Skip to content

Conversion rules

What Gedcom::convert() changes, by version pair. Conversions run along the chain 4.0 → 5.5 ↔ 5.5.1 ↔ 7.0, with 5.5.5 attached to 5.5.1 (SPEC section 7); output conventions and loss codes are in losses.md.

$result = Gedcom::convert(Gedcom::read('family.ged'), Version::V7_0);
file_put_contents('family-70.ged', $result->output());   // the converted file
$document = $result->document();                          // read from the output on demand

5.5.1 → 7.0

Upgrading loses nothing: what 7.0 has no structure for stays as it is (an extension). Implemented so far:

5.5.1 7.0
Any encoding, HEAD.CHAR UTF-8 with a BOM, CHAR removed
HEAD.GEDC.FORM Removed (7.0 is always lineage-linked)
HEAD.FILE Removed: like CHAR, it describes the file, not its data (SPEC section 7, "Source housekeeping structures")
CONC Joined into one line: 7.0 has no line length limit; CONT stays
@@ anywhere in text Only a leading @ is doubled: a@@b → a@b, @@ x stays
@#…@ in text Kept as text: @#DJULIAN@ x → @@#DJULIAN@ x. In a 5.5.1 DATE it is a calendar escape, which the date rule converts; elsewhere (a DATE inside an extension) it is text like any other
Xref with anything but A–Z, 0–9, _, or @VOID@ (@fam 1@, @note@) @X<n>@, the next free n in record order; every pointer follows. Valid xrefs (@1@, @I1@) are kept
Pointer to a record that does not exist, compared case-sensitively (@test@ ≠ @TEST@) @VOID@, without a loss entry: the source already has KV4001 and the name of a missing record is not data
NOTE whose pointer leads nowhere (NOTE @Smith family@: text written without @@) Text: NOTE @@Smith family@
NOTE record, NOTE pointer SNOTE; a pointer where 7.0 has no SNOTE (under FONE, ROMN) stays NOTE, made _NOTE below
A structure 7.0 does not have where it stands: NOTE, SOUR of a FONE or ROMN (a 7.0 TRAN has neither) An extension with its tag and substructures (_NOTE, _SOUR), KL2001; converting down gives the structure back
SUBN record, HEAD.SUBN HEAD._SUBN with the record's substructures, one per SUBN; the record and the pointer are removed
REPO of a source without a pointer (1 REPO + CALN) REPO @VOID@: 7.0 requires the pointer
EVEN Moved to Osaka of an individual or family without TYPE TYPE Moved to Osaka added first: 7.0 requires TYPE, and the value describes the event. FACT and IDNO without TYPE stay so (KV2003 in the output): nothing names their type
OBJE without a pointer A new OBJE record @X<n>@ (before TRLR, in order of appearance) with its substructures; the link keeps TITL. Not inside an extension, at any depth (_MILT.SOUR.OBJE)
SOUR with a description instead of a pointer A new SOUR record @X<n>@ with the description as NOTE; TEXT moves under a new DATA of the citation, NOTE, QUAY and the rest stay on it
AFN x EXID x + TYPE https://gedcom.io/terms/v7/AFN
RFN xyz:123abc, RFN x EXID xyz + TYPE https://gedcom.io/terms/v7/RFN#123abc (the conversion the 7.0 term file recommends), EXID x + TYPE https://gedcom.io/terms/v7/RFN
RIN x EXID x + TYPE https://gedcom.io/terms/v7/RIN#<HEAD.SOUR> (percent-encoded; no fragment without HEAD.SOUR)
_UID UID, value unchanged
Calendar escape in DATE (@#DJULIAN@, @#DFRENCH R@) Calendar name (JULIAN, FRENCH_R); @#DROMAN@, @#DUNKNOWN@ → _ROMAN, _UNKNOWN with extension months (_ROMAN 24 _IUL 15)
B.C. BCE
Dual year 30 JAN 1648/49 The year by a 1 January new year, the original as PHRASE: DATE 30 JAN 1649 + PHRASE 30 JAN 1648/49, as in the 7.0 appendix; 1701/99 (not the following year) keeps the first year
INT 1900 (text), (text) DATE 1900 + PHRASE text, DATE without a payload + PHRASE text; the whole value as PHRASE when the date also had a dual year
