Skip to content

Latest commit

 

History

History
2837 lines (2343 loc) · 140 KB

dates.md

File metadata and controls

2837 lines (2343 loc) · 140 KB

title: "Extended Legacy Format (ELF)" subtitle: Date, Age and Time Microformats date: 8 October 2019 numbersections: true ...

Extended Legacy Format (ELF):
Date, Age and Time Microformats

{.ednote ...} This is a first public draft of the microformats used for dates, ages and times in FHISO's proposed suite of Extended Legacy Format (ELF) standards. This document is not endorsed by the FHISO membership, and may be updated, replaced or obsoleted by other documents at any time.

Comments, discussion and other feedback on this draft should be directed to the [email protected] mailing list.


Latest public version: https://fhiso.org/TR/elf-dates This version: https://fhiso.org/TR/elf-dates-20181230


{/}

FHISO's Extended Legacy Format (or ELF) is a hierarchical serialisation format and genealogical data model that is fully compatible with GEDCOM, but with the addition of a structured extensibility mechanism. It also clarifies some ambiguities that were present in GEDCOM, and documents best current practice.

The GEDCOM file format developed by The Church of Jesus Christ of Latter-day Saints is the de facto standard for the exchange of genealogical data between applications and data providers. Its most recent version is GEDCOM 5.5.1 which was produced in 1999, but despite many technological advances since then, GEDCOM has remained unchanged.

{.note} Strictly, [GEDCOM 5.5] was the last version to be publicly released back in 1996. However a draft dated 2 October 1999 of a proposed [GEDCOM 5.5.1] was made public; it is generally considered to have the status of a standard and has been widely implemented as such.

FHISO are undertaking a program of work to produce a modernised yet backward-compatible reformulation of GEDCOM under the name ELF, the new name having been chosen to avoid confusion with any other updates or extensions to GEDCOM, or any future use of the name by The Church of Jesus Christ of Latter-day Saints. This document is one of five that form the initial suite of ELF standards, known collectively as ELF 1.0.0:

  • ELF: Primer. This is not a formal standard, but is being released alongside the ELF standards to provide a broad overview of ELF written in a less formal style. It gives particular emphasis to how ELF differs from GEDCOM.

  • ELF: Serialisation Format. This standard defines a general-purpose serialisation format based on the GEDCOM data format which encodes a dataset as a hierarchical series of lines, and provides low-level facilities such as escaping.

  • ELF: Schemas. This standard defines flexible extensibility and validation mechanisms on top of the serialisation layer. Although it is an optional component of ELF 1.0.0, future ELF extensions to ELF will be defined using ELF schemas.

  • ELF: Date, Age and Time Microformats. This standard defines microformats for representing dates, ages and times in arbitrary calendars, together with how they are applied to the Gregorian, Julian, French Republican and Hebrew calendars.

  • ELF: Data Model. This standard defines a data model based on the lineage-linked GEDCOM form, reformulated to be usable with the ELF serialisation model and schemas. It is not a major update to the GEDCOM data model, but rather a basis for future extension and revision.

{.ednote} At the time this draft was published, none of the other documents are yet at the stage of having a first public draft available, however FHISO's Technical Standing Committee (TSC) are working on them and hope to have first drafts available soon.

An explanation of the conventions used in this standard can be found in {§conventions}, and the general concepts associated with time, calendars and uncertainty are defined in {§concepts}. A generic syntax for expressing dates in arbitrary calendars is given in {§generic}, which {§modifiers} extends to support imprecisely known dates and dates written in natural language, and {§DatePeriod} extends to support date periods representing extended states of being. Four specific calendars are defined in {§calendars}, allowing the Gregorian, Julian, French Republican and Hebrew calendars to be used in the generic date syntax.

This standard defines five datatypes so that the formats defined here can be used in ELF. The elf:DateValue datatype defined in {§DateValue} is ELF's standard datatype for representing historical dates, while the elf:DatePeriod datatype defined in {§DatePeriod} is used to represent date periods, such as the period of coverage of a source. The elf:DateExact datatype defined in {§DateExact} is a more restricted format for expressing exact dates in the Gregorian calendar, and is used to record when ELF objects were created or last modified, typically in conjunction with the elf:Time format defined in {§time}. The elf:Age datatype defined in {§age} is used to represent individuals' ages. These datatypes contain only modest changes from GEDCOM, but should serve as a basis for future work on calendars.

Conventions used {#conventions}

Where this standard gives a specific technical meaning to a word or phrase, that word or phrase is formatted in bold text in its initial definition, and in italics when used elsewhere. The key words must, must not, required, shall, shall not, should, should not, recommended, not recommended, may and optional in this standard are to be interpreted as described in [RFC 2119].

An application is conformant with this standard if and only if it obeys all the requirements and prohibitions contained in this document, as indicated by use of the words must, must not, required, shall and shall not, and the relevant parts of its normative references. Standards referencing this standard must not loosen any of the requirements and prohibitions made by this standard, nor place additional requirements or prohibitions on the constructs defined herein.

{.note} Derived standards are not allowed to add or remove requirements or prohibitions on the facilities defined herein so as to preserve interoperability between applications. Data generated by one conformant application must always be acceptable to another conformant application, regardless of what additional standards each may conform to.

If a conformant application encounters data that does not conform to this standard, it may issue a warning or error message, and may terminate processing of the document or data fragment.

This standard depends on FHISO's Basic Concepts for Genealogical Standards standard. To be conformant with this standard, an application must also be conformant with [Basic Concepts]. Concepts defined in that standard are used here without further definition.

{.note} In particular, precise meaning of character, string, whitespace, whitespace normalisation, language tag, term, prefix notation, prefix, property, datatype and subtype are given in [Basic Concepts].

Indented text in grey or coloured boxes does not form a normative part of this standard, and is labelled as either an example or a note.

{.ednote} Editorial notes, such as this, are used to record outstanding issues, or points where there is not yet consensus; they will be resolved and removed for the final standard. Examples and notes will be retained in the standard.

The grammar given here uses the form of EBNF notation defined in §6 of [XML], except that no significance is attached to the capitalisation of grammar symbols. Conforming applications must not generate data not conforming to the syntax given here, but non-conforming syntax may be accepted and processed by a conforming application in an implementation-defined manner.

{.note} In this form of EBNF, whitespace is only permitted where it is explicitly stated in the grammar. It is not automatically permitted between arbitrary tokens in the grammar.

The grammar productions in this standard uses the S production defined in §2 of [Basic Concepts] to match any non-empty sequence of whitespace characters.

This standard defines five datatypes to represent time-related concepts in ELF, each of which is identified by a term name, which is simply an IRI. The concept of a datatype, as used by FHISO, is defined in §6 of [Basic Concepts], and the definition of each datatype in this standard includes a table listing its formal properties.

{.note} These properties include a formal statement that the datatype is datatype, define the pattern and non-trivial supertypes of the datatype, and say whether it is an abstract datatype. These concepts are defined in §6.1, §6.2 and §6.3 of [Basic Concepts]. The pattern is a regular expression written in FHISO's types:Pattern datatype defined in [FHISO Patterns]. This information forms part of an abstraction which allows applications to use a discovery mechanism to find out about unknown components, thus allowing them to be processed in more sophisticated ways than could be done with a truly unknown component. To support this, FHISO's web server has been configured to provide [Triples Discovery] on all terms defined in this standard. Such functionality is outside the scope of this standard, and is entirely optional in ELF. Most readers can safely ignore this formalism and the tables of properties given for each datatype.

This standard uses the prefix notation, as defined in §4.3 of [Basic Concepts], when discussing specific terms. The following prefix bindings are assumed in this standard:


elf https://terms.fhiso.org/elf/ xsd http://www.w3.org/2001/XMLSchema# types https://terms.fhiso.org/types/


{.note} The particular prefixes assigned above have no relevance outside this standard document as prefix notation is not used in the formal data model as defined by this standard. This notation is simply a notational convenience which makes the standard easier to read.

General concepts {#concepts}

{.ednote} It is anticipated that this section will be moved to [Basic Concepts] in a future draft of these documents.

Time {#time-concepts}

An instant is defined as an infinitesimally brief point in time.

{.note} Although defined as an infinitesimally brief point in time, it may be subject to the various forms of uncertainty described in {§uncertainty}.

{.example} King Alfred's birth occurred at some particular instant in the middle of the ninth century. Even though the year is not known with any great certainty, it is still an instant.

A time interval is defined as the section of time spanning between two specific instants.

{.example} The interval lasting from midday on 1 Feb 2018 until midday on 14 Feb 2018 is a time interval.

{.example} The lifetime of a particular individual is another example of a time interval, beginning at the instant of their birth and ending with the instant of their death.

A duration is the length of time elapsing between two instants, but without reference to any specific choice of start and end instants.

{.example} "3 days", and "34 years, 2 months" are two examples of durations expressed in natural language.

{.note} Durations differ from time intervals in that time intervals are durations with specific start and end instants. A time inteval has a duration associated with it, quantifying how long it lasts.

{.note} ELF does not provide a general-purpose duration datatype, but the elf:Age datatype defined in {§age} is a datatype customised for representing the duration of an individual's life.

Fundamental to ELF's handling of dates is a set of time intervals called calendar days, each of which spans from one midnight until the next.

{.note} A calendar day lasts for 24 hours, except when leap seconds is inserted or deleted, or when the local time zone changes, as in the transition to or from daylight saving time. In practical terms, it is a period during which the sun rises and sets exactly once, except in the polar regions.

{.note} Because midnight does not occur simultaneously around the world, the set of calendar days in one region may be offset compared to those in another region. The details vary depending on local legislation and custom. Currently, there can be three different calendar days happening simultaneously in various parts of the world: when a calendar day is just beginning in the Line Islands of Kiribati, it is still the previous calendar day in most of world's landmasses, and the calendar day before that in American Samoa. This means it is possible for a person to participate in an event on one calendar day, travel to another region, and subsequently participate in an event on the previous calendar day. If the second event is the person's death, this could theoretically result in a living person participating in an event the calendar day after their death.

{.ednote} Does this definition need loosening? Not all cultures consider the day to be begin at midnight. The Hebrew calendar defined in {§hebrew}, for example, is normally used with calendar days beginning at sunset, which is defined as 6pm using variable length hours. The Islamic and Bahá'í calendars do similarly. The definition of a calendar day is currently taken from [ISO 8691], but should it be loosened to allow such definitions?

A date is a way of identifying a particular calendar day.

{.note} ELF dates do not include an indication of either the time zone or the locale which leaves some ambiguity into the exact points in time that are meant. The [ISO 8601] concept of a date has the same ambiguity.

{.ednote} This standard deliberately does not specify whether the calendar day 1 January 2018 in Vancouver is the same calendar day as 1 January 2018 in Paris, which started nine hours earlier than the calendar day in Vancouver. This is left unspecified because [ISO 8601] is similarly vague. However if they are regarded as distinct calendar days, they are represented by the same date in ELF.

{.note} The definitions of an instant, a time interval, a duration, a calendar day and a date given here are intended to be fully compatible with the definitions of these concepts in [ISO 8601]. Any incompatibility between the definitions here and those in [ISO 8601] is unintentional.

{.ednote} These concepts have been defined here rather than by normative reference to [ISO 8601] because of the cost involved in obtaining a legal copy of [ISO 8601], and the likelihood that implementers will not do so.

A time of day is a way of identifying an instant within a calendar day, done by dividing an ordinary calendar day into 24 hours, each of which is subdivded into 60 minutes, each of which is further divided into 60 seconds. In ELF, a time of day is represented by the elf:Time datatype defined in {§time}.

{.note} A calendar days may exceptionally be divided differently if a leap second is inserted or deleted, or when the local time zone changes.

Calendars {#cal-concepts}

Many different systems for reckoning dates have been used throughout history and in different parts of the world. Such systems are called calendars, and ELF allows historical dates to be specified using many different calendars.

{.example} The simplest form of calendar is to count the number of calendar days which have elapsed since a particular day zero. The most popular such calendar is called the Julian Day (which is unconnected to the similar-sounding Julian Calendar). Its day zero is 24 November 4714 BC in the proleptic Gregorian Calendar, a day chosen to be before all recorded history. Written as a Julian Day, 1 January 2000 can be represented by the integer 2451545. Such calendars are not commonly used for writing historical dates as they are cumbersome and error-prone.

{.ednote} Nevertheless, FHISO might consider standardising the Julian day as a lightweight calendar for use as a common intermediate calendar during the conversion of dates from one calendar to another.

Many calendars make use of units of time which are longer than a calendar day, and the general framework for dates in ELF allows for two such units of time, a calendar month and a calendar year, whose definitions will be dependent on the particular calendar.

{.note} It is intended that a calendar year will typically be unit of time roughly equal to the time it takes the Earth to orbit the Sun, and a calendar month will be a unit of time intermediate in duration between a calendar day and a calendar year, and often loosely based on the time it takes the Moon to orbit the Earth. However these are not requirements, nor is it a requirement that all calendar years or all calendar months be of approximately equal length in a given calendar.

{.ednote} The flexibility to define calendar years and calendar months arbitrarily might be exploited in the future. FHISO are considering whether there is merit to defining a calendar for the Julian day. If defined it would not be for general use expressing historical dates, but rather to provide a way of expressing epochs in a calendar-neutral way when defining calendars. Because there is no applicable notion of a calendar month or calendar year with Julian days, and because the generic date syntax defined in {§generic} is most natural with calendars that have a calendar year, it is quite likely this calendar might define a calendar year and a calendar month to be identical to a calendar day. This standard does not prohibit this.

