5.21 The bibliography (TPS 0.3.0)
Instruction artifact, machine-optimized text: TPS, section 5.21 The bibliography. Part of TPS and the TPM hub. Human-only edit (TPS 5.13). TPS 0.1.0 was declared final at revision 4 on 2026-09-22, a version declaration recorded in the foundations record; the chain begins at TPX's first landing (TPS 5.6.7). TPS 0.1.1, which changed only the status lines, was declared final on 2026-09-26. This page is revision 6 of the tree, proposed as version 0.3.0 until H-DECL declares it; revision 5, proposed as 0.2.0, stays undeclared and is carried whole; the section is new at revision 5. The text below is this section of the TPS artifact at 0.3.0 (proposed), verbatim; the whole artifact and its file digest are on The Principia Machina/TPS. Human readers: the contents page is the map.
TPS 5.21 THE BIBLIOGRAPHY
purpose: the bibliography unit: the BIBLIOGRAPHY SETTING and successors,
the BIBLIOGRAPHY beside the register, the BIBLIOGRAPHY ENTRY and
the BIBLIO MARK, the stored form, identifiers, the write check,
append-only and the status lines, who adds, LITERATURE, what an
entry never holds, reading and search, holds and packages, the
lookup under REGISTRY DECLARATIONS, the comparison and the CHECK
value, notices, FK-20, rechecks, the lines, the LITERATURE VIEW,
budget and etiquette, the page rules, the functions as built
governs: 5.21 | version: 0.3.0 (proposed) | class: instruction artifact
audience: operator, agent
5.21.1 THE BIBLIOGRAPHY SETTING (D170.1). A UWC keeps a BIBLIOGRAPHY only
under its BIBLIOGRAPHY SETTING, OFF, ON or CLOSED, one of UWC STATE's
fields beside ACCESS POLICY and AUDIT TRIGGER. The new-UWC form
carries the selection, OFF or ON, OFF unless the initiator selects
ON; the DRAFT UWC API takes the same selection, default OFF; H-DECL
declares it in UWC STATE with the certificate (5.5.11, 5.6.14), and
the control panel shows it (5.12.12). At instantiation under ON, TPX
creates Bibliography Volume 1 with the register. H-DECL may later
declare OFF to ON, on which TPX creates the first volume, and ON to
CLOSED, after which no entry is added and every entry stays landed,
citable and checked; never ON or CLOSED to OFF: delete is used by no
one (5.13.3), and a record would otherwise cite an entry that no
longer shows. The setting is H-DECL's declaration, not H-OWN's and
not a tuning option: it fixes what, not how fast (5.16.12). [D170.1]
5.21.2 SUCCESSORS (D170.1 (e)). A successor of a UWC whose setting is ON or
CLOSED is ON. COPY-FORWARD (3.4.9) first carries each entry, in number
order, to the new prefix with a back-pointer and its CHECK LINE and
NOTICES LINE as they stand, then remaps the LITERATURE of every
inherited record through the back-pointers; the entries are not
looked up again at the copy, since the rechecks of 5.21.18 reach
them, so a copy costs no registry call. A successor of an OFF UWC
takes the form's choice. [D170.1]
5.21.3 THE BIBLIOGRAPHY (D171.1 (a)). The BIBLIOGRAPHY is a page class of
its own: BIBLIOGRAPHY VOLUMES at "<UWC root>/Bibliography/Volume <v>",
beside the register and not in it. No entry is a register record:
the strike rule takes no entry, SR4's counts and the held closure
(3.3.13) reach no entry, and no parrot or canary is built from one
(4.4.12, 5.17.5). The bibliography is sealed with its UWC (1.3.3): a
record cites entries of its own UWC alone. [D171.1]
5.21.4 THE ENTRY AND THE BIBLIO MARK (D171.1 (b)). A BIBLIOGRAPHY ENTRY is
"<prefix>-BIB<n>", its BIB ID, n from 1 on the BIBLIO MARK, a second
counter beside the record series' high-water mark (3.1.5, 3.1.16).