A DATE that is not a 5.5.1 date Kept when it is a 7.0 date (JULIAN 1650, written without the escape), otherwise PHRASE of an empty DATE. Keywords and months in uppercase; CHAN.DATE and HEAD.DATE, which have no PHRASE in 7.0, only change case
AGE 7.0 form: <0Y → < 0y, 11m 99y → 99y 11m, a bare number is years; CHILD, INFANT, STILLBORN → < 8y, < 1y, 0y + PHRASE; anything else PHRASE of an empty AGE
LANG name (Ukrainian, any case) BCP 47 (uk); a value that is already a tag stays; anything else und + _PHRASE (7.0 LANG has no PHRASE)
Media FORM (jpg) Media type (image/jpeg); an unknown extension application/x-<ext>
Media link with FORM beside FILE, as 5.5 writes it (and some 5.5.1 programs) FORM moves into the FILE, as 5.5 → 5.5.1 does. Not into one of several FILE, nor into one with its own FORM: then it is _FORM (KL2001)
FORM.TYPE (medium of a file) FORM.MEDI
FILE URL or relative path: /x → file:///x, c:\x → file:///c:/x, \\server\x → file://server/x, \ → /; spaces, %, ?, #, [, ] percent-encoded. A URL of ftp, http or https only has spaces and such encoded, file://C:\x → file:///C:/x. .. stays, though 7.0 does not allow it. A URI of another scheme (mailto:…, urn:isbn:…), which 7.0 FILE does not support → _FILE, KL3001
Enumerations (PEDI birth, SEX f, STAT PRE-1970, DNS/CAN) 7.0 values (BIRTH, F, PRE_1970, DNS_CAN); a value 7.0 does not have becomes OTHER + PHRASE where 7.0 has both (NAME.TYPE Screen, MEDI document, ROLE (midwife)), and stays otherwise. The event type of a citation (SOUR.EVEN Census of Galicia), whose set has no OTHER, becomes EVEN + PHRASE. SOUR.DATA.EVEN birt,deat → BIRT, DEAT
FONE, ROMN (of a name or place) TRAN with LANG from the TYPE: hangul → ko-Hang, kana → ja-Kana, romaji → ja-Latn, pinyin → zh-Latn-pinyin, wadegiles → zh-Latn-wadegile; a FONE type only for FONE, a ROMN type only for ROMN; another TYPE stays as _TYPE, with LANG und (FONE) or und-Latn (ROMN); an empty TYPE is dropped
RESI, EVEN without a value Unchanged: 7.0 requires the payload, but an empty Text is one (1 EVEN in section 1 of the specification); without substructures too, as below
An event with neither a value nor substructures (1 BURI, 1 DEAT) Y: it says the event occurred, and 7.0 has no structure without both
Any other structure with neither (1 BAPL, 1 SLGS, an empty SOUR.DATA, 1 RESI) An extension (_BAPL, _DATA), KL3001, as in 5.5.1 → 5.5.5; converting down gives the structure back
— HEAD.SCHMA lists the extension tags the conversion writes with a standard meaning: _PHRASE https://gedcom.io/terms/v7/PHRASE, unless the input already has _PHRASE; added to an existing SCHMA. Known from the prepass, so HEAD is complete when it is converted
_NO, _SDATE, _CREA, _SCHMA, _PHRASE and other extensions named after a 7.0 structure, where 7.0 has it and 5.5.1 does not; _FAMC @VOID@ and the like; SEX U + _SEX X The 7.0 structure (NO, SDATE, SEX X): how the conversion to 5.5.1 writes it. An extension 5.5.1 has the structure for (_DATE under BIRT) stays
ASSO.RELA ROLE: a 7.0 role (friend → FRIEND), a common word with PHRASE (Godmother → GODP, neighbor → NGHBR), otherwise OTHER + PHRASE

HEAD.SOUR names the program and is not a source; inside extensions (_X, _SUBN) nothing is converted but the line syntax. Values are converted by the 5.5.1 structure they belong to (DATE in CHAN is an exact date, TYPE under FORM a medium), not by tag alone. Extensions of the input are kept as they are and are not added to HEAD.SCHMA: Korin does not guess what a program meant by them. Nor are _SUBN and _TYPE, which keep 5.5.1 structures 7.0 has no standard URI for.