An incomplete date is a way of identifying a particular calendar month or calendar year without identifying a specific calendar day.

{.example} In the Gregorian calendar, "June 1953" is an incomplete date as it identifies a particular calendar month, but not a specific calendar day within that month.

{.note} Under this definition, an incomplete date is a date when it is being used to identify a particular calendar day, but with limited precision.

An epoch is an instant which serves as a reference point for a given calendar from which calendar years are numbered consecutively with an integer called the logical year, which either increases or decreases with time. When the logical year number increases with time, the epoch shall be first instant of the calendar year with the logical year number 1. When thee logical year number decreases with time, the epoch shall be the last instant of the calendar year with logical year number 1.

{.example ...} The epoch used in many forms of the Islamic Calendar is an instant during the Gregorian year AD 622, the year of the Hijrah when Muhammad moved from Mecca to Medina. The first calendar year of the Islamic calendar, called Anno Hegiræ 1 began at this epoch and has the logical year number 1. Subsequent Islamic calendar years have been numbered sequentially, AH 2, AH 3, etc., and have logical year numbers 2, 3, etc. The calendar year immediately before the epoch is commonly labelled 1 BH (standing for Before the Hijrah), and earlier calendar years are numbered backwards from the epoch. These have logical year numbers 1, 2, 3, etc. too.
{/}

{.note} This definition does not limit a calendar to having a single epoch.

A logical year number is not sufficient to identify a specific calendar year: it is also necessary to state the particular epoch from which calendar years are numbered, and whether logical year numbers increase or decrease with time. An epoch name is an identifier which represents these two things. An epoch name where logical year numbers increase with time is called a forwards epoch name, while an epoch name where logical year numbers decrease with time is called a reverse epoch name.

{.example} The Gregorian calendar has two distinct epoch names, "AD", standing for Anno Domini, which is a forwards epoch name, and "BC", standing for Before Christ, which is a reverse epoch name. Both use the same epoch, defined as midnight at start of 1 January AD 1.

The way years were counted in the past does not always fully coincide with the modern reckoning of years in that calendar, which is what defines the logical year, even though they are nominally reckoned from the same epoch. The year number according to the historical reckoning is called the historical year.

{.example} Prior to the adoption of the Gregorian calendar in many parts of the world, the year was considered to begin on the Feast of the Assumption (otherwise called Lady Day) which falls on 25 March. This was the case in England before 1752. Sources contemporary to the event record the execution Charles I as happening on 30 January 1648. This was in the month following December 1648 and would now be considered to be in 1649, as a result of which modern accounts usually describe it as happening on 30 January 1649. (This is not a result a change of the Julian calendar to the Gregorian one: in the Gregorian calendar this date is 9 February 1649.) In this example, 1649 is the logical year, while 1648 is the historical year.

{.example} The Byzantine calendar counted years since the supposed beginning of the world with an epoch on 1 September 5509 BC, however in early times some alternative dates were assigned to this epoch, including 25 March 5493 BC, sometimes known as the "Alexandrian" epoch. A definition of this calendar which used the traditional Byzantine epoch in 5509 BC would have logical years counted from then, but might also allow historical years to be counted from the Alexandrian epoch. This is an example where the logical year and historical year differ by more than one.

A calendar defines how the number of calendar days in each calendar month and the number of calendar months in each calendar year are determined. Stylistic and linguistic variations in the presentation of a date do not constitute separate calendars.

{.example} "31st August, 2018", "31 авг. 2018", "8/31/2018" and "2018年8月31日" are various ways in which the date which is represented in [ISO 8601] as "2018-08-31" might be presented. The differences between these presentations are merely stylistic or linguistic ones, and therefore these difference are not separate calendars: they are all written using the Gregorian calendar.

{.ednote} In due course, FHISO will need to clarify and perhaps revise this definition of what constitutes a distinct calendar. Does Roman day reckoning (e.g. "Prid. Kal. Sept" for 31 Aug) count as a separate calendar or is just a stylistic variation? What about regnal years (e.g. "31 Aug 67 Eliz II")? These are not strictly separate calendars, but it could be convenient to consider them as such in ELF if it is considered desirable for ELF to preserve the fact that the dates were recorded in these forms.

Uncertainty {#uncertainty}

The precision of a stated value, such as a date, is a measure of the specificity with which the value has been specified: the more specifically, the greater the precision. Values with relatively high or low precision may be described as relatively precise or imprecise, respectively.

{.example ...} It is more precise to say that the Battle of Agincourt was on St Crispin's Day, 1415, than it is to say that the battle occured during Henry V's reign. Both statements are true, but the former has greater precision because it identifies the specific day of the battle, while the latter identifies it only as falling within that nine-year reign.

Saying the battle was in the autumn of 1415 has an intermediate level of precision. This year might described as precise in comparision to the whole of Henry V's reign, or as imprecise when compared to the specific day. {/}

{.note} Values are stated imprecisely for many reasons, including when a more precise value is not known and when greater precision is considered irrelevant. Another reason is when the value being stated is inherently ambiguous.

When the value being stated is the instant at which some entity changed state, it is common for this instant not to be defined with arbitrary precision because there is a time interval during the transition when it is not well-defined what the state is. The duration of this time interval is known as the inherent ambiguity of the instant of the change.

{.example} Depending on the jurisdiction, the precise instant during a wedding when the couple become married may be ill-defined as there are several obvious possibilities. It could be argued to occur when the couple complete their vows, or when the priest declares the couple husband and wife; or it might be when the last signature is put on marriage certificate has been completed, or when the ceremony ends. If there is no single accepted definition, then there is likely several minutes of inherent ambiguity between the first and the last possibilities. This would be true even if the wedding had been videoed and carefully timed, as it is not due to lack of information on what happened and when.

A stated value is either exactly stated or approximately stated. An exactly stated value is one where it is well-defined exactly what values are considered to be consistent with the stated value.

{.example ...} The date of the Battle of Agincourt was stated in three different ways in the earlier example. In order of decreasing precision these were "St Crispin's Day, 1415", "the autumn of 1415", and "during Henry V's reign". The meaning of St Cripin's Day is well-defined: it is 25 October. Had the Battle of Agincourt in fact occurred on 24 October 1415, this would not be consistent with the statement that it happened on St Crispin's Day. "St Crispin's Day, 1415" is therefore an exactly stated value. "During Henry V's reign" is similarly exactly stated.

"The autumn of 1415" is very likely not well-defined. Some people define it as stretching from the autumnal equinox in late September to the winter solstice in late December, but this definition is by no means universal. Often it is used a more vague manner to refer to the later part of the year in the northern hemisphere. Unless context makes it clear that a specific, well-defined meaning of word "autumn" was intended, this is not an exactly stated value: it is therefore an approximately stated value.

The battle might also be described as happening in about 1415. This statement is true as the battle did in fact occur in 1415, but it is an approximately stated value. Had the battle actually been in 1414, would this be consistent with the description "about 1415"? Probably. But what about 1411? Or 1401? There is no general answer, and as a result "about 1415" is an approximately stated value. {/}

{.example} These concepts do not only apply to quantitative values. It is more precise to say that a person was born in the commune of Coutances, than to say say that person was born in metropolitan France. "The commune of Coutances" and "metropolitan France" are both exactly stated values. The person might also be described as born in northern France, which would normally be interpreted as an approximately stated value.

The precision range of an exactly stated value is defined as the set of values which would be considered consistent with the stated value.
One specific measure of precision is the precision range width, which is defined as the difference between the two most widely separated values in the precision range. When the value being specified is an instant, its precision range is a time interval, and its precision range width is a duration.

{.example} If a person is said to have died in 1967, this is consistent with the instant of their death being at any time between midnight at the start of 1 January 1967 and midnight at the end of 31 December 1967. The time interval between these two instants is the precision range and is a calendar year. In this example, the stated value is 1967 and its precison range is the duration 1 year.

{.example} If a person is said to have married in the 1910s, it is fairly clear this refers to a decade and therefore the precision range is 10 years. However if the person is said to have married in the 1900s, this might mean the decade or the century. Without further context, the intended precision range is unclear.

The accuracy of a value is a measure of how close a stated value is to the true value. A exactly stated value is said to accurate if the true value lies within the precision range of the stated value, and inaccurate otherwise. For an approximately stated value, the accuracy is relative: the further the stated value is from the true value, the less accurate or more inaccurate the stated value.

{.note} The precision of a value is unrelated to its accuracy. A value may be precise or imprecise independently of whether it is accurate or inaccurate.

{.example ...} The following table gives example instants of birth for Queen Victoria which are variously precise or imprecise, and accurate or inaccurate.


Precise Accurate 24 May 1819 at 4am Precise Inaccurate 19 Jun 1833 at 9pm Imprecise Accurate During the 1810s Imprecise Inaccurate During the 1790s


It is generally accepted that Queen Victoria was in fact born at 4.15am on 24 May 1819. {/}

{.note} In principle, the accuracy of any stated value is unknowable, though in practice some facts are so well established they can be regarded as proven for all practical purposes as the alternative would require there to have been a vast conspiracy. Much of the time the situation is not so clear.

The likelihood that a stated value is accurate is referred to as its reliability.

{.note} Although this is in theory a probability, reliabilities are usually described comparatively or quantitatively. A researcher may gauge the reliability of a stated value by considering the reliability of the sources in which it is stated, and the corroborating or contradicting evidence. Different researchers might reasonably reach different conclusions on the reliability of a stated value.

{.note} A stated value can be considered unreliable by virtue of being stated with excessive precision.

{.example ...} Suppose a man was last seen on 1 January and his corpse found on 31 January. The coroner determined the man had been dead for one to two weeks when found, but that no more precise date of death could be established. A newspaper obituary simply said he died in January, but a gravestone was erected giving his date of death as 21 January. A researcher might conclude the bare month given in the obituary is reliable because the relatively imprecise date is very likely accurate as it is consistent with the other evidence. The gravestone might be accurate, and it is not directly contradicted by the other evidence, but if the researcher believes the date "21 January" was made up, perhaps so that something could be put on the grave, it might be judged unreliable as there is a high likelihood that the true instant of death was not actually on 21 January.

This is not an example of inherent ambiguity. Depending on the circumstances of the death, there may have been a few minutes of inherent ambiguity as the man's life slowly ebbed away, but the bulk of the uncertainty is from lack of knowledge of what happened and when. {/}

Date concepts in other formats

{.ednote} This whole section may vanish in a future draft.

The datatypes defined in this standard are not recommended for use in serialisation formats other than with ELF.

{.ednote ...} In due course we need to decide FHISO's preferred way of handling dates and durations in other serialisation formats. GEDCOM X, for example, uses a format more closely aligned with [ISO 8601], and in early discussion on dates and in our call for paper submissions, we were erring in that direction too.

If we have one date format for ELF, another for GEDCOM X, and possibly even a third one for a future format of our own, we will probably want to make sure we don't end up with dates formatted for GEDCOM X appearing in ELF, or vice versa, otherwise an ELF application will need to know about every date format rather than just the ELF date formats. At some level, this requires converting dates between formats when data is transferred between systems. For data in the [ELF Data Model] or in the GEDCOM X data model, this is no problem as a data conversion stage will be required when converting between data models, and it can convert the date formats too.

A problem arises in FHISO's component standards, such as [CEV Concepts], which are intended to be usable in ELF, GEDCOM X and other data models. An ELF application will not necessarily know about CEV and will then see the CEV structures as unknown extensions, so there needs to some way of indicating that the ELF structure it is reading contains a date. This could be done by requiring an ELF schema to be present and have it specify the datatype for the payload, though that might be too onerous a requirement. Another option is to require a specific tag like DATE to be used, and special case this. The same issue may exist for ages too, but it is not general to all datatypes — just those which have to be formatted differently in different serialisations, which will hopefully be a minority of datatypes. {/}

Date formats {#dates}

ELF uses three different datatypes to represent dates, depending on the context.

  • elf:DateValue is used for historical dates, and is defined in {§DateValue}.
  • elf:DatePeriod is used to record the period of coverage a source, and is defined in {§DatePeriod}.
  • elf:DateExact is used to record the creation or modification date of various objects in the data model, and is defined in {§DateExact}.

As the first two of these datatypes are used to record historical dates, and ELF allows historical dates to be expressed using many different calendars, these two datatypes each allow dates in arbitrary calendars. This is achieved by providing a generic date syntax which all dates must match, regardless of calendar, and which begins with a calendar escape indicating the specific calendar in use. This generic syntax is defined in {§generic}, and extensions to it to support imprecisely known dates and dates written in natural language are given in {§modifiers}.

{.ednote ...} An earlier draft of this standard used a separate datatype for each calendar, and [ELF Serialisation] used @#D@ as a way of tagging the datatype. This approach was eventually abandoned because it could not cope with date ranges and date periods where the two end points used different calendars. [GEDCOM 5.5.1] permits this, and although many applications do not support it, the TSC considered the following use case to be important enough that ELF needed to support it too:

0 INDI
1 NAME George II
1 TITL King of Great Britain
2 DATE FROM @#DJULIAN@ 11 JUN 1727 TO @#DGREGORIAN@ 25 OCT 1760

Were there datatype per calendar, what should the datatype of the above DATE element's payload be? It's neither wholly Julian nor wholly Gregorian. Because of the difficulties with such constructs, we dropped the idea of making each calendar a separate datatype.