TPX allocates n at landing, in push order; no lease and no VOID:
entries are numbered by TPX alone (3.1.13 does not apply). Entry n
lives on volume ceil(n / BIBLIO VOLUME SIZE) (5.21.22). [D171.1]
5.21.5 THE STORED FORM (D171.1 (c)). An entry is one CSL-JSON item in a
declared subset of the CSL data schema 1.0, with TPM fields beside
it: BIB ID, PROPOSED BY, LANDED BY, LANDED, SUPERSEDES and
IDENTIFIERS (DOI, ARXIV, ISBN, ISSN, PMID, PMCID, each normalized as
5.21.6 states, at most one per kind). THE SUBSET, in this fixed key
order for serialization: type, id, title, title-short, author,
editor, translator, container-author, collection-editor, issued,
accessed, original-date, container-title, container-title-short,
collection-title, collection-number, publisher, publisher-place,
volume, issue, page, number, number-of-pages, edition, genre,
medium, event-title, event-place, chapter-number, version, language,
DOI, ISBN, ISSN, PMID, PMCID, URL, original-title, archive,
archive_location. Names carry family, given, dropping-particle,
non-dropping-particle, suffix and literal; dates carry date-parts
(integers) and circa only; page, volume, issue, number, edition and
version are strings; type is one of the schema's 45 types; id equals
the BIB ID. Every other property of the schema is refused, abstract,
annote, note, custom, keyword, references and raw or literal dates
among them (5.21.11). THE CANONICAL SERIALIZATION: the item as JSON in
the subset's key order, names and dates in their fixed order,
strings NFC and escaped only for the quotation mark, the backslash
and control characters, no white space outside strings, integers in
decimal. THE DIGEST, digest_fields (2.5.10) over ["TPC-BIB", "1", BIB
ID, PROPOSED BY, LANDED BY, LANDED, SUPERSEDES or "", then each
identifier kind's name and stored value or "" in the order DOI,
ARXIV, ISBN, ISSN, PMID, PMCID, then the canonical text], covers the
item and these fields and not the status lines of 5.21.8; TPX writes
it into the entry's own text. The manifest tree's leaf for a
bibliography volume carries the volume page's content hash, as for
every page, and no per-entry digest list (5.11.13). [D171.1]
5.21.6 IDENTIFIERS (D171.1 (c), (d)). Each identifier is normalized on
write and refused when it does not fit. DOI: the bare form, "doi:"
in any ASCII case, or a doi.org or dx.doi.org URL (http or https), a
URL form percent-decoded once; then the shape "10." + a registrant
code of digits and dots + "/" + a suffix of graphic code points with
no white space; stored as given after the wrapper is removed and
NFC; compared and keyed after A-Z to a-z folding only, the DOI
Handbook's equivalence. A DOI of prefix 10.48550 is the arXiv DOI:
it is refused as a DOI and taken as its ARXIV identifier; the item's
DOI field may carry it, the write check then requiring ARXIV and no
DOI. A directory indicator other than 10 is refused, a named limit.
ARXIV: bare, "arXiv:" in any ASCII case, or an arxiv.org/abs URL; the
new scheme YYMM.NNNN for YYMM 0704 to 1412 and YYMM.NNNNN from 1501,
MM 01-12; the old scheme archive[.SC]/YYMMNNN, the archive
[a-z][a-z-]{1,15} and SC two capitals, YY 91 to 07 and MM 01-12; an
optional vN, N from 1 with no leading zero; stored with its version,
keyed without it. No archive catalogue is kept: an unknown archive is
the lookup's NOT FOUND. ISBN: ten or thirteen characters after ASCII
hyphens and spaces are removed; ISBN-10 by the weighted sum modulo
11 with X for 10, ISBN-13 by the alternating 1 and 3 weights modulo
10 with prefix 978 or 979; an ISBN-10 is stored converted to 13.
