inf/-inf now serialise as null (was only NaN). _workspace_path
  committed only after successful write.

• Image measurement — default series ID IMG_01 (was IMG_SERIES_01,
  violated 8-char validation), DPI read from QSettings, edge-replication
  padding (no artificial dark bands), grayscale 2-D auto-stacked to 3-channel.

• Preferences — DPI and theme settings are now actually read from
  QSettings on startup (were saved but never applied).

• CI — .github/workflows/ci.yml installs xvfb, libegl1, libgl1,
  libxkbcommon-x11-0 and runs xvfb-run -a python -m pytest tests/. The
  8 new qtbot interaction tests run in CI.

Test suite

• 234 tests collected, 213 passing, 21 skipped (need DISPLAY/libEGL),
  1 pre-existing failure (test_command_stack_cleared_on_session_replacement,
  libEGL import — passes in CI).
• 12 test files: test_regression.py, test_session_serialization.py,
  test_commands.py, test_crossdate.py, test_tucson_parser.py,
  test_io_formats.py, test_tucson_fuzz.py, test_itrdb.py,
  test_detrend.py, test_export_r.py, test_ui_fixes.py, test_ui_smoke.py.

Documentation

• README.md and docs/USER_GUIDE.md comprehensively rewritten for v0.3.0.
• docs/research-papers/Dendrochronology Algorithm Reference Validation.md
  has a Status: FIXED in v0.3.0 note appended to each algorithm section.
• CHANGELOG.md has a full v0.3.0 entry (Added/Changed/Fixed/Removed).

Known limitations

• Direct dplR numerical verification of EPS/spline/RCS/biweight is pending
  (R cannot be installed in the development sandbox). Formulas match published
  definitions; closed-form property tests verify the math.
• Spline period parameter is a Butterworth approximation to dplR::ffcsaps's
  FFT-based spline.
• EPS without tree_id metadata falls back to an all-pairs formula that
  over-estimates EPS when multiple cores per tree exist (a warning is logged).
• The aggregate per-series COFECHA health metric is still per-segment only.

Historical release notes: release_notes_v0.1.3.txt, release_notes_v0.1.4.txt,
release_notes_v0.1.5.txt.

What's Changed

• chore(deps): update pyqtgraph requirement from <0.14,>=0.13 to >=0.13,<0.15 by @dependabot[bot] in #1
• v0.3.0: Comprehensive code review fixes (79 findings, 100% fix rate) by @mabo-du in #5
• chore(deps): update numpy requirement from <2,>=1.26 to >=1.26,<3 by @dependabot[bot] in #3
• chore(deps): update lxml requirement from <6,>=5.0 to >=5.0,<7 by @dependabot[bot] in #4

New Contributors

• @dependabot[bot] made their first contribution in #1

Full Changelog: v0.2.0...v0.3.0

Fritts v0.2.0

Full Changelog: v0.1.5...v0.2.0

Fritts v0.1.5

Full Changelog: v0.1.4...v0.1.5

Fritts v0.1.3

Full Changelog: v0.1.2...v0.1.3

Fritts v0.1.2

Full Changelog: v0.1.1...v0.1.2

Fritts v0.1.1

Full Changelog: https://github.com/mabo-du/fritts/commits/v0.1.1