One option for solving this which the TSC seriously considered is to introduce compound calendars, along the lines of the proposal in CFPS 38, but this adds complexity due to the need to add a mechanism for defining compound calendars.

A separate complication comes from the fact that elf:DateValue and elf:DatePeriod are separate datatypes, and calendar-specific subtypes of each would likely be required. This could be solved by removing periods from the data model entirely, perhaps by having the serialisation layer split up DATE tags containing a period. This may make sense at a date model level too, if we model periods as two implicit events: one intiating and one concluding the period being discussed. There is less of a case for doing the same wtih ranges, so a solution to the problem of ranges with multiple calendars would still be required.

The TSC believe an approach along these lines could be made to work, but it would be too big a change to include in ELF 1.0. We also feel we should investigate alternative approaches before committing to a specific solution. Deferring this functionality until a future version of ELF will give us time to do give suitable consideration to the options available. {/}

Generic date syntax {#generic}

A date is represented in the generic date syntax as a sequence of five components: a calendar escape, followed by encodings of the calendar day, calendar month, calendar year and epoch name. Only the calendar year is required; the other components are optional, except that the calendar day cannot be present if the calendar month has been omitted. It matches the Date production.

Date   ::= (CalEsc S)? ((Day S)? Month S)? Year (S? Epoch)?

CalEsc ::= "@#D" [A-Z] [A-Z ]* "@"
Day    ::= [0-9]+
Month  ::= [A-Z] [A-Z0-9] [A-Z0-9]+
Year   ::= "-"? [0-9]+ ( "/" "-"? [0-9]+ )?
Epoch  ::= [A-Z] ( [A-Z] | [A-Z0-9._]* [._] [A-Z0-9._]* ) 
           | "$" [^ #x9#xA#xD]+

{.example ...} The following are examples of dates which match the Date production:

63 B.C.
21 JAN 1793
@#DJULIAN@ 29 MAY 1453

The first of these includes only a calendar year and epoch name. The following two both have a calendar day, calendar month and calendar year, and neither specifies an epoch name. Only the third date includes a calendar escape. {/}

{.note} The Month and Epoch productions are more complex than might at first seem necessary to ensure that no string can match both productions. This ensures a string containing just a day and a month, such as "1 JAN", cannot match the Date production. A future version of ELF might allow the year to be omitted in dates where it is unknown.

A conformant application serialising a date using this syntax should use a single space character (U+0020) wherever whitespace is permitted in the Date production.

{.note} [GEDCOM 5.5.1] under-specifies how whitespace is allowed in dates. The Date production is somewhat permissive in its treatment of whitespace, though does not allow it to be omitted entirely. The preceding recommendation ensures that conformant ELF applications will be maximally compliant with current GEDCOM application which typically use a single space character.

Calendar escapes {#cal-escapes}

The CaleEsc production encodes the calendar escape, which identifies the particular calendar being used in the date.

{.note} Syntactically, the calendar escape is an ELF escape, as defined §XX of [ELF Serialisation]. When such escapes occur in the payload of a DATE line, they are passed through to the data model unaltered.

{.ednote} Check the above is accurate once [ELF Serialisation] has been updated, and update the reference. Note that this means historical dates must appear on DATE lines.

The following calendar escapes are defined in {§calendars} of this standard.


@#DGREGORIAN@ The Gregorian calendar defined in {§gregorian} @#DJULIAN@ The Julian calendar defined in {§julian} @#DFRENCH R@ The French Republican calendar defined in {§french} @#DHEBREW@ The Hebrew calendar defined in {§hebrew}


{.note} The calendar escape for the French Republican calendar contains exactly one space character (U+0020). It must not be written with any alternative form of whitespace.

{.ednote} An earlier draft of this standard included @#FRENCHR@ (without a space) as an alias for @#DFRENCH R@, but we made no record of the use case that lead to its inclusion and have removed it again.

{.note} [GEDCOM 5.5.1] includes one further calendar escape, @#DROMAN@, which it reserved for future use, presumably for use with Roman Republican calendar. This standard does not reserve this calendar escape.

{.ednote ...} Should we standardise a basic Roman calendar using the calendar escape @#DROMAN@? The lengths of the months would be left unspecified, for the simple reason that they are not always known, though we might safely limit the calendar day to being a two digit number. It would probably use the twelve month names of the Gregorian calendar, including JUL and AUG, despite them being called Quintilis and Sextilis at the time. We would need a month name for Intercalaris, the intercalary month inserted in or after February. As INT risks conflicting with the date modifier for interpreted dates, MER might be best as Intercalaris was also called Mercedonius. We would also need month names for Intercalaris Prior and Intercalaris Posterior, the two exceptional inserted in 46 BC to realign the calendar with the equinoxes in preparation for the introduction of the Julian calendar. IC1 and IC2 would work. The A.D. and B.C. epoch names would be supported. A.U.C. (for ab urbe condita – since the founding of Rome) would need careful consideration.

The need for this calendar arises surprisingly often. Pompey, the famous Roman general, was born on 29 September 106 BC, as reckoned in the Roman calendar, but this date cannot be converted into the proleptic Julian or Gregorian calendar with certainty because the precise lengths of the years around this time are not known. A comprehensive discussion of this can be found in [Roman Dates]. Even though no genealogies back to Roman times can currently be proved, they are commonly encountered, and the fact that a genealogy cannot be proved does not necessarily mean it is not useful to record it, perhaps for the purpose of critical review. {/}

The @#DUNKNOWN@ calendar escape is permanently reserved. Third parties must not define calendars with this name and applications should not generate dates using this calendar escape, however conformant applications must not discard dates using this calendar escape. Conformant applications must not assume that two dates with a @#DUNKNOWN@ calendar escape, whether explicitly written or inferred as described below, are expressed in the same calendar.

{.example} The generic date syntax puts sufficient constraints on how calendars can be represented that applications may make certain assumption about dates written in unknown calendars, for example that @#DISLAMIC@ 1 RAJ 1420 is an earlier date than @#DISLAMIC@ 9 RAM 1422 if both are well-formed dates. This is because year numbers must increase with time. Conformant applications must not make similar assumptions for two dates with the @#DUNKNOWN@ calendar escape as they might not be represented using the same calendar.

{.note} It is the intention of FHISO to allow third parties to define their own calendar escapes in order to support additional calendars. However this version of ELF provides no means of avoiding conflict between separate third-party calendar escapes. This is particularly problematic when there a several variants of a calendar, and if different vendors choose to implement a different variant using the same calendar escape. FHISO intend to introduce a mechanism to avoid such conflicts in a future version of ELF. This is likely to work by assigning a term to each calendar, and a syntax for binding calendar escapes to term name IRIs.

{.ednote ...} This functionality was dropped from ELF 1.0 because the specification proved more complicated than expected. The intention is to add calendar bindings to the ELF schema, such as this:

1 SCHMA
2 PRFX elf https://terms.fhiso.org/elf/
2 IRI elf:JulianCalendar
3 DTYPE JULIAN

Because dates are likely to copied around, once in the data model the date needs to reference the calendar term name rather than the calendar escape. This means the serialisation layer needs to convert calendar escapes to term names, and vice versa. Finding a clean way of doing this proved problematic.

The TSC had been considering making calendars a specific sort of datatype, and then have the serialisation layer treat the calendar escape as way of tagging the datatype of the DATE tag's payload. However, as noted in an earlier editorial note, this caused problems with date periods and date ranges which used multiple calendars, and has been deferred to a future version of ELF.
{/}

This generic date syntax defines only some basic syntactic constraints on the representation of the calendar day, calendar month, calendar year and epoch name components. The party defining each calendar should define further constraints on these components to define what constitutes a well-formed date in that calendar. Where possible, the set of well-formed dates should be the same as the set of dates that actually existed.

{.example} The date "12 AUGUST 2000" is not a well-formed date in the Gregorian calendar, as defined in {§gregorian}, because the specified month, "AUGUST", is not one of the twelve allowed months names: it ought to have been written "12 AUG 2000".

{.example} The date "29 FEB 1973" is not a well-formed date in the Gregorian calendar, because February 1973 only had 28 days.

{.note} This standard does not prohibit calendars from defining dates which never occurred to be well-formed, though this is generally not recommended. It is allowed to accommodate calendars where the exact sequence of dates is either unknown or cannot be determined algorithmically.

{.example} Certain versions of the Islamic calendar define the start of each calendar month by when the new moon is actually observed. This results in unpredictable month lengths. If such a calendar were defined for use in ELF, it would likely regard days beyond the expected end of the month as well-formed dates to accommodate the possibility that bad weather had prevented the new moon from being observed.

All dates written using a calendar escape with which the application is not familiar are assumed to be well-formed. This includes all dates using the @#DUNKNOWN@ calendar escape.

Conformant applications must not generate dates which are known not to be well-formed. Applications encountering dates which are known not to be well-formed may delete the date or signal an error to the user. If an application cannot determine whether or not a date is well-formed, it must assume it is well-formed.

{.note} Other than for dates written in the Gregorian calendar, no part of this standard requires an application to be able to determine whether a date is well-formed. In some calendars, such as the Hebrew calendar defined in {§hebrew}, the rules for determining this are fairly complex. Applications are encouraged to implement these rules in full, but are not required to.

The calendar escape is optional in the generic date syntax. If the calendar escape is omitted and if the date if well-formed in the Gregorian calendar, then the date is treated as if it used the @#DGREGORIAN@ calendar escape. If the calendar escape is otherwise omitted, the date is treated as if it used the @#DUNKNOWN@ calendar escape.

{.note} [GEDCOM 5.5.1] simply says that dates written without a calendar escape default to the Gregorian calendar. ELF's approach is more nuanced. If a date explicitly uses the Gregorian calendar escape then an invalid date may be deleted; if it is written without a calendar escape then it must not be, assuming it conforms to the generic date syntax. ELF deviates from [GEDCOM 5.5.1] in this regard because many current applications fail to include a calendar escape when the Julian calendar is used. As a result, dates like "29 FEB 1700" can be found written without a calendar escape. This date did not exist in the Gregorian calendar and is not a well-formed date in that calendar, however this rule prevents applications from deleting it as invalid.

It is recommended that, where possible, dates should be entered in ELF datasets using the calendar in which they were written in the source.

{.example} A contemporary record of an event occurring in seventeenth century Massachusetts would almost certainly be recoded using the Julian calendar, as Massachusetts, like all the British colonies, did not adopt the Gregorian calendar until 1752. The date of this event should therefore be recorded in ELF using the Julian calendar and not converted to the Gregorian calendar.

Days

The Day production encodes the calendar day component of the date. It is an positive integer which calendars should use to count how many calendar days into the calendar month the specified calendar day is, with the first calendar day being "1". Leading zeros preceding a non-zero digit are permitted; conformant applications must attach no significance to them and may remove them.

{.ednote} An earlier draft of this standard allowed non-integer calendar day components, providing the first character was a decimal digit. The motivation for allowing this came from consideration of Roman day reckoning. In this system, days are reckoned backwards from three fixed points in each month, called the kalends, nones and ides. For example, the day described as "ante diem quintum kalendas Septembres" or "a.d. V Kal. Sept", meaning five days before the kalends of September (1 Sept), counting inclusively, is 28 August. If this were considered a separate calendar and non-integer calendar day components allowed, the day could be represented as "5K SEP". However on balance it was felt this was an unnecessary complication and the facility was removed. The framework is still flexible enough to handle Roman day reckoning by using separate calendar month components for the kalends, nones and ides, thus giving 28 August a representation similar to "5 KSEP".

{.ednote} Should there be a requirement that days numbers increase monotonically? Roman day reckoning, which counts days backwards, can still be supported if negative day numbers are allowed: e.g. "-5 KSEP". This has the advantage of allowing applications to sort calendar days without knowing the calendar, and all that would be required to sort days completely would be to know the order of month names and epoch names.

Months

The Month production encodes the calendar month component of the date. The set of permitted calendar month components in a given calendar is called the set of month names for the calendar. Month names should normally be abbreviated forms of their common names, and must be at least three characters long.

{.example} The French Republican calendar has twelve months named Vendémiaire, Brumaire, Frimaire, Nivôse, Pluviôse, Ventôse, Germinal, Floréal, Prairial, Messidor, Thermidor and Fructidor. In ELF, as described in {§french}, these are abbreviated VEND, BRUM, FRIM, NIVO, PLUV, VENT, GERM, FLOR, PRAI, MESS, THER and FRUC. In addition, each year had five or six consecutive intercalary days, or jours complémentaires, which were not part of any month. For the purposes of representing this calendar in ELF, the intercalary days are considered a thirteenth month, COMP.

{.note} Month names are always upper-case, a constraint guaranteed by the Month production. Lower-case, mixed-case, or non-ASCII month names must not be used.

{.note} Month names are required to be at least three characters long to avoid conflicting with epoch names.

The following words are reserved and must not be used as month names in any calendar: ABT, AFT, AND, BCE, BEF, BET, CAL, EST, EVERY, FOR, FROM, INT, POS, REP, TIME, UNCERT, UNK and ZONE.