Not everything the gedcom7code converter does is done here: it strips @#…@ escapes from text (obsolete @#OLD FORM@ escape → obsolete escape), which drops data, and renames valid xrefs (@1@ → @S1@); it drops SUBN and its RIN, and puts extracted records before the record that points to them. It converts standard tags inside extensions (LANG under _ALL_LANGUAGES), which 7.0 gives their standard meaning but 5.5.1 does not, documents extensions of the input it knows (_LOC) in HEAD.SCHMA, and reads 1701/99 as 1699 (Korin keeps 1701: the second part is not the following year, so it is not a new-year dual date) and writes Dogri as dgr, which is Dogrib (Korin: doi). Where the two agree, its 7.0 file is the corpus golden file.

7.0 → 5.5.1

Downgrading drops nothing by default: what 5.5.1 has no structure for becomes an extension and a loss report entry. Converting the result up again gives the 7.0 document back, except for what the loss report lists. Implemented so far:

7.0 5.5.1
UTF-8 UTF-8, HEAD.CHAR UTF-8, no BOM
— HEAD.GEDC.FORM LINEAGE-LINKED, after GEDC.VERS
No HEAD.SOUR 1 SOUR Korin first in HEAD: 5.x requires the system that wrote the file (as in every conversion to 5.x)
Text with @ Every @ doubled: a@b → a@@b. @#DJULIAN@ in a 7.0 text is text, written @#DJULIAN@ again (converting up reads it as text too)
Lines of any length Split into CONC at 255 bytes, between characters, never before a space
SNOTE record, SNOTE pointer NOTE; HEAD.SNOTE stays as _SNOTE (KL2001): 5.5.1 HEAD.NOTE is text, never a pointer
HEAD._SUBN A SUBN record @X<n>@ before TRLR with its substructures; HEAD.SUBN points to the first in its place
UID _UID
EXID + TYPE https://gedcom.io/terms/v7/AFN AFN
EXID xyz + TYPE https://gedcom.io/terms/v7/RFN#123abc, …/RFN RFN xyz:123abc, RFN xyz (not for a value with :)
EXID + TYPE https://gedcom.io/terms/v7/RIN#<HEAD.SOUR> RIN
Any other EXID, or one on a record without that tag in 5.5.1 _EXID with its TYPE, KL2001. Not REFN: a number the user assigned is another kind of identifier, and it would stay REFN when converted up
A pointer to @VOID@ (FAMC @VOID@ + PHRASE) An extension with its tag and substructures (_FAMC @VOID@), KL4001
SLGC with FAMC @VOID@ (5.5.1 requires the family) _SLGC with its substructures in their 7.0 form, KL4001
A record HEAD.SCHMA defines as a standard one (TAG _USER https://gedcom.io/terms/v7/record-SUBM) That record: 0 @U1@ _USER → 0 @U1@ SUBM
RESI Kyiv (5.5.1 RESI has no value) RESI + NOTE Kyiv, KL3005
REPO @VOID@ of a source (a repository known by its call numbers) REPO without a pointer, which 5.5.1 allows
EVEN with a TYPE equal to its value (how 5.5.1 → 7.0 writes an EVEN without TYPE) The TYPE removed
Calendar names in DATE (JULIAN, FRENCH_R, _ROMAN 24 _IUL 15) Escapes (@#DJULIAN@, @#DFRENCH R@, @#DROMAN@ 24 IUL 15); BCE → B.C.