ISSN: NNNN-NNNC, the first seven weighted 8 to 2 modulo 11, X for
10. PMID: digits with no leading zero. PMCID: "PMC" and digits. A
second ISBN or ISSN is a second entry's or the container's, never a
second identifier of one entry, a named limit. [D171.1]
5.21.7 THE WRITE CHECK (D171.1 (d)). MECHANICAL, TPX's at landing and
PRB's advisory pre-check before the push (5.4.14), in this order:
the item's keys within the subset with the allowed types; REQUIRED:
type, title, issued with a year, and at least one author or editor,
a literal name serving for an institution or a collaboration; THE
CAPS, counted in code points after NFC: title 1,000, every other
text 500, each name part 200, URL 2,000, at most 100 names per role,
and the canonical serialization at most BIBLIO ENTRY MAX bytes
(7.2.27); every identifier normalized as 5.21.6 states; the item's
DOI, ISBN, ISSN, PMID or PMCID equal, after normalization, to its
IDENTIFIERS entry (an ISBN-10 converted to 13; a 10.48550 DOI in the
item requiring ARXIV and no DOI); no field outside the subset; and
THE DUPLICATE KEY, the folded DOI, else the ARXIV identifier without
version, else the ISBN-13, else none, not among the UWC's landed
entries unless SUPERSEDES names that BIB ID: a duplicate is refused
with the existing BIB ID. A preprint and its published version are
two entries. An entry with no duplicate key has no duplicate check:
a title-only duplicate is INSPECTION. A string that fails the text
safety of 2.5.13 is refused as its field. The refusal names the
failing check and field. [D171.1]
5.21.8 APPEND-ONLY AND THE STATUS LINES (D171.1 (e)). An entry's digested
part is never edited; a correction is a new entry naming SUPERSEDES,
and the superseded entry stays citable. TPX alone writes the status
lines into the entry's own text and nothing else there: SUPERSEDED
BY, and the CHECK LINE and NOTICES LINE of 5.21.19, written with the
entry at landing and rewritten only when they change, so a
raw-wikitext reader sees them (3.2.13 applied to the class). A
bibliography volume whose BIBLIO VOLUME SIZE entries are all landed
is FROZEN; its status lines stay writable, as a register volume's
do (3.1.14). [D171.1]
5.21.9 WHO ADDS (D171.1 (f)). Only the mutex holder's landing adds entries,
through TPX's API: a coworker proposes entries in its RETURN
(5.8.11), vetted by the lead under 6.2; a person proposes them
through the RR INPUT BUFFER (5.4.8). PROPOSED BY names the proposer
and LANDED BY the landing account. No falcon and no agent writes an
entry on any page (5.20.3, X5). [D171.1]
5.21.10 LITERATURE (D171.1 (g)). A record lists the entries it relies on in
LITERATURE, a field of the record field schema (3.2.14, C-16),
citing BIB IDs; a verification-class M record (3.3.6) names there
the entry its external value comes from. TPX refuses a LITERATURE
member that is not a landed entry of the same UWC, so PRB lands
entries in a push before the records that cite them (5.4.14). With
the setting OFF a record has no LITERATURE field. A LITERATURE
member is not a citation of a record: it enters no closure, no
distance and no hold (4.4.12). [D171.1]
5.21.11 WHAT AN ENTRY NEVER HOLDS (D171.1 (h)). Never: an abstract, a note,
full text or a figure, most of which is copyrighted; data a source's
terms forbid redistributing, or licensed data without its licence
named; a value a derivation is later compared with (AC-4.24). An
entry whose role is the source of a verification not yet run lands
with that verification record, after the derivation's step ends,
the lead's INSPECTION under benchmark quarantine. Text pasted into
a title or another free field under its cap is not detected:
INSPECTION. [D171.1]
5.21.12 READING AND SEARCH (D171.1 (i)). A bibliography volume is a page of
TPX's own content model, showing the entries as formatted references
above the machine form; the search index takes a summary per entry
(BIB ID, title, first author, year, DOI or ARXIV identifier), never
the machine form; TPX's reading view renders the whole bibliography,
or any range, on demand, stored nowhere (5.3.42). The volumes are in
the index of record's scope (5.11.13). [D171.1]
5.21.13 HOLDS AND PACKAGES (D171.1 (j)). After a failed canary TPX re-runs
the lookup of 5.21.14 on every entry the account in question
proposed or landed in a block of the hold's scope and lists in the
HOLD record's bibliography field those whose CHECK value is not
CONFIRMED (3.3.11, 5.14 CR6); entries are never HELD and no record is
held for its LITERATURE. A PACKAGE's HISTORICAL VIEW includes the
bibliography as it stood (5.17.4). [D171.1]
5.21.14 THE LOOKUP (D172.1 (a)). TPX's landing job, inside the push in LANE
2 (5.3.33), looks up each member of a new entry's IDENTIFIERS that a
declared registry takes, before the entry is saved, at the
registries H-DECL declares as REGISTRY DECLARATIONS on the
DECLARATIONS page, outside D(IS) (5.6.14, 7.2.26). A REGISTRY
