Natural-language compatibility review

This review covers the supplied English/French examples and all eight edtfy test files, including its disabled shared-component collection examples. Selected edtfy examples are credited to Nicolas Barbotte (MIT). We use the examples to identify syntax gaps; their older EDTF outputs are not a conformance oracle.

The tables are generated by node packages/natural/scripts/generate-compatibility.cjs. The phrase expectations live in packages/natural/tests/fixtures/compatibility.json and are enforced by compatibility.test.ts. English numeric examples use en-US, French uses fr-FR, and rolling short years use January 1, 2026. All alternatives are checked, in confidence order. Separate shared fixtures and round-trip tests cover Spanish and all six regional locale presets.

Corrected patterns

• A known day survives an unspecified month: 1870-XX-12 renders as 12th of unknown month, 1870, día 12 de mes desconocido, 1870, or 12 d'un mois inconnu, 1870. These phrases parse back without losing the day.
• Dates and seasons share interval endpoints: march 1988 - spring 1990 becomes 1988-03/1990-21. This is a supported extension, excluded by the optional strict interoperability profile. Reversed intervals are rejected. Season calendar bounds remain application conventions.
• French De … à …, articles, and apostrophe contractions work with shared date semantics. Season qualifiers work before or after the season. Century-wide qualifiers apply to both boundaries.
• Missing-year forms, semicolon lists, short range end years, ordinal/roman centuries, selected abbreviations, and validated weekday prefixes have regression coverage. January 12 produces only XXXX-01-12; January 0012 selects the historical month/year.

Modern EDTF distinctions

The current LOC specification replaced u with X, combined ?~ into %, and replaced interval keywords with empty/.. endpoints. We do not accept obsolete literal spellings as modern EDTF.

Since 1970 means 1970/..; an unknown end is 1970/ and renders explicitly as unknown. 20th century means 1901/2000, whereas 19XX covers 1900–1999. January 1999 to December 1999 is an interval, not a synonym for the unspecified month 1999-XX. 2020; 2021 is an inclusive list, {2020,2021}; bare 2020,2021 is not valid EDTF. Strict event-date cutoffs use sets and exclude the cutoff unit. Fuzzy cutoff phrases remain errors.

Bare one- or two-digit years require an era marker or four-digit spelling: write 90 CE or 0090 for year 90. Write 1990 explicitly or use '90 for the rolling window. Short end years in a range inherit the start century, advancing to the next century if necessary: 1851-52 → 1851/1852. Valid literal EDTF takes precedence: 1851-12 remains December 1851.

Preferred rendering equivalents

These are our preferred phrases for the valid modern equivalents of the supplied rendering examples. Exact wording and capitalization may differ; tests check the meaning after parsing the preferred rendering. Storage should retain EDTF, because display options can suppress qualifiers or eras.

EDTF	English	French	Spanish
1975-07-01	July 1, 1975	1 juillet 1975	1 de julio de 1975
1975?	1975 (uncertain)	1975 (incertain)	1975 (incierto)
1975-22	Summer 1975	été 1975	verano 1975
199X	1990s	Les années 1990	Los años 1990
19XX	1900s	19XX	19XX
1970/1980	1970 to 1980	1970 à 1980	1970 a 1980
1970/..	1970 to open end	1970 à fin ouverte	1970 a fin abierto
1970/	1970 to unknown	1970 à inconnu	1970 a desconocido
1970~/1980~	1970 (approximate) to 1980 (approximate)	1970 (approximatif) à 1980 (approximatif)	1970 (aproximado) a 1980 (aproximado)
1999-XX	some month in 1999	un mois de 1999	algún mes de 1999
{2020,2021}	All of: 2020, 2021	Toutes ces dates: 2020 et 2021	Todas estas fechas: 2020 y 2021
1870-XX-12	12th of unknown month, 1870	12 d'un mois inconnu, 1870	día 12 de mes desconocido, 1870