DATE + PHRASE One 5.5.1 date: the PHRASE when it is the date it came from (30 JAN 1649 + PHRASE 30 JAN 1648/49 → 30 JAN 1648/49), otherwise INT 5 MAR 1900 (phrase), or (phrase) without a date. A PHRASE of an approximate date or a range stays
A date 5.5.1 cannot write (an extension calendar) _DATE, KL3001
TIME 15:43:20Z (5.5.1 times have no time zone) _TIME, KL3001
AGE < 8y → <8y; < 8y + PHRASE CHILD → CHILD. An age in weeks, or a PHRASE alone that is no 5.5.1 age (AGE + PHRASE Zero) → _AGE, KL3001
HEAD.LANG, SUBM.LANG 5.5.1 names: uk → Ukrainian, und + _PHRASE Hmong → Hmong. A tag without a 5.5.1 name (en-GB) → _LANG, KL3001. A SUBM.LANG after the three 5.5.1 allows keeps its tag (_LANG fr, KL2002)
Media FORM A file extension: one 5.5.1 lists (image/jpeg → jpg, image/tiff → tif), another known one (image/png → png), application/x-x → x; others stay
FORM.MEDI FORM.TYPE
TITL of a media link by pointer (OBJE @O1@ + TITL) _TITL, KL2001: the 5.5.1 link by pointer has no substructures
FILE A path: file:///c:/My%20Photos/a.jpg → c:\My Photos\a.jpg, file:///x → /x, file://server/x → \\server\x, a relative path without percent-encoding. Other URLs stay
Enumerations 5.5.1 case and spelling (BIRTH → birth, PRE_1970 → PRE-1970, DNS_CAN → DNS/CAN); OTHER + PHRASE → the phrase ((midwife) for a role in an event)
A value a closed 5.5.1 list does not have (MEDI OTHER + PHRASE Portrait: FORM.TYPE lists photo, book…; STAT INFANT) The 7.0 structure as an extension (_MEDI OTHER + PHRASE Portrait, _STAT INFANT), KL3001
ASSO.ROLE RELA: GODP + PHRASE godmother → godmother, FRIEND → Friend, OTHER + PHRASE land-lord → land-lord
TRAN of a name or place FONE/ROMN with TYPE where 5.5.1 has the type: ko-Hang → FONE hangul, ja-Kana → FONE kana, ja-Latn → ROMN romaji, zh-Latn-pinyin → ROMN pinyin, zh-Latn-wadegile → ROMN wadegiles; und/und-Latn + _TYPE x → FONE/ROMN + TYPE x. Another language → _TRAN, KL2001
SEX X SEX U + _SEX X, KL3002
A structure 5.5.1 does not have where it stands: NO, SDATE, CREA, INIL, CROP, HEAD.SCHMA, MIME/LANG of a note, ASSO of an event, a PHRASE no value could take An extension with its tag and substructures (_NO, _SDATE, _SCHMA), KL2001; converting up gives the structure back
_NOTE, _SOUR and other extensions named after a 5.5.1 structure, where 5.5.1 has it and 7.0 does not (under TRAN), or that 7.0 cannot write empty (_BAPL) The 5.5.1 structure: how the conversion to 7.0 writes it. An extension 7.0 has the structure for stays
More of a structure than 5.5.1 allows (a second RFN of a record) The extra ones as extensions, KL2002 (an EXID that would be a second RFN stays _EXID, KL2001)
Y of an event 5.5.1 writes without a value (BURI Y, CREM Y) No value: an event without substructures means the same, and converting up gives Y back; one with substructures is reported, KL3003