DECLARATION states: NAME; IDENTIFIER KIND (DOI-RA, DOI-CROSSREF,
DOI-DATACITE, ARXIV, PMID or PMCID); the URL TEMPLATE, https, with
one slot "{ID}" in the path or the query, never the host and never
a fragment; the ACCEPT header; the RESPONSE SCHEMA name; PACING, the
least interval between calls; TIMEOUT; the CREDIT line shown
wherever its data appears, the CHECK date beside it; and the
CONTACT parameter whose value lives in TPX's configuration, never on
a page. TPX fills the slot with the identifier, every UTF-8 byte
percent-encoded in upper-case hexadecimal except A-Z, a-z, 0-9, "-",
".", "_", "~", "/", ":", "(" and ")", one rule for path and query
slots; an identifier whose encoded form leaves a "." or ".." segment
in the path is not fetched and lands LOOKUP FAILED. No agent and no
person supplies an address. The fetch runs under the SOURCE
DECLARATION's guards (5.19.3): HTTPS with certificate verification,
no private or loopback address, redirects only to declared hosts
and re-validated, at most 3 hops, the timeout and size caps, a
response refused whole when it fails its schema. THE STARTING SIX:
the DOI system's agency lookup (whether the DOI exists, and its
registration agency); Crossref's works endpoint; DataCite's dois
endpoint; the arXiv API; PubMed's E-utilities summary; and the PMC
identifier converter, PubMed's credit line shown wherever its data
is. ROUTING: a DOI goes first to the agency lookup, "does not exist"
being NOT FOUND; agency Crossref to the works endpoint, agency
DataCite to the dois endpoint, any other agency FOUND UNCOMPARED
unless a REGISTRY DECLARATION names its endpoint; a failure or a
not-found answer at the agency's own registry after the agency said
the DOI exists is LOOKUP FAILED, never NOT FOUND. An ARXIV identifier
goes to the arXiv API: a feed with no entry is NOT FOUND (VERIFY on
the host), a feed whose entry names the API's error address is
LOOKUP FAILED, except an incorrect-format error, which is NOT FOUND;
a versioned identifier is fetched with its version and its year is
the retrieved version's date, an unversioned one bare and its year
the first version's. A PMID goes to the summary endpoint; a PMCID to
the identifier converter, then to the summary by the PMID it
returns. [D172.1]
5.21.15 THE COMPARISON AND THE CHECK VALUE (D172.1 (b)). MECHANICAL, one
result per identifier looked up. The PRIMARY identifier is the
entry's DOI, else its ARXIV identifier, else its PMID, else its
PMCID. TPX parses each registry's payload into a REGISTRY RECORD
(year, title, subtitle, the first author's family name with its
name form, DOI, ISSN and ISBN lists, arXiv DOI, arXiv identifier,
and notices), a parse tested on recorded payloads; the comparison
is TPC's (5.1.12). The PRIMARY's record is compared with the entry
on the issued year, the DOI after A-Z folding where both carry one,
and the title and the first author's family name after THE KEY
FORM: markup runs removed; the five XML entities and numeric
entities decoded; NFD; every code point of the categories Mn, P,
S, Cf and Cc removed; A-Z folded to a-z, a letter that does not
decompose to ASCII keeping its case; NFC; every run of white space
collapsed to one space; trimmed; for a title, one leading marker of
the closed set "RETRACTED", "RETRACTED ARTICLE", "WITHDRAWN" or
"WITHDRAWN ARTICLE" followed by ":" stripped first, other markers,
other named entities and non-ASCII case a named limit. The family
name is compared whole against a Crossref or DataCite record, as a
suffix at a token boundary against an arXiv full name and as a
prefix at a token boundary against a PubMed "Family Initials" name;
a subtitle is joined to the title with one space; an organisation
stands as the family name. PubMed is compared, its summary carrying
title, year, first author and DOI; a PMCID is compared through the
PMID it converts to. An arXiv record beside a PRIMARY DOI is
compared on its arXiv DOI and the family name; without an arXiv DOI
it is CONFIRMED when its keyed title and family name both equal the
entry's, else FOUND UNCOMPARED. A DataCite record's arXiv identifier
is compared with the entry's ARXIV key where both carry one. The
entry's ISSN and ISBN consult the PRIMARY's found record alone:
CONFIRMED when listed there, FIELDS DIFFER when the record lists
others and not it, NO LOOKUP when the record lists none or the
PRIMARY was not found. RESULTS per identifier: CONFIRMED (resolved
and every compared field equal; an absent field is not equal);
FIELDS DIFFER (one or more compared fields unequal); FOUND
UNCOMPARED (resolved at a registry whose record TPX does not
compare); NOT FOUND; NO LOOKUP (an ISSN or ISBN no fetched record
carries, or an entry with no identifier a declared registry takes);
LOOKUP FAILED (no answer, or a lookup the bounds of 5.21.21 left
unmade). The entry's CHECK value is the first present of NOT FOUND,
FIELDS DIFFER, LOOKUP FAILED, FOUND UNCOMPARED, NO LOOKUP, CONFIRMED.
