NEWS.md
In xemd: eXtended Empirical Mode Decomposition

The conditions the library signals carry the classes of the family console: xemdZeroEnergy, xemdInsufficientModes and xemdZeroEnvelope inherit from moduleFailure, and the warning of decomposeTSL() about a non-uniform time grid is a moduleWarning. Callers that caught the own classes keep working.
The command-line interface reads its arguments from the launcher and leaves the rendering of every message, warning and error to the installer's console renderer: an error is written as xemd: <message>, exit status 2 for a usage error (class xemdUsage) and 1 otherwise; the launcher accepts --verbose and --quiet. The package code is otherwise unchanged.

The maintainer scripts of inst/ (cran-check.R, cran-release.R, cran-submit.R, rhub-check.R) travel in the source package, as in the other modules of the family; the package code is unchanged.
The introductory vignette is titled "Introduction", and the vignettes, the articles, the README and two help pages write the package name in code font or with \pkg{} instead of as a plain word.

decomposeEEMD(), decomposeCEEMD(), decomposeMaskedEMD() and decomposeBEMD() no longer return a mode whose sifting lost the extrema needed for its envelopes. Each realization of EEMD and CEEMD stops before that candidate, masked EMD discards the stage when one of its phases loses them, and BEMD discards the incomplete vector candidate. The content of the attempt stays in the residue, the modes extracted before it are unchanged, and the divisors, phases and directions are preserved. A candidate that reaches the iteration limit is kept and flagged converged = FALSE, as the reference implementations do.
Those four decompositions return termination, a data frame that records the attempted mode index, the subtractions of the attempt and its stopReason (terminalResidue, modeLimit or insufficientExtrema, and iterationLimit for a phase of masked EMD listed with the one that failed), with the realization and its sign, the phase or the first direction without an envelope where the method has them. decomposeTSL() carries the table with the keys of the series.
The tests gain a script that checks, on internal inputs recorded from the decomposition of two horizontal components of a strong-motion record, that an interrupted attempt leaves the residue unchanged and reports its cause.

The package carries eight contract test scripts on toy cases. The suite exercises the exported functions and checks output structures, reconstruction identities, preservation of the caller's random stream with explicit noise, and rejection of invalid inputs. The conformance tests shipped in version 0.1.1 are no longer part of the package.
The examples of decomposeCEEMD(), decomposeCEEMDAN(), decomposeEEMD(), decomposeICEEMDAN() and resiftModes() use 1000 samples instead of 3000; decomposeMEMD() uses 256 instead of 512.

Code adapted from pytvfemd and from curve_ext of the Synchrosqueezing Toolbox, which version 0.1.0 did not declare in DESCRIPTION, NOTICE or the file headers, is credited; executable code is unchanged. DESCRIPTION adds Stefano Bianchi and Alessandro Longo, authors of pytvfemd, and Jianfeng Lu, original author of curve_ext, and gives the maintainer the role trl.
NOTICE adds pytvfemd 0.3.14, with its copyright and license notice and the seven files that adapt its code for decomposeTVFEMD(), and lists in the SynchWave entry the two files of extractSSTModes() that adapt curve_ext and curve_ext_multi. Those nine files carry a header with the source, its authors, its license and the date of modification.
The six files adapted from SynchWave keep LGPL (>= 2) as the license of their source and state that xemd distributes them under GPL-3, as section 3 of the GNU Library General Public License version 2 permits.
The help pages of the methods whose article inst/REFERENCES.bib lists cite it, with its DOI where the entry gives one, and the pages of decomposeTVFEMD(), transformSST() and extractSSTModes() name the adapted code and its license.

decomposeEWT() takes detection, a list that selects the boundary detector of the EWT authors' code (locmax, locmaxmin or locmaxminf), the logarithm and the regularization of the spectrum, and with bandCount = NULL counts the bands by the relative threshold of Gilles (2013); detection = NULL keeps the previous partition unchanged. extractSSTModes() traces ridges of a synchrosqueezed transform by the penalized dynamic programme of the Synchrosqueezing Toolbox and reconstructs one component per ridge from a band around it. assessFractionalNoise() compares the energy of each mode with the fractional Gaussian noise reference and threshold tabulated by Flandrin, Gonçalves and Rilling (2005).
The help page of each function that uses code adapted from a public library names the library, its author and its license, as NOTICE does.
Decompositions are named decompose<METHOD>: EMD; EEMD, CEEMD, CEEMDAN and ICEEMDAN, with resiftModes(); masked EMD, with estimateMaskFrequency(), and iterated masking EMD; VMD, MVMD and SVMD; BEMD, MEMD and NA-MEMD; LMD and iterative filtering; EWT; TVF-EMD. The synchrosqueezed transform is transformSST().
Decompositions receive signal without a time vector and return lists with the samples in the first dimension: modes and, where the method defines them, residue or fitResidual, iterations, converged, stopReason and the fields of its family. Functions whose formulation depends on the time scale receive samplingInterval in seconds, and frequencies are in Hz. Noise-assisted methods receive their noise realizations and draw no random numbers. Undefined formulas stop with the condition classes xemdZeroEnergy, xemdZeroEnvelope and xemdInsufficientModes.
Analysis of modes: computeInstantaneousAttributes(), buildHilbertSpectrum(), assessIMF(), assessWhiteNoise(), measureDispersion(), measureOrthogonality(), measurePMSI(), measureQRF(), measureFCF(), measureDTW() and selectModesEnergy().
decomposeTSL() decomposes the series of a long table of records by record, quantity (ID) and component (OCID), regularizes grids that are not uniform with a warning, and returns homogeneous tables for every method. It replaces TSL2IMF(), retired without an alias. listDecompositions() returns the methods it accepts.
The command-line interface, the skill and the operations catalogue inst/api/operations.json follow decomposeTSL(): xemd decompose passes the JSON control object as the named list of arguments of the method, with nested objects as lists, arrays as vectors and numbers as doubles, and writes the list of tables to an RDS file; xemd methods prints listDecompositions(). Warnings, such as the regularization of a time grid, reach standard error as they occur.
xemd doctor reports, one line per check (--json for programs), whether R meets the minimum of xemd, xemd loads, its R packages and jsonlite are available, the installation receipt names the version that loads, the first xemd on PATH is the launcher of the same prefix and the working directory is writable; a failed check sets exit status 1. xemd doctor --install installs the unavailable R packages from CRAN and checks again; with --source DIR it also reinstalls the xemd library from that checkout, without sudo, whether or not its version differs. The report names no library or file path. describeModule() still returns the package and its version.
measureModesTSL(), selectModesTSL() and reconstructTSL() work on the tables of decomposeTSL() for every method: the metrics of each mode (frequency and bandwidth in Hz, energy, energy fraction and mean period) and of each series (number of modes and energy fractions of the remainder and of the cross products); the selection of modes with a declared criterion (explicit indices, a minimum energy fraction or a closed frequency band, to keep or to remove); and the sum of the retained modes, with or without the remainder. The command-line interface adds the verbs compare, metrics, select, reconstruct, export, run and batch, and every verb answers --help with its keys: run chains steps on one input and writes one RDS file per step and an index, batch runs the same steps on several inputs and stacks their tables, and export writes the tables as CSV files, long or with one column per mode. A series of t and s is read as one record, and a method that draws noise requires seed, which the output keeps. The verbs after compare work method by method and stack their tables with the stacking keys first: method, or source and then method after batch. A criterion modes whose entries mix method with none is a usage error. xemd doctor --install --source DIR reinstalls the library whether or not the version differs.