Every value is chosen so that converting it up gives the same 7.0 value and PHRASE again (so ROLE GODP without a phrase is RELA Godp: Godparent would come back with a PHRASE). Where no 5.5.1 value does, the structure is kept as an extension with a loss entry, or, for a PHRASE alone, left to the rule for structures 5.5.1 does not have. Values 5.5.1 programs write although 5.5.1 does not list them stay as they are (PEDI guardian, FORM png, RESN confidential, locked).

Xrefs are kept: every 7.0 xref is a valid 5.5.1 one. Extensions (_X) and what is inside them only get the 5.5.1 line syntax.

5.5.1 → 5.5

5.5 is written in UTF-16 (big-endian, with a BOM, HEAD.CHAR UNICODE). Nothing is dropped by default; converting the result up again gives the 5.5.1 document back, except for what the loss report lists with a KL3 code.

5.5.1 5.5
A structure 5.5 does not have where it stands: EMAIL, FAX, WWW, ADR3, MAP, FONE, ROMN, NAME.TYPE, FACT, FAMC.STAT, RELI and RESN of an event, RESN of a family, MEDI An extension with its tag and substructures (_EMAIL, _MAP), KL2001
QUAY of an inline citation (SOUR Told by his sister + QUAY) _QUAY, KL2001: only the 5.5 citation by pointer has it
More of a structure than 5.5 allows: a second REPO of a source, a second FILE of a media link The extra ones as extensions (_REPO, _FILE), KL2002
REPO of a source without a pointer (1 REPO + CALN) _REPO, KL3001: 5.5 requires the pointer
A value 5.5 does not list: SEX U, RESN confidential, STAT EXCLUDED The structure as an extension with its value (_SEX U), KL3001
Media FORM jpg, tif jpeg, tiff, the 5.5 spelling
EVEN Served in the navy 5.5 EVEN has no value; it describes the event with TYPE, the element 5.5.1 writes in the value: 2 TYPE Served in the navy (KL3004, moved). An EVEN with a TYPE already: 2 NOTE Served in the navy (KL3004, note)
Media link: OBJE + FILE + FORM FORM belongs to the OBJE: the FORM of the first FILE moves up, before it
Media record: FILE a.jpg + FORM + TITL A 5.5 record has no FILE: it holds the object itself (BLOB). The FORM and TITL of the first FILE move up to the record, the file is _FILE a.jpg (KL2001). The 5.5 validator reports the missing BLOB (KV2003)
ASSO TYPE INDI added as its first substructure: 5.5 requires it
BIRT Y and the other events Unchanged: 5.5 allows Y on every event

5.5 → 5.5.1

Upgrading keeps everything: what 5.5.1 has no structure for becomes an extension and a loss report entry.

5.5 5.5.1
Any encoding UTF-8, HEAD.CHAR UTF-8
A structure 5.5.1 does not have where it stands: BLOB and OBJE (continuation) of a media record, NOTE of a media link, SOUR of a note or a place, AGE of a family event An extension with its tag and substructures (_BLOB, _AGE), KL2001
A second PEDI of a FAMC _PEDI, KL2002
STAT CLEARED, QUALIFIED, INFANT where 5.5.1 does not list it _STAT, KL3001
Y of an event 5.5.1 writes without a value (BURI Y, DIV Y) No value, as 7.0 → 5.5.1 (KL3003 with substructures)
Media FORM jpeg, tiff jpg, tif
Media link: OBJE + FORM + FILE FORM moves into the FILE
Media record with _FILE (as 5.5.1 → 5.5 writes it) or a FILE (as some 5.5 programs write) A FILE with the record's FORM and TITL. A record without either keeps them as _FORM, _TITL (KL2001); the 5.5.1 validator reports the missing FILE
ASSO + TYPE INDI TYPE removed; an association with another record (TYPE FAM) → _ASSO, KL2001
_EMAIL, _MAP, _SEX U and other extensions 5.5.1 could write as the structure it has there The structure (EMAIL, MAP, SEX U): how 5.5.1 → 5.5 writes it

A conversion between 5.5 and 7.0 goes through 5.5.1: 7.0 → 5.5 is 7.0 → 5.5.1 and then 5.5.1 → 5.5, with one loss report for both, located in the input. Converting a 7.0 file to 5.5 and back gives the same fingerprint, except for KL3 losses, as it does through 5.5.1 (tested on every 7.0 input of the corpus).

A structure 5.5.1 does not have where it stands (5.5 and 7.0 NOTE.SOUR, 7.0 FAM.NCHI.DATE) goes through it as an extension, and the next step makes it the structure again. Its value and what is inside are still in the form of the version before 5.5.1. They are converted as the 5.5.1 structure of that tag that the value fits (_DATE INT 1900 (about then) → DATE 1900 + PHRASE), a 5.5 media link inside has its FORM moved into the FILE, and a value the target does not list becomes an extension (7.0 ROLE NGHBR → 5.5 _ROLE, KL3001). The KL2001 of the intermediate extension is not reported: the output has the structure. Under LossPolicy::Fail the first step still stops at it. An extension of the input that a conversion makes the structure has its value converted so too (5.5.1 FAM.NCHI._DATE INT 1900 (about then), as 7.0 → 5.5.1 writes it, → 7.0 DATE 1900 + PHRASE). What is inside is in the target's form already.