CONFIRMED needs every compared field equal, which an invented
reference cannot produce; every other value sends the identifier to
FK-20 or to the retry. [D172.1]
5.21.16 NOTICES (D172.1 (b)). A notice is the registry's own statement
about the work: Crossref's "updated-by" entries, which carry the
Retraction Watch data (retraction, correction, expression of
concern, reinstatement) and the publishers' own update types;
PubMed's retracted-publication type and its retraction and
expression-of-concern references; an arXiv withdrawal, read from the
retrieved version's comment field holding the whole token
"withdrawn" in any ASCII case, a pattern with a named limit. Each is
recorded as type, in the key form, identifier (the notice's DOI or
PMID where given), date and source, as the registry gives them; TPX
keeps no catalogue of kinds, and a notice moves no status. [D172.1]
5.21.17 FK-20 BIBLIOGRAPHY CHECK (D172.1 (c)). Every identifier whose result
is neither CONFIRMED nor LOOKUP FAILED goes with its entry to FK-20
BIBLIOGRAPHY CHECK, a FALCON KIND of FALCON CLASS FINDINGS, not
CONFIDENTIAL, declared under 5.20.2 with the first catalogue; LOOKUP
FAILED is retried at each W1 run and goes to FK-20 only when still
failed after one BIBLIO RECHECK period. Its declaration: write
targets none; material served, the entry's canonical text, each
REGISTRY RECORD TPX fetched, served as SERVED entries of the packet's
source list, and the FALCON SOURCE LIST's catalogue pages; output per
identifier sent, RESULT (SOURCE AGREES, SOURCE DIFFERS or NOT FOUND;
MECHANICAL), SOURCE FIELDS (rows keyed year, title, family, doi,
issn, isbn and arxiv, each with a quote; MECHANICAL; the QUOTE CHECK
against the served record or the catalogue page) and NOTE
(INSPECTION); agreement EXACT MATCH; risk ALLOWED; cadence on
demand; rate cap none. Two agents, blind, compare the entry with the
served records or, where there is none, with the catalogue pages;
the SOURCE MATCH's SERVED kind (5.20.11) reads a served record from
TPX's store only as served, so quote checks run on text TPX holds;
fields are reconciled on the ladder; the finding, UNRESOLVED fields
shown as such, is published on the FALCON PAGE and linked from the
entry's reading view, never written into the entry. FK-20 changes
no page (5.20.12). A SOURCE DIFFERS finding is shown to the UWC's
lead as untrusted input, vetted under 6.2; the lead may land a
superseding entry, which the landing lookup checks again. FK-06 and
FK-17 read other pages; FK-20 alone reads the bibliography. The
agents' rules are AC-4.45's. [D172.1]
5.21.18 RECHECKS (D172.1 (d)). Every entry with an identifier a declared
registry takes is looked up again each BIBLIO RECHECK period for new
notices and a changed result, spread over W1's runs in LANE 5
(5.3.33), at most BIBLIO RECHECK BATCH identifiers a run under the
time bound of 5.21.21, the oldest CHECK date first when BIBLIO
LOOKUP CAP binds; the backlog is reported on the scorecard
(5.18.15). [D172.1]
5.21.19 THE LINES (D172.1 (e)). At landing TPX writes into the entry's text
one CHECK LINE, "CHECK <date> <value>" followed by one " ; <kind>
<identifier> <result>" clause per identifier looked up, the PRIMARY
kinds DOI, ARXIV, PMID and PMCID first, then ISBN and ISSN; and,
when there is a notice, one NOTICES LINE, "NOTICES <date>" followed
by one " ; <source> <identifier> <date> <type>" clause per notice,
"-" for an absent identifier or date. Either line is rewritten only
when it changes (5.21.8). The CHECK LINE shares the entry's save at