{.note} Many of these words have specific meanings in the ELF date datatypes. The words EVERY, FOR, POS, REP, TIME, UNCERT, UNK and ZONE are reserved for possible future use because they describe concepts in [GEDCOM X Dates] or [ISO 8601-2] which are not currently in ELF. This does not necessarily mean FHISO will add such functionality to a future version of ELF.

Years

The Year production encodes the calendar year component of the date. The string matching this production shall be an integer, and may be followed by a solidus (U+002F) and another integer. Either integer may begin with a minus sign (U+002D).

{.note} Negative years and the year zero are supported by this generic date syntax, but not by any calendar defined in this standard. The intention is that they provide a way of representing a dates before the epoch of a particular calendar where no reverse epoch name has been defined. This is not expected to be common with real calendars.

{.ednote} One specific example where it might occur is if a future version of ELF defines a calendar to represent Julian days, for use in defining epochs. The epoch for the Julian day is in the Gregorian year 4714 BC, but some other calendars, such as the Byzantine calendar, have earlier epochs meaning their epoch occurs on a negative Julian day.

When the calendar year component contains no solidus and second integer value, the integer given is the logical year number. A conformant application must attach no significance to its particular lexical form, and may add or remove leading zeros preceding a non-zero digit.

A calendar year component with two integers is called a dual year, and is used when the historical and modern reckoning of years differ. The first integer in a dual year is the historical year; the second integer is the logical year which may be given in abbreviated form.

{.example} The principal use of dual years is to encode dates which were recorded using years beginning on 25 March, as was the practice in many parts of the world in mediæval and early modern times. Charles I's execution occurred on 30 January in the logical year 1649, but the historical year 1648. To avoid ambiguity, the year can be written 1648/1649. ELF supports this and allows the date to be recorded as “30 JAN 1648/1649”. In this case the logical year is not abbreviated.

The dual year syntax is only used when there are genuine differences in the conventional reckoning of years. Dual years should be used during periods when use of a historical year differing from the logical year was common, even if the source in question did not use the historical year. This is to remove any potential ambiguity.

{.example} The date of Charles I's execution should not be written "30 JAN 1649", even though this is technically correct as 1649 is the logical year, as reckoned with 1 January as the start of the new year. This is true even if the date was quoted from a modern book which wrote it using the logical year. The form "30 JAN 1648/1649" avoids any possible ambiguity over which new year was being used.

Dual years must not be used simply to record an error in the year as stated in a source.

{.example} If a parish register includes a baptism entry dated 12 Jan 1842, but the context and other circumstantial evidence makes it clear that the year was incorrectly written in the register and was in fact 1845, this must not be recorded in ELF as "12 JAN 1842/1845".

When serialising a date, a dual year must be used if the historical year and logical year differ, and must not be used if they are equal.

When serialising a dual year, applications may abbreviate the logical year to just the last two digits if the difference between the historical year and the logical year is less than 10 years, or may abbreviate it to just the last one digit if the difference between the historical year and the logical year is no more than 1 year. If the historical year and the logical year differ by 10 or more years, abbreviated form must not be used. When the logical year is abbreviated, any minus sign is dropped in the abbreviated form. Where possible, logical years should be abbreviated to two digits.

{.example} The date of Charles I's execution may be written in abbreviated form as "30 JAN 1648/49" or as "30 JAN 1648/9". The former is recommended as this is compatible with the [GEDCOM 5.5.1] standard, though both forms can be found in current GEDCOM files, as can the unabbreviated form.

{.example} Abbreviated form must not be used if dual years are being used in the Byzantine calendar to express years using standard Byzantine epoch and the Alexandrian epoch. This is because the historical year and logical year will differ by 16 or 17 years, depending on the specific date.

A specific calendar should normally place further restrictions on how dual years can be used, and any dates using dual years which fail to conform to any such calendar-specific restrictions are not well-formed dates.

{.note} This is necessary if applications are to have understanding of what the historical date means. A date of "@#DJULIAN@ 1740/1620" is perfectly well-defined insofar that it represents the Julian year A.D. 1620 and which for some reason was conventionally recorded as 1740 at the time, but without an understanding of why this was done, it is difficult for an application to know what to do with this information. In fact, the Julian calendar defined in {§julian} only allows dual years when the historical date and logical date differ by exactly one year. "@#DJULIAN@ 1740/1620" is not therefore a well-formed date.

Calendars must not place further restrictions on when the abbreviated format can be used.

{.note} This is because abbreviated form is considered part of the generic date syntax, and not specific to any calendar. It is intended that applications can parse dual years to extract the historical date and logical date even in unknown calendars.

{.note ...} [GEDCOM 5.5.1] includes support for dual years, though only with dates in the Gregorian calendar. This is unfortunate as dual years are principally used with dates in the Julian calendar, and the example given in [GEDCOM 5.5.1] is seemingly in the Julian calendar. Nevertheless, many applications do implement dual year support. In [GEDCOM 5.5.1], dual years must be written in abbreviated form with the logical year shortened to two years. ELF relaxes this restriction as dual years are commonly found unabbreviated or with a single-digit logical date.

Dual years in [GEDCOM 5.5.1] were probably only intended for years beginning on Lady Day, though the standard is not quite explicit in saying so; however the same facility readily accommodates years beginning on arbitrary days, and some current applications implement this. ELF generalises this further by allowing the historical year and logical year to differ arbitrarily in the generic date syntax, though of the calendars currently defined, only the Julian calendar allows dual years, and then only when the historical year and logical year differ by exactly one year. {/}

When parsing a dual year, a conformant application must treat the second integer as an abbreviated logical year if it could have been produced by the serialisation rule for abbreviated logical years, above, and otherwise must not.

{.example} The dual year "1616/8" is not in abbreviated form because there is no logical year number ending in 8 which is at most one year away from the historical year 1616. This is therefore considered an unabbreviated dual year, representing the historical year 1618 and the logical year 8. Very probably this will result in the date not being well-formed in the applicable calendar, in which case the application may reject it.

Conformant applications may alter the lexical form of a dual year in any way, providing the logical year and historical year it represents is unchanged, and must not attach any significance to its particular lexical form.

{.note} This allows applications to store the calendar year component of the date as two integers internally, and regenerate its lexical form upon serialisation.

The calendar may define further restrictions on the range of integers permitted as logical years. Any such restrictions apply regardless of whether dual years are in used.

Epochs

The Epoch production encodes the epoch name component of the date, which specifies both the particular epoch from which calendar years are numbered and the direction of the numbering. Each calendar specifies the set of permitted epoch names components for that calendar.

{.example} The Julian calendar defined in {§julian} includes the epoch names "B.C." and "A.D.", standing for Before Christ and Anno Domini, respectively. Both epoch names state that calendar years are counted from an epoch on the Julian date 1 January A.D. 1, nominally the date for the birth of Jesus Christ, but for "B.C." calendar years are counted backwards from that epoch, while for "A.D." calendar years are counted forwards.

{.note} The epoch name always comes after the calendar year component in the generic date syntax, despite certain epoch names being conventionally written before the logical year number in English. For example, it is normal to write "AD 1752" in English, but the ELF representation is "1752 A.D.". A conformant application must not accept "A.D. 1752" as a valid date, as it does not match the Date production.

The Epoch production requires epoch names either to be exactly two characters long, or to include at least one full stop (U+002E) or underscore (U+005F), or to begin with a dollar sign (U+0024).

{.note} This ensure that no month name is syntactically allowed as an epoch name, or vice versa. This ensures that a string like "1 JAN" cannot match the Date production.

Epoch names should normally be an acronym or initialism with full stops between letters. The two-letter form of epoch names, without a full stop, underscore or leading dollar, is deprecated. If an application encounters such a epoch name, it should convert it into an initialism by inserting a full stop between the two letters and another after the second letter.

{.note} Two-letter epoch names without full stops are allowed in ELF because "BC" is moderately frequently found in current GEDCOM files, even though [GEDCOM 5.5.1] requires it to be spelled "B.C.". This rule converts "BC" to "B.C.".

Epoch names beginning with a dollar sign are reserved for future use. Conformant applications must accept them, but calendars must not define any such epoch names.

{.ednote} This specification of epoch names is designed to be usable to implement regnal years. In such a scheme, the instant of accession of each monarch would be an epoch, and a forwards epoch name would be used to reference it, for example "25 OCT 3 HENRY_V". However to accommodate monarchs with a single name and no regnal number (e.g. Victoria or Tutankhamun), as well as to allow monarch names in foreign scripts, the dollar sign is reserved as a possible sigil for epoch names. This would allow "$昭和", for example, to represent the epoch at the start of the Shōwa period in Japan.

The words "TO" and "AT" are reserved and must not be used as epoch names in any calendar.

{.note} The word "AT" is included here for future compatibility; the word "TO" is reserved because it is used in date periods in {§DatePeriod}.

{.ednote ...} An earlier draft of this standard included the following paragraph:

If a calendar defines multiple epoch names for the same epoch, and which are either all forwards epoch names or all reverse epoch names, applications should treat those epoch names as interchangeable.

This was added to allow the Julian and Gregorian calendars to define "B.C.E." as an alias for "B.C.", however the current draft does not do that, and it seems an unnecessary complication. If the BCE form is added, this paragraph needs reinstating. {/}

A calendar may require an epoch name to be present, may allow it to be omitted, and may define no epoch names thereby requiring it to be omitted. In calendars where the epoch name is optional, the calendar should define an explicit epoch name that is equivalent to an omitted epoch name. This epoch name is called the default epoch name for the calendar.

{.example} The Julian calendar defined in {§julian} provides the "A.D." epoch name to serve as its default epoch name.

{.ednote} The French Republican calendar, as currently defined in {§french}, does not properly define an epoch name, or equivalently leaves its sole epoch name anonymous. This is allowed, but is not recommended. If the French Republican epoch name remains anonymous, the above text should perhaps be changed.

Date modifiers {#modifiers}

The generic date syntax described in {§generic} provides a means of specifying a particular date, but does not provide a way of describing dates which are approxiately stated, nor of giving natural language descriptions, nor of stating a imprecise date via its precision range. ELF provides several extensions to the generic date syntax to provide this functionality. Syntactically, these are expressed by prefixing the generic date syntax with a token called date modifier.

{.example} An "ABT" date modifier is introduced in {§approx}. It is placed before the date, as in "ABT JAN 1901".

{.ednote} This draft, like [GECOM 5.5.1], does not allow more than one of these facilities to be used on the same date. It seems potentially useful to remove this restriction, for example by allowing "FROM ABT 1798 TO 2 JAN 1842". There would be no ambiguity in the grammar with such constructs and their meaning would generally be well defined. Backwards compatibility could be handled by saying that applications should not produce such constructs but must be able to read them. This would also allow constructs such as "EST ABT 1881", which the GEDCOM grammar does not allow but which appears as an example in [GEDCOM 5.5.1], albeit as an example of an unnecessary circumlocution. Should the initial version of ELF support such things?

Approximated dates {#approx}

A date may be preceded by one of the tokens ABT, CAL or EST, as shown in the following DateApprox production:

DateApprox  ::= ( "ABT" | "CAL" | "EST" ) S Date

The ABT token indicates that the date is approximately stated, and that its precision is lower than would have been the case without the ABT token. It is typically used when there is evidence that the date given is roughly correct, and should not be used when the date is estimated using statistical likelihoods or cultural norms.

{.example} If it is known that the first and third children in a family were born in 1897 and 1900, respectively, the second child's birth may be recorded as "ABT 1899" as the date of birth is fairly well bounded, even if twins were involved.

{.note} The ABT token is currently by far the most commonly used of the three approximated tokens.

The EST token also indicates that date is approximately stated, and is an estimate perhaps based only on statistical likelihoods or cultural norms. An application may assume a date written with the EST token has lower precision than one with the ABT token.

{.example} If it is known that the first three children in a family of four were born between 1897 and 1900, a researcher might conclude that it is probable that next child was born shortly afterwards and estimate the date of birth as "EST 1903". The ABT token should not be used in this case, unless there is additional evidence that the fourth child was not very much younger.

{.example} It can sometimes be useful to provide a crude estimate of an individual's date of birth. If a baptism register records a baptism in 1692 and gives the father's name, a researcher might wish to record an estimate of the father's date of birth, perhaps to help disambiguate him from other people of the same name. If such an estimate is to recorded, it should use the EST token, for example, "EST 1660".

The CAL token is used to record a date that is not directly stated in a source, but instead has been calculated from other known values, typically another date and the duration separating the two dates. An application may assume a date written with the CAL token has a similar or higher precision than one written with the ABT token.

{.example} If a newspaper reports on a golden wedding celebration in the summer of 1948, this is evidence that the wedding happened 50 years previously, and the date of the marriage may be entered as "CAL 1898".

When a date is calculated from another date and an imprecise duration separating them, use of the ABT token is recommended instead of the CAL token.

{.example} Like many censuses, the 1880 census of Iceland records individuals' ages on their most recent birthday. A person recorded as 72 could have been born in 1807 or 1808. Because the age has been rounded down and is therefore somewhat imprecise, it is recommended that the date of birth be recorded as "ABT 1808" rather than "CAL 1808".

{.note} A CAL token might sometimes be an indicator that the date is of lower reliability than one without this token, everything else being equal, due to it not being directly stated.

{.ednote} Should FHISO deprecate the CAL token and allow uses of it to be replaced with ABT? It's not clear it adds anything useful.

Date phrases and interpreted dates {#date-phrases}

ELF allows dates to be recorded in natural language. The natural language description of a date is called a date phrases and is written between parentheses (U+0028 and U+0029).