4.0 → 5.5

GEDCOM 4.0 (1989) has a dictionary of tags, but no grammar of where they go, and writes values in words. A 4.0 file (HEAD.GEDC.VERS 4.0, 4.0.4.18 as PAF writes it, or 4+ as Roots and Ultimate Family Tree do) is read as Version::V4_0 and validated with the 5.5 profile (KV1009); it is read only, so it converts to 5.5 and from there to any version. Best effort: the values take the form 5.5 has for them, and nothing is reported for that, as with other spellings.

4.0 5.5
SEX Male, Female M, F
A date in words: month names (11 June 1861, Sept.), Abt., c., Before, After, Between … and … The 5.5 date when it then fits 5.5: 11 JUN 1861, ABT 1850, BET 1880 AND 1882
Any other date that does not fit 5.5, which 4.0 calls irregular and keeps as written (4/5 January 1751/52, 7-12-84, 2 days after Easter, 1690) A date phrase: (4/5 January 1751/52)
A date with @ (a 4.0 calendar escape) or a parenthesis Stays; the 5.5 validator reports it
A structure 5.5 does not have where it stands (HEAL, SYMB) An extension, KL2001
HEAD without GEDC.FORM, CHAR IBMPC GEDC.VERS 5.5, FORM LINEAGE-LINKED, CHAR UNICODE

5.5 requires a submitter, which a 4.0 file may not have; the conversion does not invent one. The + continuation and the @#L and @#C escapes of 4.0 are not read: such lines are kept as they are (KV1001).

A 5.0–5.4 draft (and 5.01, the 5.1 of some programs) is read as 5.5, with KV1009. Converted to 5.5 it is written as 5.5 (GEDC.VERS 5.5, CHAR UNICODE), not copied as a 5.5 document is; no rule changes its structures.

What a file declares in GEDC.VERS (4+, 4.0.4.18, 5.01) describes the file, not its data: each step reads the version it converts from there, and the output names the target.

5.5.1 → 5.5.5

5.5.5 is 5.5.1 made strict: UTF-8 with a BOM, GEDC.FORM LINEAGE-LINKED with its own VERS 5.5.5, no inline sources or media, no structure with neither a value nor substructures, and stricter grammars for values. Nothing is dropped by default; converting the result up again gives the 5.5.1 document back, except for what the loss report lists with a KL3 code, the inline sources and media (records in 5.5.5), new ADR lines and respelled values.

5.5.1 5.5.5
HEAD.GEDC.FORM FORM LINEAGE-LINKED with VERS 5.5.5, added when missing
HEAD.SUBM and the SUBM records The record HEAD.SUBM points to (or else the first; a missing HEAD.SUBM then points to it, unless HEAD has _SUBM, a pointer to no record) moves directly after HEAD; any other SUBM record → _SUBM, KL2002. A file without a SUBM record keeps none
SUBN record, HEAD.SUBN HEAD._SUBN with the record's substructures, as 5.5.1 → 7.0
Inline SOUR Family Bible + TEXT A SOUR record @X<n>@ with TITL Family Bible; the citation points to it, its TEXT under a new DATA
Inline OBJE + FILE + FORM + MEDI + TITL An OBJE record @X<n>@ with the FILE, TITL under the first FILE, MEDI as FORM.TYPE; a FORM beside FILE moves into it. A FORM 5.5.5 does not list (ole) stays (KV3006, see the matrix limits)
ADDR 12 Main St + CONT apt. 4 5.5.5 ADDR has no value: its lines are ADR1–ADR3. Lines an ADR line repeats are dropped, missing ADR lines added. A value made of CITY, STAE POST and CTRY is dropped. Any other value (more than three lines, an empty line, a line other than its ADR) → a first substructure _ADDR, KL3001
A structure with neither a value nor substructures (BURI, BIRT Y: 5.5.5 has Y only for MARR, CHR, DEAT) An extension (_BURI, _BIRT Y), KL3001
SEX U + _SEX X (7.0 → 5.5.1 writes it for SEX X) SEX X
A structure 5.5.5 does not have where it stands: LDS ordinances, RESN, ALIA, ANCI, DESI, AFN, RFN, HEAD.PLAC, HEAD.CHAR.VERS, SUBM.LANG, SOUR of a note An extension with its tag and substructures, KL2001
More than 5.5.5 allows: a second FILE of a media record, a second CALN The extra ones as extensions, KL2002
Dates: 05 JAN 50, ABT 50 Days without a leading zero, years of at least 3 digits: 5 JAN 050, ABT 050
Ages: <01y 0m, 42m < 1y: a space after < and >, no leading zeroes, no zero parts, months past 11 in years (42m → 3y 6m; days stay). An age with no part left (0y) → _AGE, KL3001
Times: 09:05:00.5 9:05:00.50
Places: Kyiv,Ukraine Kyiv, Ukraine; a place with an empty part (Kyiv, , Ukraine) stays as it is
A value no 5.5.5 spelling fits (HEAD.SOUR.VERS 7.0 beta) The structure as an extension, KL3001