Phrase-by-phrase results

Locale	Input	EDTF candidates / decision	Explanation / source
en-US	July 1, 1975	1975-07-01
en-US	1975?	1975?
en-US	summer 1975	1975-22
en-US	1990s	199X
en-US	20th century	1901/2000	Numbered centuries use historical boundaries; 19XX covers 1900–1999 instead.
en-US	1970 to 1980	1970/1980
en-US	since 1970	1970/..	An open endpoint is distinct from the unknown endpoint in 1970/.
en-US	circa 1970 to circa 1980	1970~/1980~
en-US	January 1999 to December 1999	1999-01/1999-12	An explicit interval is not the unspecified month 1999-XX.
en-US	2020; 2021	{2020,2021}	An inclusive list uses braces; adjacent years compact.
en-US	January 12, 1940	1940-01-12
en-US	90	Rejected	A bare one- or two-digit year requires an era marker or four-digit spelling.
en-US	'90	1990
en-US	January 2008	2008-01
en-US	the year 1800	1800
en-US	10/7/2008	2008-10-07; 2008-07-10	US ordering ranks first; both valid interpretations survive.
en-US	1860?	1860?
en-US	1862 (uncertain)	1862?
en-US	circa Feb 1812	1812-02~
en-US	c.1860	1860~
en-US	ca1860	1860~
en-US	approx 1860	1860~
en-US	ca. 1860s	186X~
en-US	circa 1840s	184X~
en-US	ca. 1860s?	186X%	Modern EDTF combines uncertainty and approximation as %.
en-US	c1800s?	18XX%; 180X%	Both century-block and decade readings remain ambiguous; a qualifier does not disambiguate them.
en-US	January 12	XXXX-01-12	A month and day without a year. Write January 0012 for the historical year.
en-US	January	XXXX-01
en-US	7/2008	2008-07
en-US	month in 1872	1872-XX
en-US	day in January 1872	1872-01-XX
en-US	day in 1872	1872-XX-XX
en-US	Autumn 1872	1872-23
en-US	Fall 1872	1872-23
en-US	earlier than 1928	[..1927]	Strict event-date cutoff excludes 1928; it is a date-choice set.
en-US	later than 1928	[1929..]	Strict event-date cutoff excludes 1928.
en-US	before January 1928	[..1927-12]	Strict cutoff shifts one month.
en-US	after about the 1920s	Rejected	A fuzzy cutoff has no exact exclusive boundary; use an exact cutoff or literal EDTF with the intended meaning.
en-US	1st century	0001/0100	Historical century boundaries differ from 00XX.
en-US	10c	0901/1000
en-US	19th century?	1801?/1900?	The qualification applies to both boundaries of the century.
en-US	a day in about Spring 1849?	Rejected	EDTF has no day component on a season: 1849-21-XX is invalid. Use circa Spring 1849? (1849-21%) if season precision is sufficient, or specify calendar dates.
en-US	1851-1852	1851/1852
en-US	1851-1852; printed 1853-1854	Rejected	Bibliographic event annotations are not date syntax. The parser rejects the whole input instead of discarding the printing date.
en-US	1851-52	1851/1852
en-US	1856-ca. 1865	1856/1865~
en-US	1860s-1870s	1860/1879	Interval endpoints expand the decade start/end according to the existing parsing convention.
en-US	1920s - early 1930s	1920/1933	The early-decade convention retains the explicit early modifier.
en-US	1938, printed 1940s-1950s	Rejected	No silent truncation of annotated prose; parse each intended event separately.
en-US	12th of unknown month, 1870	1870-XX-12
en-US	march 1988 - spring 1990	1988-03/1990-21
fr-FR	1 Juillet 1975	1975-07-01
fr-FR	1975?	1975?
fr-FR	été 1975	1975-22
fr-FR	Les années 1990	199X
fr-FR	20e siècle	1901/2000	Numbered centuries are not equivalent to 19XX.
fr-FR	De 1970 à 1980	1970/1980
fr-FR	Depuis 1970	1970/..	Open, not unknown, end.
fr-FR	De 1970 environ à 1980 environ	1970~/1980~
fr-FR	De Janvier 1999 à Décembre 1999	1999-01/1999-12	Preserves explicit interval phrasing.
fr-FR	2020; 2021	{2020,2021}
en-US	mar the 29 1988	1988-03-29	edtfy test
en-US	bet the 03/28/1988 and the 03/30/1988	1988-03-28/1988-03-30	edtfy test
fr-FR	le 29 mar 1988	1988-03-29	edtfy test
fr-FR	du 28 mars 1988 au 30/03/1988	1988-03-28/1988-03-30	edtfy test
en-US	monday 03/29/1988	Rejected	March 29, 1988 was Tuesday; weekdays must match. edtfy test
fr-FR	jeudi 29/03/1988	Rejected	The weekday contradicts the date. edtfy test
en-US	from sep 1988 to 04/30/1988	Rejected	Reversed intervals are invalid. edtfy test
en-US	march 1988 - winter 2005?	1988-03/2005-24?	edtfy test
en-US	from around sep 10 1988? to unknown	1988-09-10%/	edtfy test
fr-FR	environ l'hiver 1988?	1988-24%	edtfy test
fr-FR	1988 - en cours	1988/..	edtfy test
en-US	XIXth century	1801/1900	edtfy test
fr-FR	XVIIIème siècle	1701/1800	edtfy test
fr-FR	IV siecle	0301/0400	edtfy test
fr-FR	2nd siècle après JC	0101/0200	edtfy test
en-US	12th March 24BC	-0023-03-12	edtfy test
fr-FR	printemps 4AEC	-0003-21	edtfy test
en-US	bf. 1988	[..1987]	edtfy test
fr-FR	ap. 1988	[1989..]	edtfy test
en-US	<1988	[..1987]	edtfy test
fr-FR	>1988	[1989..]	edtfy test
en-US	198u	Rejected	Legacy u masks are not modern EDTF. Use 198X. edtfy test
fr-FR	11/0u/1988	Rejected	Legacy slash masks are intentionally unsupported. Use modern EDTF 1988-0X-11. edtfy test
en-US	11/10/21988	Rejected	The selected EDTF profile supports extended years as year-only values, not extended full calendar dates. edtfy test

Remaining scope differences

The upstream tests include incorrect weekdays and reversed intervals; we reject them. Weekday prefixes constrain the full calendar date instead of being discarded. Draft u/* slash masks should be converted explicitly to modern EDTF with X; extended years remain year-only in the selected profile.

Shared-component collection syntax such as spring or summer 2015 and April 1 or 2 2015 remains outside this change (these cases are disabled in edtfy's suites). Use explicit members, such as One of: spring 2015, summer 2015 or One of: April 1, 2015, April 2, 2015. A future extension needs contextual component propagation through all collection members and language-specific word order, rather than guessing a year from the last number.

Bibliographic event labels such as printed are not discarded. Parse each event separately so dates and event roles are retained. A day within an approximate season cannot be encoded by appending -XX to a season code; use season precision if sufficient, or explicit calendar dates. No broad text-stripping or obsolete-EDTF compatibility mode was added.