DateInterp  ::= ( "INT" S Date S )? "(" DatePhrase ")"
DatePhrase  ::= [^#xA#xD()]*

The date phrase is an arbitrary language-tagged string except that it must not contain parentheses (U+0028 or U+0029), line feeds (U+000A) or carriage returns (U+000D).

{.note} [GEDCOM 5.5.1] only requires that date phrases be "enclosed in matching parentheses", which could be interpreted as requiring that any parentheses in the date phrase be balanced. ELF goes further and prohibits parentheses from occurring in date phrases at all.

{.ednote} The prohibition on these characters is not strictly needed, at least while the grammar does not allow date phrases within date ranges or date periods, though that is a possible extension which might be made in a future version of ELF. A data model layer escape mechanism along the lines of the %{} syntax used in FHISO's draft Creator's Name microformat could allow arbitrary characters to appear in it.

A date phrase should be a fragment of text quoted from a source, possibly in translation, and should only normally be used if the text cannot readily be converted into a date, or where that conversion would lose useful information. Date phrases should not be used for arbitrary annotations or commentary, and must not be used in a negative sense to say the event did not occur.

{.example} If a source describes an event as happening on "The Feast of St John" in a particular year, this might be recorded in a date phrase if there was insufficient context to establish whether it referred to the Feast of Nativity of St John the Baptist on 24 June, or the Feast of St John the Evangelist on 27 December.

{.note} Historically, some applications have used a date phrase like "Not married" to indicate a marriage did not occur. Such usage is prohibited in ELF and was not allowed in [GEDCOM 5.5.1]. Conformant applications are permitted to assume a date phrase describes an actual date, even if the date phrase cannot be interpreted. Fortunately, uses like "Not married" is largely confined to [GEDCOM 5.3] and earlier which omitted the parentheses around date phrases. Date phrases written without parentheses will not normally match the generic date syntax, and will therefore not be parsed as dates in ELF.

The date phrase may be accompanied by a date in the generic date syntax preceded by the INT token. The combination is called an interpreted date and its date should be an interpretation of the associated date phrase in the context of source containing it.

{.example} "INT @#DJULIAN@ 18 JUNE 1502 (Saturday before the Feast of the Nativity of St John the Baptist)" is a valid interpreted date. In this case the date phrase does not mention the year: perhaps it was inferred from its context or was stated elsewhere in the document.

{.note} An interpreted date might sometimes be of lower reliability than if the date were written without a date phrase, as the presence of a data phrase typically indicates some subtlety in interpreting the date.

Being a language-tagged string, a date phrase has a language tag associated with it. This language tag is not embedded in the date but is provided externally by the serialisation format.

{.note ...} In [ELF Serialisation], the language tag is provided by a LANG tag, which may be a substructure of the DATE tag, failing which it may be a substructure of one of the superstructure of the DATE tag, failing which it may be in the ELF header. For example,

2 DATE INT 25 JAN 1840 (L'an mil huit cent quarante, le vingt-cinquième jour du mois de janvier)
3 LANG French

{/}

{.ednote} Review the previous note once [ELF Serialisation] has been developed further.

Date ranges {#date-ranges}

A date range is a time interval used to record an unknown date which can be placed within certain bounds. A date range may be bounded from below, from above, or both. It matches the DateRange production:

DateRange   ::= ( "BEF" | "AFT" ) S Date | "BET" S Date S "AND" S Date

A date range beginning with a BEF token represents an unknown instant on or before the specified date. A date range beginning with a AFT token represents an unknown instant on or after the specified date.

It is recommended that the BEF and AFT forms of date ranges are only used if it is believed the specified date is probably within a few years of the unknown date being represented.

{.example} If an individual was elected mayor of a city in 1745 and the cultural norms of the time mean the person was likely a middle-aged adult, their date of birth should not be recorded as "BEF 1745". Even though this would literally be true, insofar as the person was indeed born before 1745, and this representation is not prohibited, it is not recommended. The person could well have been born in the previous century which would be stretching the definition of "a few years". If there is need to give a date of birth, a crude estimate given with the EST token might be preferable.

A date range using the BETAND construct represents an unknown instant on or after the first specified date, and on or before the second specified date. This is a way of specifying an unknown date in terms of its precision range. In this form of date range, the first specified date must not be later than the second specified date.

{.example} If someone's age is recorded as 26 on a census conducted on 3 April 1881, their date of birth could be recorded as "BET 4 APR 1854 AND 3 APR 1855". However the usual practice in GEDCOM is to record this as "ABT 1855".

{.ednote} Should ELF encourage users to prefer the date range form in the previous example?

Date ranges using the BETAND format contain two dates and therefore potentially two calendar escapes. The second date does not inherit the calendar escape from the first date, and it must be respecified when required. It is recommended that, where possible, the same calendar escape be used on both dates.

{.note} Although [GEDCOM 5.5.1] supports date ranges where the two dates use different calendar escapes, not all current implementations support them. They are best avoided where possible.

The elf:DateValue datatype {#DateValue}

The elf:DateValue datatype is used to record historical dates. It allows dates to be represented using the generic date syntax or any of the modifier forms given in {§modifiers}. It may also be a date period as defined in {§DatePeriod}. Its lexical space is the set of strings which match the following DateValue production:

DateValue ::= Date | DateApprox | DateInterp | DateRange | DatePeriod

{.ednote} In the many places in ELF, it is wrong to use a date period where an elf:DateValue is required. This is not the case with the other options allowed in the DateValue production. The difference is that a date period represents a state which persisted for an extended time interval, while the others represent a single instant (or at least an event of short duration that is conveniently approximated to an instant). FHISO are considering removing DatePeriod from the DateValue production, above, and allowing it as an explicit option in certain contexts. However this introduces the same complications as introducing separate elf:AgeWord datatype, and which are discussed in an editorial note at the end of {§age}.

The elf:DateValue datatype is a language-tagged datatype, however the associated language tag is ignored unless the date contains a date phrase. Conformant applications are not required to preserve the language tag of dates without a date phrase, and may replace it with an arbitrary language tag.

{.note} This allows applications to store the language tag as part of the date phrase, and allows them to serialise an elf:DateValue that does not contain a date phrase using an arbitrary language tag. In particular, this means a date read from an ELF file in one language may be serialised as part of an ELF file in another language without explicitly annotating it with a LANG substructure, unless the date contains a date phrase.

Formally, the elf:DateValue datatype is a structured language-tagged datatype which has the following properties:

: Datatype definition


Name https://terms.fhiso.org/elf/DateValue Type http://www.w3.org/2000/01/rdf-schema#Datatype Pattern See below Supertype http://www.w3.org/1999/02/22-rdf-syntax-ns#langString Abstract false


The pattern for this datatype is as follows:

\begingroup\footnotesize

{#include DateValue.md}

\endgroup

{.note} The grammar productions given in this standard provide a more useful description of datatype for a human reader. The pattern above is used during datatype correction in certain optional contexts. It was generated automatically from the grammar productions, and any discrepancy between the grammar and this pattern is unintentional.

The elf:DatePeriod datatype {#DatePeriod}

A date period is a time interval used to record a state of being that persisted throughout the stated time interval which represents an extended period of history; they are also used to record the period of coverage of a source. They are represented in ELF using the elf:DatePeriod datatype whose lexical space is set set of strings which match the following DatePeriod production:

DatePeriod  ::= "FROM" S Date ( S "TO" S Date )? | "TO" S Date

Use of a date period does not necessarily mean the state was believed to have begun on the specified "from" date, nor that it ended on the specified "to" date, though where possible this is recommended. Either the start date or the end date may be omitted from the date period, in which case the missing date is interpreted as an unknown date.

{.example} Abraham Lincoln was President of the United States from 1861 to 1865. This could recorded using the date period "FROM 1861 TO 1865", or more precisely as "FROM 4 MAR 1861 TO 15 APR 1865". If it were unknown when Lincoln's presidency began, it could be written "TO 15 APR 1865". If it was also known that Lincoln was president at the time of the Battle of Gettysburg in July 1863, the date period of his presidency could be written "FROM JUL 1863 TO 15 APR 1865". This does not imply Lincoln became president in 1863, only that he was president for the whole of the period from 1863 to 1865.

{.ednote} Neither GEDCOM nor this draft of ELF has a way of saying whether the date period is believed to be the complete period during which the state persisted. This is perhaps a shortcoming that a future version of ELF should resolve. The obvious solution is to say that a date period always represents the full period, and allow the various date modifiers from {§modifiers} to be used in the end instants of the date period. In the previous example where Lincoln was known to have been president for the Battle of Gettysburg, and to have died in office on 15 April 1865, this could be written "FROM BEF JUL 1863 TO 15 APR 1865". This could be done without allowing arbitrary nested date modifiers.

{.note} The difference between a date range and a date period is sometimes misunderstood, and applications should be alert to the wrong representation being used. A date range is used to record an instant, or an event of short duration, when its date is not precisely known, whereas a date period is used to record a state that persisted throughout the specified time interval.

{.example ...} Civil birth, marriage and death registrations in England and Wales are indexed in quarterly volumes which do not list the precise date of the event. The coverage of this source is represented in ELF using a date period:

0 @S1@ SOUR
1 TITL GRO Marriages Index 
1 DATA
2 EVEN MARR 
3 DATE FROM 1 JAN 1898 TO 31 MAR 1898

However the event itself must not be recorded using a date period because the marriage did not take place over an extended period: it happened on one particular date which is not precisely known. For this purpose a date range should be used.

0 FAM
1 MARR
2 DATE BET 1 JAN 1898 AND 31 MAR 1898

A less precise alternative would be to just give a year. {/}

: Datatype definition


Name https://terms.fhiso.org/elf/DatePeriod Type http://www.w3.org/2000/01/rdf-schema#Datatype Pattern See below Supertype http://www.w3.org/1999/02/22-rdf-syntax-ns#langString Abstract false


The pattern for this datatype is as follows:

\begingroup\footnotesize

{#include DatePeriod.md}

\endgroup

Calendar definitions {#calendars}

This section defines four standard calendars for use with ELF: the Gregorian, Julian, Hebrew and French Republican calendars. Conformant applications are required to support all four.

{.ednote} There is no technical need for applications to support anything other than the Gregorian calendar. Would it be better to make only this one required and the others recommended? Ideally, applications would do more than simply check dates are syntactically valid and would offer conversions between supported calendars, date arithmetic, the ability to look up the day of the week and the dates of important feast days. Few applications currently offer this. Would reducing the number of required calendars make it easier for vendors to provide such functionality?

{.ednote} The definitions given in this section only specify those details which are necessary to determine whether a date is well-formed in the given calendar. Should these descriptions be expanded to give more details to help applications convert between calendars?

The Gregorian calendar {#gregorian}

{.note} The Gregorian calendar is the name given to the now ubiquitous calendar introduced by Pope Gregory XIII in 1582 to correct the Julian calendar which was slowly drifting relative to the seasons. It is represented by the @#DGREGORIAN@ calendar escape, and is also ELF's default calendar, used whenever a date has no calendar escape and is well-formed in the Gregorian calendar.

The Gregorian calendar has an epoch at the start of the calendar day 1 January 1 AD, as expressed in the Gregorian calendar, and two epoch names relative to that epoch: a reverse epoch name "B.C.", and a forwards epoch name "A.D.". The latter is the default epoch name for the calendar, and its name should be omitted.

{.example} The date "24 DEC 2018 A.D." is equivalent to "24 DEC 2018". The latter is recommended for compatibility with [GEDCOM 5.5.1] which does not support the "A.D." epoch name.

{.ednote} Do we want "B.C.E." and "C.E." (standing for Before Common Era and Common Era, respectively) as aliases? There is no technical justification for adding them. Supporting "BCE" as an alias for "B.C.E." is problematic as it is three letters long, meaning matches the Month production rather than Epoch production. However "BCE" has been reserved so it cannot be used as a month name.

Regardless of epoch name, the logical year shall be an integer greater than 0.

{.note} This prohibits negative or zero year numbers as they are not needed. The year before "1 A.D." is "1 B.C.".

{.note} [GEDCOM 5.5.1] says that logical years must be 3 or 4 digits long, and presumably requires dates in the first century to be zero padded. This standard has no such requirement, and many current applications do not enforce this requirement.

Dual years must not be used in the Gregorian calendar.

{.note} This is a deviation from [GEDCOM 5.5.1] which allows dual years only on Gregorian dates. In this standard, a date with a dual year is not well-formed in the Gregorian calendar. This means a date using a dual year and no explicit calendar escape will be assigned the @#DUNKNOWN@ calendar escape.

{.ednote} In practice there is a very strong likelihood that the Julian calendar is intended. This draft could have altered the default calendar rules in {§cal-escapes} so that dates using dual years and no explicit calendar escape were automatically labelled @#DJULIAN@. The reason this was not done is that an ELF file containing such dates is likely to have many other miscalendared dates but which are well-formed in the Gregorian calendar and so go undetected. Flagging those with dual years with @#DUNKNOWN@ will hopefully bring this to the researcher's attention, with the result that all the miscalendared dates are fixed.

Every calendar year in the Gregorian calendar consists of 12 calendar months. Their month names are given in the table below in order of their occurrence in the calendar year. The table also gives the usual form of their name in English, and the number of calendar days in each month. The calendar days in each calendar month are numbered sequentially starting with 1.