landing; a recheck's rewrite is a status write outside a landing,
one save per page per run, nothing written when unchanged, the
volume's manifest path and the badge's MANIFEST REVISION rewritten
with it (5.11.11). [D172.1]
5.21.20 THE CONSEQUENCES AND THE LITERATURE VIEW (D172.1 (f)). A CHECK LINE,
a NOTICES LINE and an FK-20 finding are information: no record is
struck, held or regraded by one, and no audit block opens on one by
TPX (4.1.11). TPX's LITERATURE VIEW, a reading view per UWC computed
on demand and stored nowhere, linked from the control panel, lists
every standing record (not STRUCK, HELD or REFUTED) whose LITERATURE
cites an entry with a NOTICES LINE, a CHECK value NOT FOUND or an
FK-20 RESULT NOT FOUND reached as AGREED, with the reason, sorted by
identifier in code point order; a record's reading view shows the
same for the entries it cites. A second list, for H-DECL, names
every entry found NOT FOUND by both FK-20 (AGREED) and the lookup
where the entry has a PRIMARY identifier; H-DECL may open an audit
block on the landing that carried it. A fabrication the audit
confirms is an audit finding against the account in PROPOSED BY
when it is an AGENT, an AGAINST event at m_2 (5.14 CR11); an entry a
USER proposed is listed and moves no score. [D172.1]
5.21.21 BUDGET AND ETIQUETTE (D172.1 (g)). TPX keeps each registry's
published rate and its PACING: the Crossref pool with a contact
address and the DataCite and NCBI contact parameters, each in TPX's
configuration and never on a page; arXiv one request per 3 s.
Bounds, tuning options (7.2.27): BIBLIO LOOKUP TIME, the wall-clock
time one landing job or one recheck run spends on lookups, after
which the rest land LOOKUP FAILED and the recheck job retries them;
BIBLIO LOOKUP CAP, the fleet's lookups a day, landings and rechecks
counted together, landings first, inside every registry's limit;
BIBLIO RECHECK BATCH; BIBLIO RECHECK. Landing lookups run in LANE 2
and are never refused on budget (7.1.41); rechecks run at CLEAR and
BUSY only. [D172.1]
5.21.22 THE PAGE RULES (D173.1). BIBLIO VOLUME SIZE is 5 entries a volume,
a declared parameter beside VOLUME SIZE, which is unchanged
(7.2.25); entry n lives on volume ceil(n / 5), fixed for the life
of the UWC, so an entry's page follows from its number alone. The
bibliography is kept like the register, in volumes with status
lines, and is not an APPEND VOLUME class: 5.3.42's list names it as
the exception, and COMPRESSION, SEARCH AND VIEWS, the MANIFEST TREE
and ONE WRITE PER PAGE PER LANDING apply to it. Measured on the
register's own efficiency analysis, 5 is the least storage of the
sizes asked about (1, 5, 10, 20, 25, 50, 100) at 10,000 and 100,000
entries, the register's 10 costing 3 to 6 per cent more in bulk and
steady arrival and about 18 per cent more where entries trickle;
page saves per landing about 1.46 at 5 against 1.27 at 10. A page of
5 is about 4.2 KB on average, at most about 11 KB. [D173.1]
5.21.23 THE FUNCTIONS, AS BUILT. The identifier functions, the key form,
the duplicate key, the write check, the canonical text and digest,
the lookup address, the comparison with the CHECK LINE and NOTICES
LINE renderings, the withdrawal pattern, the LITERATURE VIEW's
selection and the HOLD record's bibliography list are TPC functions
in both residences with vectors (5.1.12); the parse of a registry
payload and the fetch are TPX's (5.3.44). [D171.1] [D172.1]
Pages for this section, added by landing round 8's page batch: Walk through/Bibliography; Inputs/Bibliography; Agent/Bibliography.