Lengths (5.5.5 {Size=min:max}, KV3010) are not converted: a value too long or too short for 5.5.5 is still its value. FACT and IDNO without the TYPE 5.5.5 requires get none: there is nothing to name it by.

5.5.5 → 5.5.1

5.5.5 5.5.1
GEDC.FORM.VERS Removed
ADDR without a value The value: the first substructure _ADDR (removed), or else ADR1–ADR3 as lines, or else CITY, STAE POST, CTRY
HEAD._SUBN A SUBN record, as 7.0 → 5.5.1
_SUBM records SUBM records
SEX X, SEX N SEX U + _SEX X, _SEX N, KL3002
Dates: ABT 050, 1000 BCE ABT 50, 1000 B.C.
A day and month without a year (5 MAR) _DATE, KL3001: 5.5.1 has no such date
Ages: < 1y <1y
Media FORM JPEG, TIFF jpg, tif; other formats 5.5.1 does not list (PNG, PDF) stay, as 5.5.1 programs write them
_BURI, _BIRT Y, _DATE and other extensions 5.5.1 could write as the structure it has there The structure

5.5.5 ↔ 7.0 goes through 5.5.1. A 7.0 file converted to 5.5.5 and back has the same fingerprint, except for KL3 losses and what 5.5.5 writes its own way: the SUBM records directly after HEAD (with HEAD.SUBM), ADR1–ADR3 lines that repeat the ADDR value, and times without a leading zero (8:56) or with a two-digit fraction (.50), the same times in 7.0 (tested on every 7.0 input of the corpus).

Conversion matrix

Every corpus input, and random valid documents of every version (tests/Generative), is converted to every version (tests/Corpus/MatrixTest.php, SPEC section 10): the conversion is deterministic, converting the output to its own version changes nothing, and the output validates without errors other than those of the input and these limits of the target versions. An error of the input stays one where it was: with its code, at the same path, or, where the conversion moved a structure (an inline OBJE that becomes a record), with the same message; an error of that code elsewhere is a new one.

Target Error the output may have Why
5.x HEAD requires SUBM (KV2003; 5.5.5: the file requires a SUBM record) 7.0 does not require a submitter, and the conversion does not invent one
5.x FORM does not permit … (KV3006) on a media FORM 5.x lists a few formats from the 1990s; others (png, pdf) stay, as programs write them
5.5, 5.5.1 RESN does not permit "confidential, locked" (KV3006) 5.x has one restriction; programs write the 7.0 list, and keeping one of them would lift a restriction
5.5 OBJE requires BLOB (KV2003) A 5.5 media record holds the object itself; a file reference is _FILE (KL2001)
5.5.5 … is longer than 248 characters, … is shorter than … (KV3010) Lengths are not converted: a 7.0 text has no limit, and cutting it would lose it; a short one (EMAIL x) is not padded
7.0, 5.5.5 FACT requires TYPE, IDNO requires TYPE (KV2003) 5.5.1 does not require TYPE, and nothing names it
7.0 EVEN requires TYPE (KV2003) An EVEN without a value has nothing to name its type with (one with a value gets it as TYPE)
5.5.1, 5.5.5, 7.0, from 5.5 OBJE requires FILE (KV2003) The 5.5 media record holds the object itself (BLOB), which is not a file
5.5.1, 5.5.5, 7.0, from 5.5 REPO requires NAME, STAT requires DATE (KV2003) 5.5 requires neither the name of a repository nor the date of an LDS status, and the conversion does not invent them