JAN January 31 days FEB February 28 or 29 days — see below MAR March 31 days APR April 30 days MAY May 31 days JUN June 30 days JUL July 31 days AUG August 31 days SEP September 30 days OCT October 31 days NOV November 30 days DEC December 31 days


The number of calendar days in February varies depending on the logical year. The rules for determining this number in years with the "A.D." epoch name are as follows:

  • If the logical year number is exactly divisible by 400, then February has 29 days.
  • Otherwise, if the logical year number is exactly divisible by 100, then February has 28 days.
  • Otherwise, if the logical year number is exactly divisible by 4, then February has 29 days.
  • Otherwise, February has 28 days.

In the Gregorian calendar, a calendar year in which February has 29 calendar days is called a leap year, while a calendar year in which February has only 28 calendar days is called a non-leap year.

{.note} Other calendars defined the phrases leap year and non-leap year differently.

For years with the "B.C." epoch name, the logical year number is subtracted from one to get zero or a negative number, which is then used in place of the logical year in the preceding rules.

{.example} The year 5 BC was a leap year in the proleptic Gregorian calendar, meaning February had 29 calendar days. This is because subtracting 5 from 1 gives −4 which is exactly divisible by 4.

{.note} Although the Gregorian calendar was first introduced in 1582, ELF allows its use proleptically, including for dates BC. This is a partial departure from [GEDCOM 5.5.1] which only allows incomplete dates referencing only a calendar year to use the "B.C." epoch name.

A date which uses a calendar day number which is greater than the number of calendar days in the specified year and month is not a well-formed date.

{.example} The date "29 FEB 2018" is not well-formed in the Gregorian calendar because February only had 28 days in 2018 due to 2018 not being exactly divisible by 4.

The elf:DateExact datatype {#DateExact}

The elf:DateExact datatype is used to record the creation or modification date of various objects in the ELF data model using the Gregorian calendar. The lexical space of this datatype is the set of strings which match the following GregDate production and which are additionally well-formed dates in the Gregorian calendar defined in {§gregorian}.

GregDate  ::= GregDay S GregMonth S GregYear

GregDay   ::= [0-9] [0-9]?
GregMonth ::= "JAN" | "FEB" | "MAR" | "ARP" | "MAY" | "JUN" | "JUL"
              | "AUG" | "SEP" | "OCT" | "NOV" | "DEC"
GregYear  ::= [0-9] [0-9] [0-9] [0-9]

{.example} The string "24 DEC 2018" matches the GregDate production and is a well-formed date in the Gregorian calendar. It is therefore in the lexical space of elf:DateExact. "54 NOV 2018" is not in the lexical space of elf:DateExact despite matching the GregDate production because it is not a well-formed date in the Gregorian calendar due to November only having 30 days.

{.note ...} This datatype is more restricted than a Gregorian date written in the generic date syntax due to the following additional constraints:

  • There must not be a calendar escape.
  • There calendar day and calendar month must be present.
  • The calendar year component must be four digits long.
  • The epoch name must not be present.
  • The date modifiers defined in {§modifiers} must not be used.
  • It must not be a date period. {/}

Formally, the elf:DateExact datatype is a structured non-language-tagged datatype which has the following properties:

: Datatype definition


Name https://terms.fhiso.org/elf/DateExact Type http://www.w3.org/2000/01/rdf-schema#Datatype Pattern [0-9]{1,2}[ \t\r\n](JAN|FEB|MAR|ARP|MAY|JUN|JUL|AUG|SEP|OCT |NOV|DEC)[ \t\r\n][0-9]{4} Supertype No non-trivial supertypes Abstract false


{.note} The pattern in the table above has been split on to three lines for convenience of presentation; it is, however, really one pattern and contains no whitespace or line breaks. Any functional difference between the GregDate production and the pattern specified above is unintentional.

{.note} The elf:DateExact datatype is not specified as a subtype of the elf:DateValue datatype because the latter is a language-tagged datatype.

The Julian calendar {#julian}

{.note} The Julian calendar is the name given to the calendar introduced by Julius Caesar in 45 BC and subsequently amended by Augustus in about 8 BC to correct an error in the application of its leap year rule during its first three decades. The Augustan form of the Julian calendar is represented by the @#DJULIAN@ calendar escape. Because the Gregorian calendar derived from the Julian calendar, many of the details here are the same as those in {§gregorian}.

The Julian calendar has an epoch at the start of the calendar day 1 January 1 AD, as expressed in the Julian calendar, or 30 December 1 BC, as expressed in the proleptic Gregorian calendar. The Julian calendar has the same two epoch names, "B.C." and "A.D.", as the Gregorian calendar, but defined relative to the Julian epoch instead of the Gregorian epoch. The "A.D." epoch name is the default epoch name for the calendar, and its name should be omitted.

As for the Gregorian calendar, regardless of epoch name, the logical year shall be an integer greater than 0.

Dual years are only allowed when the logical year and the historical year differ by exactly one year. They are used to represent the differing conventions for the first day of the year.

{.note} This is a deviation from [GEDCOM 5.5.1] where dual years are not permitted with the Julian calendar.

{.note} Logical years always begin on 1 January, but historical years can begin on an arbitrary day of the year which can result in the historical year being ahead or behind the logical year. Because the use of dual years is not restricted to just years beginning on Lady Day, it is not in general possible to tell simply from the fact that a dual year was used what convention for the start of the year. This can usually be inferred from context, and the logical year allows an application to process the date without knowing this.

{.example ...} In France, the year was reckoned to begin on Easter Sunday until 1564 (or 1565 — the change did not happen uniformly). An application cannot tell when reading the date "10 FEB 1521/22" whether it has been written using the French convention of starting the year on Easter Sunday or the English convention of starting the year on Lady Day, but it is not normally necessary to know this as this date refers to the same calendar day regardless of when the historical year began.

This is an example where the historical year alone could be ambiguous due to Easter being a movable feast. In France, there were two days which would have been written "1er jour d'avril, 1522", as Easter was early in the logical year 1522 and late in 1523. For both days described as 1 April 1522, the historical year is 1522, but only for the first is the logical year also 1522: for the second, it is 1523. This means the first date must be encoded in ELF as "1 APR 1522", while the second should be encoded as "1 APR 1522/23" (though it may alternatively be encoded as "1 APR 1523", without using a dual year). {/}

Every calendar year in the Julian calendar, as reckoned by logical year, consists of the same 12 calendar months as the Gregorian calendar. Their month names, the usual English form of their name, and the number of calendar days in each calendar month is as given for the Gregorian calendar in the table in {§gregorian}. The only difference between the calendars is the rule for determining the number of calendar days in February.

In the Julian calendar, the rules for determining the number of calendar days in February are as follows:

  • If the logical year number is exactly divisible by 4, then February has 29 days.
  • Otherwise, February has 28 days.

The terms leap year and non-leap year are defined as for the Gregorian calendar.

{.example} The year 1900 was a leap year in the Gregorian calendar, as 1900 is divisible by 100, but it was not a leap year in the Julian calendar. This means "29 FEB 1900" is a well-formed date in the Gregorian calendar, but not in the Julian calendar.

For years with the "B.C." epoch name, the logical year number is subtracted from one to get zero or a negative number, which is then used in place of the logical year in the preceding rules.

{.note} The year 5 BC was a leap year according to ELF's definition of the Julian calendar, because subtracting 5 from 1 gives −4 which is divisible by 4. In fact, the year 5 BC was almost certainly not a leap year as the Augustan reform was still taking effect at this point. There is disagreement between scholars on the exact details of the reform, but is generally accepted that by AD 8 the rule for leap years given above was being applied correctly. Any use of the Julian calendar (in its final Augustan form) before about AD 5 should therefore be regarded as proleptic.

A date which uses a calendar day number which is greater than the number of calendar days in the specified year and month is not a well-formed date.

The French Republican calendar {#french}

{.note} The French Republican calendar or French Revolutionary calendar are the names given to the new calendar adopted in 1794 by the French National Convention. It is represented in ELF by the @#DFRENCH R@ calendar escape.

The French Republican calendar has an epoch at the start of the Gregorian calendar day 22 September 1792, the date the First French Republic was founded. This date is identified as 1 Vendémiaire I in the new calendar.

{.note} It is common to write French Republican calendar years using Roman numerals.

The calendar has a single anonymous forwards epoch name. It does not provide a backwards epoch name for referring to dates before the founding of the First Republic, nor are zero or negative logical years permitted.

{.ednote} If it proves undesirable to have an anonymous epoch name, the usual phrases are "l'époque républicaine" and "l'ère de la republique". An epoch name of "E.R." would therefore seem appropriate.

The logical year shall be an integer greater than 0 and should be no greater than 18. Applications may consider any date with a logical year greater than 18 to be not well-formed.

{.note} The placement of leap years in the French Revolutionary calendar was never defined satisfactorily due to an ambiguity in the legislation that had not been resolved when the calendar was abolished. The calendar was in actual use during years II to XIV, and the placement of leap years in the period between years I and XVIII is unambiguous. The calendar was used again very briefly during the Paris Commune in May 1871 (Floréal and Prairial LXXIX), and applications are encouraged to support this, but ELF does not require this as the number of sources of genealogical relevance from the Paris Commune is likely to be small.

Dual years must not be used in the French Republican calendar.

Every calendar year in the French Republican calendar consists of 12 calendar months, which are followed by 5 or 6 intercalary days or jours complémentaires which ELF treats as a thirteenth month. Their month names are given in the table below in order of their occurrence in the calendar year. The table also gives the usual form of their name in French, and the number of calendar days in each month. The calendar days in each calendar month are numbered sequentially starting with 1.


VEND Vendémiaire 30 days BRUM Brumaire 30 days FRIM Frimaire 30 days NIVO Nivôse 30 days PLUV Pluviôse 30 days VENT Ventôse 30 days GERM Germinal 30 days FLOR Floréal 30 days PRAI Prairial 30 days MESS Messidor 30 days THER Thermidor (or Fervidor) 30 days FRUC Fructidor 30 days COMP Jours complémentaires 5 or 6 days — see below


{.note} The month of Thermidor was also called Fervidor, however ELF uses the month name THER regardless of which form the source uses. FERV must not be used.

In the French Republican calendar, a calendar year with 6 jours complémentaires is called a leap year, while a calendar year with only 5 jours complémentaires is called a non-leap year. If the logical year number is 3, 7, 11 or 15, the calendar year was a leap year; all other years with logical year numbers between 1 and 18, inclusive, were non-leap years.

{.note} This standard does not specify when leap years occurred after year XVIII because the legislation was ambiguous. The legislation required leap years to be arranged such that the autumnal equinox would fall on the first day of the year, 1 Vendémiaire, as observed from the Paris Observatory. Under this rule, the leap year after year XV would have been five years later, in year XX. However the legislation also said that leap years were every four years, in which case the next leap year would have been XIX. A solution was proposed by Charles-Gilbert Romme, one of the creators of the calendar, which would have placed a leap year in year XX, and then one every four years with similar rules as the Gregorian calendar applying every century. No decision had been made when Napoléon abolished the calendar, so no definitive statement can be made on which subsequent years were leap years, and this standard does not require any particular interpretation. Applications may do any of the above and remain conformant; they may also reject anything after year XVII as not being well-formed.

A date which uses a calendar day number which is greater than the number of calendar days in the specified year and month is not a well-formed date. This provision applies to jours complémentaires too.

The Hebrew calendar {#hebrew}

{.note} The Hebrew calendar is the name given to the calendar used by Jewish peoples around the world which developed into its current form in the early ninth century. It is represented in ELF by the @#DHEBREW@ calendar escape.

The Hebrew calendar has an epoch at the start of the calendar day referred to as 7 September 3761 BC in the proleptic Gregorian calendar. This date is identified as 1 Tishrei AM 1 in the Hebrew calendar.

{.note} The date of this epoch is more commonly quoted as 7 October 3761 BC, which is its date in the proleptic Julian calendar. This is the date of the start of the Hebrew calendar year traditionally believed to contain the Creation.

{.ednote} A day in the Hebrew calendar is normally reckoned to begin at sunset, which means the epoch should properly be at sunset on 6 September 3761 BC in the proleptic Gregorian calendar, or 6 October 3761 BC in the proleptic Julian calendar. However, as noted in {§time-concepts}, the definition of a calendar day given in [ISO 8601] and adopted here says a calendar day starts at midnight.

The calendar has a single forwards epoch name, "A.M.", which is the default epoch name for the calendar. It should be omitted when serialising dates in ELF. The calendar does not provide a backwards epoch name for referring to dates before the Hebrew epoch, nor are zero or negative logical years permitted.

{.note} [GEDCOM 5.5.1] does allow the "B.C." epoch name to be used with the Hebrew calendar, but it seems almost certain this is an error in that standard. ELF does not permit "B.C." to be used with this calendar, and any Hebrew dates using of it will not be well-formed date.

It is not recommended for dual years to be used with the Hebrew calendar.

{.note} This is not an outright prohibition on dual years, and alternative calculations of the year of Creation were used to number calendar years in early times. These could be used as historical years, typically differing from the logical year by no more than three years.

Every calendar year in the Hebrew calendar consists of either 12 or 13 calendar months. Their month names are given in the table below in order of their occurrence in the calendar year. The table also gives the usual form of their name in English and Hebrew, and the number of calendar days in each month. The calendar days in each calendar month are numbered sequentially starting with 1.


