edf2csv 0.6.56 → 0.6.58

package/docs/CHANGELOG.md

@@ -3,6 +3,47 @@
 Notable changes to edf2csv. Versions follow [semantic versioning](https://semver.org); while the
 major version is 0, a minor bump may contain breaking changes.
 
+## 0.6.58
+
+### Added: CI runs the six sweeps it has always claimed as its evidence
+
+The correctness page names seven checks as how this project knows what it claims — the estimate
+sweep, the layout sweep, the round trip, the corruption fuzzer, the batch trees, the narrowing
+sweep, and the pyEDFlib cross-check. CI ran none of them. `npm test` covers the suite; the
+harnesses are separate scripts, and nothing invoked them on a push, so a regression in an
+invariant that only a sweep can see would have shipped in silence.
+
+Six of the seven now run as their own job, each as a named step so a failure says which invariant
+broke rather than "sweeps". A separate job because they take minutes where the suite takes one,
+and GitHub runs jobs concurrently — a pull request waits for the slowest, not the sum. One Node
+version, since these assert what the tool computes and `core` is what covers computing it on
+both.
+
+The seventh, `crossvalidate`, stays manual: it needs pyEDFlib installed, and a check whose value
+is that it compares against an independent implementation should not be silently skipped when
+that implementation fails to install.
+
+## 0.6.57
+
+### Added: a sweep checking that narrowing a conversion returns the part it names
+
+`--channels` selects columns and `--start`/`--end` selects rows. Both are documented as
+selections rather than transformations — output-files puts it as "a section converted on its own
+lines up with the full conversion" — and nothing ran it. The layout sweep compares wide against
+long, the estimate sweep compares predicted row counts against written ones, and neither asks
+whether the rows a narrowed run writes are the rows the full run wrote.
+
+`npm run narrowing` takes each fixture's full conversion as the truth, converts every channel
+alone and three windows out of it, and requires the narrowed output to be the corresponding slice
+byte for byte: 106 single-channel selections and 253 windows over 50 recordings. Everything
+agrees, which is the result to want from a check written after the fact.
+
+Its first run did not agree, and the disagreement was the sweep's. Window bounds were read back
+off the CSV, where `time_s` is rounded — 8/17 is 0.47058823… and prints as `0.47059` — while a
+conversion filters the exact sample times, so a bound taken from the file sits just past the
+sample it came from and the conversion rightly keeps a row the check excluded. Bounds now sit
+halfway between two samples, where no rounding on either side can move a row across them.
+
 ## 0.6.56
 
 ### Fixed: the `UNREADABLE` case list, four of the six things it covers

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "edf2csv",
-  "version": "0.6.56",
+  "version": "0.6.58",
   "description": "Convert EDF, EDF+ and BDF biosignal recordings (European Data Format) to CSV from the command line. Local, streaming, and never resamples or alters units.",
   "keywords": [
     "edf",
@@ -67,7 +67,8 @@
     "fuzz:batch": "npm run build && npm run fixtures && node test/fuzz/trees.mjs",
     "estimate": "npm run build && npm run fixtures && node test/fuzz/estimate.mjs",
     "roundtrip": "npm run build && node test/fuzz/roundtrip.mjs",
-    "layouts": "npm run build && npm run fixtures && node test/fuzz/layouts.mjs"
+    "layouts": "npm run build && npm run fixtures && node test/fuzz/layouts.mjs",
+    "narrowing": "npm run build && npm run fixtures && node test/fuzz/narrowing.mjs"
   },
   "devDependencies": {
     "@types/node": "^22.10.0",