TSH Tishrei תשרי 30 days CSH Cheshvan חשוון 29 days, or sometimes 30 days — see below KSL Kislev כסלו 30 days, or sometimes 29 days — see below TVT Tevet טבת 29 days SHV Shevat שבט 30 days ADR Adar I אדר א 30 days, if present — see below ADS Adar II אדר ב 29 days NSN Nisan ניסן 30 days IYR Iyar אייר 29 days SVN Sivan סיוון 30 days TMZ Tammuz תמוז 29 days AAV Av אב 30 days ELL Elul אלול 29 days


{.note} The English names given above are transliterations of the Hebrew, and many variants of the transliterations can be found. Cheshvan is a shortened form of Marcheshvan (מרחשון) produced by dropping the first syllable. Adar I and Adar II are otherwise known as Adar Rishon (אדר ראשון) and Adar Sheni (אדר שני), respectively, using the Hebrew words for first and second. Adar I and II are also sometimes called Adar Aleph and Adar Bet (אדר א and אדר ב) after the first and second letters in the Hebrew alphabet.

{.note} For religious purposes, the calendar year is sometimes said to begin with the calendar month of Nisan. ELF uses the usual civil definition of the year beginning with the calendar month of Tishrei.

Calendar years containing 13 calendar months are called leap years, and fall in a 19 year cycle called the Metonic cycle. Calendar years which are not leap years are called non-leap years. To determine whether a year is a leap year, the logical year number is divided by 19 and the remainder taken to find the year's place in the Metonic cycle. If the remainder is 0, 3, 6, 8, 11, 14 or 17 then the year is a leap year; otherwise it is a non-leap year.

In a non-leap year, the calendar month of Adar I is omitted and Adar II is known simply as Adar (אדר). Nevertheless, the month name used for Adar in a non-leap year is "ADR", the same month name as Adar I in a leap year. "ADS" is not well-formed when used as a month name in a non-leap year, and it is recommended that applications rewrite it to "ADR".

{.note} The intended handling of Adar in non-leap years is very poorly unspecified in [GEDCOM 5.5.1]. As Adar I is the extra month inserted into the calendar, it would make sense if Adar were represented as "ADS" in non-leap years, but in practice this seems not to be what current vendors do. Possibly because [GEDCOM 5.5.1] describes "ADR" as standing for Adar rather than Adar I or Adar Rishon, vendors have used "ADR" to represent Adar in non-leap years and Adar I in leap years. ELF standardises that behaviour.

The rule for deciding when Cheshvan is extended in length to have 30 calendar days, and when Kislev is reduced in length to have 29 calendar days, are somewhat complex and based on when in the week the Tishrei molad falls.

{.note} Although an application needs to know how many calendar days are in each calendar month in order to determine whether a date is a well-formed date, an applications can be conformant without being able to determine whether a date is well-formed. This would allow a conformant application to support the Hebrew calendar without full knowledge of how long each calendar month is. An application which cannot determine how many calendar days are in Cheshvan and Kislev in a particular year, must consider "30 CSH" and "30 KSL" to be well-formed regardless of year.

The molad is the calculated instant of the new moon, as seen from Jerusalem. The first molad occurred 5 hours and 204 halakim into Monday, 1 Tishrei AM 1. A helek (plural: halakim) is a traditional Hebrew subdivision of the hour equal to 3⅓ seconds: there are 1080 halakim in an hour. The week used by the Hebrew calendar is the standard seven day week.

{.note} This time is stated relative to the traditional Hebrew start of the day, which was at sunset the previous evening; it also uses unequal hours so that the time between sunset and sunrise is always 12 hours, regardless of the time of year. Applications do not need to be aware of this detail as all times are expressed using Hebrew time reckoning in this calculation. All they need to know is that there are 1080 halakim (also spelt "chalakim") in an hour, 24 hours in a calendar day, and 7 calendar days in a week.

In the Hebrew calendar calculations in this standard, instants within a week are expressed with as three integers separated with colons. These being the day number, the number of hours into the day, and the number of halakim into the hour. Days of the week are numbered with Saturday being 0.

{.example} The instant of the first molad can be written 2:5:204, meaning 5 hours and 204 halakim into a Monday.

The molad for any given calendar month and calendar year is calculated from an earlier molad whose time and day of the week are known by adding 4 weeks, 1 day, 12 hours, and 793 halakim per elapsed calendar month between molads.

{.note} This duration is commonly written 4:1:12:793, where the components are weeks, days, hours and halakim, though the number of weeks can be dropped for the purpose of determining the length of the calendar year. As 793 halakim does not equate to an integer number of seconds, doing the calculation in halakim rather than seconds avoids rounding errors if floating point arithmetic is used.

{.note} The computation of the molad is simplified by knowing that a complete 19 year Metonic cycle always contains exactly 235 calendar months.

{.example} The year AM 5779 began in September AD 2018, 5778 years after the first molad. 5778 divided by 19 is 304, with a remainder of 2, so from the first molad to the Tishrei molad for AM 5779, there were 304 complete Metonic cycles plus two extra years, which are both non-leap years, being years 1 and 2 in the Metonic cycle. This means 71 464 calendar months elapsed between the first molad and the Tishrei molad for AM 5779. Adding 71 464 lots of 4:1:12:793 to 2:5:204 gives 301482:2:14:316. The number of weeks can be discarded as it is irrelevant when determining the year length, leaving just 2:14:316. This says the Tushrei molad for the year AM 5779 happened on a Monday, being the second day after the Sabbath, at 08:17:33⅓, when converted to hours, minutes and seconds after midnight, rather than hours and halakim after sunset.

Once the Tishrei molad is known, the length of the calendar year can be determine using the following table. If the Tishrei molad is on or after the instant in the first column, and before the instant in the second column, then the length of the calendar year can be found in one of the last four columns, depending on the position of the year in the Metonic cycle as shown in the column headings.

Molad ≥ Molad < 1, 4, 9, 12 or 15 7 or 18 2, 5, 10, 13 or 16 0, 3, 6, 8, 11, 14 or 17


0:00:000 0:18:000 355 355 355 385 0:18:000 1:09:204 353 353 353 383 1:09:204 1:20:491 355 355 355 383 1:20:491 2:15:589 355 355 355 385 2:15:589 2:18:000 354 354 355 385 2:18:000 3:18:000 354 354 354 384 3:18:000 4:11:695 354 354 354 383 4:11:695 5:09:204 354 354 354 385 5:09:204 5:18:000 355 355 355 385 5:18:000 6:00:408 353 353 353 383 6:00:408 6:09:204 355 353 353 383 6:09:204 6:20:491 355 355 355 383 6:20:491 7:00:000 355 355 355 385

Cheshvan has 29 calendar days except when the calendar year is 355 or 385 calendar days long, in which case it has 30 calendar days. Kislev has 30 calendar days except when the calendar year is only 353 or 383 calendar days long, in which case it only has 29 calendar days.

{.example} The previous example calculated the Tishrei molad for AM 5779 to be at 2:14:316, which means the fourth row of the table applies. As 5779 divided by 19 is 304 with a remainder of 2, the year AM 5779 is the second year in the Metonic cycle, so the fifth column, whose heading includes "2", gives the year length as 355 calendar days. That means both Cheshvan and Kislev have 30 calendar days. "30 CSH 5779" and "30 KSL 5779" are therefore both well-formed dates.

{.note} This table is called the "Table of Four Gates". Its complexities are the result of the competing requirements that Yom Kippur does not fall on a day adjacent to the Sabbath (i.e. that 10 Tishrei not fall on a Friday or Sunday); that the new moon represented by the molad cannot be observed in daylight until it is six hours old and Tishrei cannot start until the new moon it deemed observable; that all months have 29 or 30 calendar days; and that only Cheshvan and Kislev can vary in length.

A date which uses a calendar day number which is greater than the number of calendar days in the specified year and month is not a well-formed date.

{.ednote} The TSC decided to include a detailed specification of the Hebrew calendar in this standard rather than referencing a third party description largely because of the lack of a good, accessible description of calendar presented in a way that would make it easy to implement the calendar in ELF. Most sources describing the calendar do so in terms of the postponement rules which makes it easier to understand why the calendar is as it is, but does not make it straightforward to implement. It is hoped that by providing these details here, vendors will be encouraged to provide full implementations of the calendar.

The elf:Time datatype {#time}

ELF uses the elf:Time datatype to represent times of day using the 24-hour clock in hours, minutes and seconds, with these components separated by a colon (U+003A). The hours and minutes components are required, and the seconds component should be provided. A fractional seconds component may be provided.

{.example} The value "15:30:00" is a valid time of day, represented using the elf:Time datatype. It represents half past three in the afternoon.

The elf:Time datatype is typically used in conjunction with a value of type elf:DateExact called its associated date.

{.note} The elf:DateExact datatype is limited to expressing dates in the Gregorian calendar, so the associated date will always be a Gregorian date.

{.example ...} In [ELF Data Model], times of day are found as TIME substructures of a DATE element which has a payload of datatype elf:DateExact.

2 DATE 10 DEC 2018
3 TIME 13:52:00

In this example, "10 DEC 2018" is the associated date for the time "13:52:00". {/}

The lexical space of the elf:Time datatype is the set of strings which match the following Time production, as well as the other constraints given here on the numerical value of each component. Whitespace is not permitted anywhere in an elf:Time value.

Time     ::=  HH ":" MM (":" SS)? TZD?  

HH       ::= [0-9] [0-9]
MM       ::= [0-9] [0-9]
SS       ::= [0-9] [0-9] ("." [0-9]+)?
TZD      ::= "Z" | ("+" | "-") HH ":" MM

The HH production encodes the hours component of the time of day, zero-padded to two digits. It shall be a decimal integer between 00 and 24, inclusive. Values outside this range are outside the lexical space of elf:Time. The hours component shall only be 24 if the minutes and seconds components are both zero or absent; any other uses of 24 as an hours component is outside the lexical space of the datatype.

{.example} The strings "30:00" and "24:30" are both outside the lexical space of this datatype. The former is invalid because the hours component of 30 is not between 00 to 24; the latter is invalid because it has an hours component of 24 and a non-zero minutes component.

The string "24:00:00" is the end-of-day instant and denote the final instant of a calendar day. It is the same instant as the first instant of the following calendar day (which is denoted 00:00:00). Conformant applications must accept end-of-day instants as valid input but must not create new instances of them. A conformant application may convert an end-of-day instant to 00:00:00, but only if it has an associated date and that is simultaneously incremented by one day.

{.example} A time of day of "24:00:00" with an associated date of "30 NOV 2018" may be converted to a time of day of "00:00:00" with an associated date of "1 DEC 2018".

{.note} [GEDCOM 5.5.1] does not specify whether or not the end-of-day instant is legal, and existing applications are unlikely to produce it. It is supported by elf:Time for compatibility with [ISO 8601] and the xsd:time datatype defined in [XSD Pt2].

The MM production encodes the minutes component of the time of day, zero-padded to two digits. It shall be a decimal integer between 00 and 59, inclusive. Values outside this range are outside the lexical space of elf:Time.

The SS production encodes the seconds component of the time of day, zero-padded to two digits and followed by an optional fractional component. It shall be a decimal greater than or equal to 00 and strictly less than 61. Values outside this range are outside the lexical space of elf:Time.

Applications must preserve at least the first three decimal digits of a fractional seconds component, but may truncate or round the fractional part of the seconds component beyond that. Applications may add or remove trailing zeros on the frational part of the seconds component, and should add a seconds component of :00 if none was given.

{.note} These provisions allows applications to process times of day using the standard data structures and facilities provided in many program languages, without preserving the original lexical form of the time of day.

{.note} A future version of this standard is likely to make the seconds component required to make this datatype fully compatible with the xsd:time datatype defined in [XSD Pt2].

A seconds component greater than or equal to 60, and strictly less than 61 is called a leap second component. Any use of a leap second component is part of the lexical space of elf:Time, but applications shall only create new leap second to represent leap seconds inserted by the International Earth Rotation Service or its successor.

Conformant applications encountering a leap seconds component may convert it to an ordinary seconds component by subtracting one from its value. This should not be done if the application supports leap seconds and knows the specified time of day was a leap second.

{.example} The string "12:56:60.800" is in the lexical space of this datatype and has a leap second component of 60.800. As leap seconds are always inserted at midnight UTC and there is no timezone in which this time of day is midnight UTC, this time of day could not arise in practice and therefore conformant applications must not generate such a time of day. Conformant applications must accept such times of day but may subtract one second to convert it to "12:56:59.800".

{.note} [GEDCOM 5.5.1] makes no mention of leap seconds, but existing applications are likely to generate such times of day if they happen to save a file during the inserted leap second. Leap seconds are also supported by [ISO 8601], but not in the xsd:time datatype defined in [XSD Pt2]. These rules allow compatibility with these standards and with existing use, while also allowing applications to ignore the leap second.

The lexical space of the elf:Time datatype allows the inclusion of a time zone designator matching the TZD production. Conformant applications must accept time zone designators on dates in input, but should ignore them and may remove them. Conformant applications must not include time zone designators on newly generated times of day.

{.note} Syntactic support for time zone designators is an extention to [GEDCOM 5.5.1], and is included in ELF for forwards compatibility. This standard does not specify the meaning conveyed by a time zone designator, but the syntax used is compatible with [ISO 8601] and the xsd:time datatype defined in [XSD Pt2], and a future version of this standard is likely to define it by reference to the latter standard.
Application choosing not to ignore time zone designators are encouraged to follow [XSD Pt2].

Formally, the elf:Time datatype is a structured non-language-tagged datatype which has the following properties:

: Datatype definition


Name https://terms.fhiso.org/elf/Time Type http://www.w3.org/2000/01/rdf-schema#Datatype Pattern (([01][0-9]|2[0-3]):[0-5][0-9](:([0-5][0-9]|60)(\.[0-9]+)?)? |24:00:00(\.0+)?)(Z|(\+|-)([01][0-9]|2[0-3]):[0-5][0-9])? Supertype No non-trivial supertypes Abstract false


{.note} The pattern in the table above has been split on to two lines for convenience of presentation; it is, however, really one pattern and contains no whitespace or line breaks. really one single line. Any functional difference between the Time production and the pattern specified above is unintentional.

The elf:Age datatype {#age}

The age of a living individual is defined as the duration which has elapsed since the instant of their birth.

{.note} It is normal to stop counting age at the instant of an individual's death, therefore an age given for a posthumous event such as a burial is likely to be the age of the person when they died, i.e. the duration of the time interval from birth to death, rather than from birth to burial.

The elf:Age datatype is used to represent ages in ELF, which it does by recording number of calendar years, calendar months and calendar days to have elapsed during the duration.

The lexical space of this datatype is the set of strings which match the following Age production:

Age      ::= ( [<>] S? )? ( Duration | BareYear ) | AgeWord

Duration ::= [0-9]+ "y" ( S? [0-9]+ "m" )? ( S? [0-9]+ "d")?
             | [0-9]+ "m" ( S? [0-9]+ "d" )? | [0-9]+ "d"
BareYear ::= [0-9]+
AgeWord  ::= "CHILD" | "INFANT" | "STILLBORN"

{.note} These productions are case-sensitive, so the string "17y" matches the Duration production, while "17Y" does not.

{.note} Because the elf:Age datatype is currently only used in ELF in contexts where whitespace normalisation has been carried out, the S production will only ever match exactly one space character.

A conformant application serialising an age using this datatype should use a single space character (U+0020) wherever whitespace is permitted in the Age production.

{.note} [GEDCOM 5.5.1] under-specifies how whitespace is allowed in ages. The Age production is permissive in its treatment of whitespace, including allowing it to be omitted entirely. The preceding recommendation is intended to make conformant ELF applications maximally compliant with current GEDCOM implementations which typically use a single space character.

An age which uses the elf:Age datatype shall either consist of a quantitative duration conforming to the Duration or BareYear productions, or an qualitative age keyword conforming to the AgeWord production.

When an age is a quantitative duration matching the Duration production, it shall contain one, two or three integers, each followed by a "y", "m" or "d" suffix to denote a number of calendar years, calendar months or calendar days, respectively, the sum of which is the age.

{.example} The age "15y 2m" represents a duration of 15 years and 2 months.

If there is only a number of calendar years present, the BareYear production allows the "y" suffix to be omitted, though this is not recommended.

{.example} The age "42y" only specifies a number of calendar years and so can be written "42" without the "y" suffix. The age "15y 2m" must not be written "15 2m" as this duration includes both a number of calendar years and calendar months.

{.note ...} Allowing the "y" suffix to be omitted as shown in the Duration production is a change from [GEDCOM 5.5.1] where it was required. This has been relaxed because GEDCOM files fairly commonly contain lines like the following, despite them being illegal:

2 AGE 58

{/}

An age which is written including a (possibly zero) number of calendar days may be presumed to have a precision of a few calendar days; an age which includes a (possibly zero) number of calendar months, but no explicit number of calendar days, may be presumed to have a precision of a few calendar months; an age containing only a number of calendar years may be presumed to have a precision of a few calendar years.

{.example ...} The age "40y" may be assumed to have a precision of a few years. The actual precision will depend on context. In many situations the precision will be exactly one year, and so assuming the age has been given correctly, the individual was born at least 40 years ago but not as long as 41 years ago.

At other times, a lower precision may apply. For example, on the 1841 census of Britain, adults were asked to round their age down to the previous multiple of five years. In such a context, an age of "40y" means the individual was born at least 40 years ago but not as long as 45 years ago. As ELF does not provide a means of stating a range of years in an age, nor of explicitly stating the precision, an age recorded as "40" on the 1841 census should be encoded in ELF as "40y". {/}

The precise meaning of an age which is quantitative duration is dependent on context, cultural considerations and the calendar in use.

{.example} The previous example gave an example of how the interpretation of "40y" depended on context, due to different precisions.

{.example} An elderly person reckoning their age using the Islamic calendar could consider their age in years to be several years greater than if they were reckoning their age using the Gregorian calendar because the Islamic calendar year is around 11 calendar days shorter than the Gregorian calendar year. ELF does not provide a means of specifying which calendar was used when recording an age.

{.note} A future version of ELF may add facilities to specify which calendar is being used for ages and also to allow an explicit precision to be stated, for example to say that an age has been given within a five-year age bracket.

Ages should generally be rounded down when expressed using just a number of calendar years, or using a number calendar years and calendar months, but an application should not assume this is necessarily the case without additional information or context.

{.example} In many parts of East Asia, ages are rounded up rather than down, so a newborn infant is considered to be one year old, and becomes two years old on the first anniversary of their birth.

A conformant application may remove any leading zeros preceding a non-zero digit, change how whitespace is used in an age, or append any omitted "y" suffix, but must not otherwise alter the age. In particular, applications must not insert or remove zero components, except as allowed below, nor convert between days, months and years.

{.example} "2y 0m", "2y" and "24m" are all distinct ages and applications must not rewrite one to another. However a conformant application may convert "2" to "2y", adding the "y" suffix; may convert "02y" to "2y", removing leading zero; and may convert "2y0m" to "2y 0m", a simple change in whitespace.

In addition, a conformant application may insert or remove components which are the number zero followed by "y", "m" or "d" suffix, providing doing so does not alter the presumed precision of the age.

{.example ...} The ages "2y 0m" and "2y" are presumed to have different precisions: the former has a precision of a few months, while the latter has a precision of a few years. Converting one to the other by inserting or removing a "0m" component would change the precision of the age, and of this reason a conformant application must not do it.

However it is only the least significant component present (representing the smallest unit of duration) which affects the precision, and other zero components may be inserted or removed. For example, "2m" and "0y 2m" are equivalent in ELF, and applications may convert either one to the other. {/}

Ages which are quantitative durations may be prefixed with with a "<" or a ">". These are interpreted as meaning the individual is at most the specified age, or is at least the specified age, respectively.

{.note ...} This means the "<" and ">" token are actually interpreted as ≤ and ≥ operators, though in practice the precision of ages in ELF is such that difference between ≤ and <, or between ≥ and > does not matter.

In common usage, a person may be said to over 21 as soon as they've had their 21st birthday, at which point they would normally round their age down to 21. ELF allows for this usage, and "> 21y" may be used to refer to someone whose age is exactly 21. This may seem to be a deviation from [GEDCOM 5.5.1] which defines them as meaning less than or greater than the specified age, respectively, but because of the uncertainty in the intended precision of an age, there is little practical difference.
The case for interpreting "<" as less than or equal to is weaker, but there are examples in current GEDCOM files where "<" has been used on ages inferred from an age given a few months later. {/}

{.example} If a source gives an individual's age as 52 in March, their age might be inferred to be "< 52y" the preceding January. In practice, they might be 51 or 52, assuming the source is accurate. However, it is normally best not to infer ages and only to record ages when they are given in sources.

Instead of a quantitative duration, an age may be expressed qualitatively using an age word which matches the AgeWord production. This standard defines three age words: CHILD, INFANT and STILLBORN. These should be used when a source describes an individual as a child, an infant, or stillborn, respectively, or using other words which are broadly equivalent. They should not be used when a source describes an individual using a quantitative age.

{.note} The words "child", "infant" and "stillborn" can be considered societal roles which are to some extent culturally dependent, rather than well-defined age brackets. A modern source might describe a 15-year-old as a child, while in mediæval times, a person of this age would be working and unlikely to be described as a child. This standard does not therefore put firm limits on the ages meant by these terms. This is a deviation from [GEDCOM 5.5.1] which states that a child is less than 8 years old, an infant is less than 1 year old, and a stillborn individual is approximately 0 days old. Nevertheless, these ages may be useful in deciding whether a foreign word or phrase can reasonably be translated as one of these word.

The age word STILLBORN is not only a statement of age but also conveys the fact that the individual died just prior to, at, or immediately after the time of birth. This age word must not be used to refer to infants whose age is about 0 days unless they died around the time of birth.

{.ednote} Having exactly three age words seems unsatisfactory as it fails to cope with other descriptions that might occur in a source, such as "elderly". The TSC have discussed whether age words should be deprecated, or whether they should be developed into a more general and extensible age word mechanism, but have reached no conclusion so far.

Formally, the elf:Age datatype is a structured non-language-tagged datatype which has the following properties:

: Datatype definition


Name https://terms.fhiso.org/elf/Age Type http://www.w3.org/2000/01/rdf-schema#Datatype Pattern ([<>][ \t\r\n]*)?([0-9]+|[0-9]+y([ \t\r\n]*[0-9]+m)? ([ \t\r\n]*[0-9]+d)?|[0-9]+m([ \t\r\n]*[0-9]+d)?|[0-9]+d) |CHILD|INFANT|STILLBORN Supertype No non-trivial supertypes Abstract false


{.note} The pattern in the table above has been split on to three lines for convenience of presentation; it is, however, really one pattern and contains no whitespace or line breaks. Any functional difference between the Age production and the pattern specified above is unintentional.

{.ednote} Depending on the details of how the ELF data model is specified, it may make sense to split this into two datatypes: the elf:Age datatype which would be restricted to just quantitative ages and would therefore serve as a general duration datatype, and a separate elf:AgeWord datatype.
The various ELF structures having an elf:Age payload would then be changed to have payloads which were a union of datatypes. This introduces several technical complications, including the need make datatype correction work with multiple default datatypes. In principle this works fine when the default datatypes have disjoint lexical spaces, as they would here, but it would need careful specification.

References

Normative references

[Basic Concepts] : FHISO (Family History Information Standards Organisation). Basic Concepts for Genealogical Standards. First public draft. (See https://fhiso.org/TR/basic-concepts.)

[RFC 2119] : IETF (Internet Engineering Task Force). RFC 2119: Key words for use in RFCs to Indicate Requirement Levels. Scott Bradner, 1997. (See http://tools.ietf.org/html/rfc2119.)

[XML] : W3C (World Wide Web Consortium). Extensible Markup Language (XML) 1.1, 2nd edition. Tim Bray, Jean Paoli, C. M. Sperberg-McQueen, Eve Maler, François Yergeau, and John Cowan eds., 2006. W3C Recommendation. (See https://www.w3.org/TR/xml11/.)

Other references

[ELF Data Model] : FHISO (Family History Information Standards Organisation). Extended Legacy Format (ELF): Data Model. Exploratory draft.

[ELF Serialisation] : FHISO (Family History Information Standards Organisation). Extended Legacy Format (ELF): Serialisation Format. Exploratory draft.

[GEDCOM 5.3] : The Church of Jesus Christ of Latter-day Saints. The GEDCOM Standard, draft release 5.3. 4 Nov 1993.

[GEDCOM 5.5] : The Church of Jesus Christ of Latter-day Saints. The GEDCOM Standard, release 5.5. 2 Jan 1996, as amended by the errata sheet dated 10 Jan 1996.

[GEDCOM 5.5.1] : The Church of Jesus Christ of Latter-day Saints. The GEDCOM Standard, draft release 5.5.1. 2 Oct 1999.

[GEDCOM X Dates] : Intellectual Reserve Inc. The GEDCOM X Date Format. Stable draft, accessed December 2018. (See http://gedcomx.org/.)

[ISO 8601] : ISO (International Organization for Standardization). ISO 8601:2004. Data elements and interchange formats — Information interchange — Representation of dates and times. 2004.

[ISO 8601-2] : ISO (International Organization for Standardization). ISO 8601-2:2009. Data elements and interchange formats — Information interchange — Part 2: Extensions. Draft, 15 Feb 2016.

[FHISO Patterns] : FHISO (Family History Information Standards Organisation). The Pattern Datatype. First public draft. (See https://fhiso.org/TR/patterns.)

[Roman Dates] : Chris Bennett. "Roman Dates." Website, accessed October 2019.
(See http://www.instonebrewer.com/TyndaleSites/Egypt/ptolemies/chron/chronology.htm.) Last updated 2012.

[XSD Pt2] : W3C (World Wide Web Consortium). W3C XML Schema Definition Language (XSD) 1.1 Part 2: Datatypes. David Peterson, Shudi Gao (高殊镝), Ashok Malhotra, C. M. Sperberg-McQueen and Henry S. Thompson, ed., 2012. W3C Recommendation. (See https://www.w3.org/TR/xmlschema11-2/.)

[Triples Discovery] : FHISO (Family History Information Standards Organisation). Simple Triples Discovery Mechanism. First public draft. (See https://fhiso.org/TR/triples-discovery.)

\vfill


Copyright © 2018–19, Family History Information Standards Organisation, Inc. The text of this standard is available under the Creative Commons Attribution 4.0 International License.