Citation Cross-Check (conditionally double-blind bib-vs-web field audit)
Codex mirror adaptation (normative). This skill has no cross-model jury by
design — it is an evidence instrument, not an acquittal step (see the acceptance-gate
note in Step 2, and pair it with citation-audit for the cross-model verdict). A and B
are read-only extraction shards and C emits a mechanical per-field delta, so every
verdict is assigned by the main agent from that delta plus its own primary-source
fetch — not by any shard and not by a reviewer model.
Blindness is NOT automatically identical across hosts — it is conditional on
isolation, and the gate enforces that. The whole cross-check rests on B never seeing
the .bib. On the mainline, the Agent tool gives B a fresh context. On a Codex host
the frontmatter grants no delegation tool (no Agent, no spawn_agent — see
allowed-tools), so:
- If the host can dispatch B in a genuinely isolated context that does not inherit
the main agent's history (which has already read the
.bib in Step 1) — passing B only
the title+first-author list — B may set blind:true and the run certifies normally.
- Otherwise (B would run in the main context, or "fresh sequential context" cannot be
guaranteed to drop the parent's
.bib view), B MUST set blind:false. The Step-3.5
gate then downgrades every MATCH/MINOR to ESCALATE: the run still flags
discrepancies for the main agent's Step-4 webcheck, but it certifies nothing. It is a
labelled heuristic, not a double-blind certification.
So the mirror never claims blindness it cannot deliver: it either has real isolation and
attests it, or it declares blind:false and drops to flag-only. Do not add a
spawn_agent-based reviewer/continuation route — the trust decision is the main agent's
WebFetch, and it must not be delegated to a shard.
🔒 Do not wrap this skill in /loop, /schedule, or CronCreate. It is
verdict-bearing — it emits a per-reference verdict that gates .bib edits (an evidence
instrument, not a cross-model acquittal; see the acceptance-gate note in Step 2).
Re-running it on a timer adds no signal (the verdict changes only when the
bibliography changes).
Schedule the external wait that precedes it: bibliography finalized → run once. See
../shared-references/external-cadence.md.
Why this exists (and how it differs from citation-audit)
citation-audit is a single cross-model reviewer that judges three things holistically:
existence, metadata, and context-fit (is the cited paper used for a claim it
actually supports?). This skill does not do context-fit. It does one thing that a
single reviewer cannot do reliably: an adversarial, field-level diff (double-blind when the host can isolate the web agent, a labelled flag-only heuristic otherwise)
of the local .bib entry against an independently reconstructed record.
The failure mode it targets is the hand-entry error that survives a plausibility read:
a dropped or re-ordered author, a wrong year, a swapped venue, a DOI that
404s. A single reviewer that sees the .bib entry tends to anchor on it and rate it
"looks correct." Two agents that never see each other's output cannot anchor — a
discrepancy only survives if both independent records agree, and the verifier reports
exactly where they diverge.
In practice it catches dropped/re-ordered authors, wrong years, and bad DOIs that a
single-pass audit rated "looks clean" — while NEEDS_WEBCHECK stops it from
false-flagging a legitimate non-arXiv (blog / tech-report) citation as fabricated. Use it
in addition to citation-audit, not instead: citation-audit protects your review
score (context-fit); this protects
against the integrity red flag of a wrong or fabricated reference (existence + fields).
|
citation-audit |
citation-crosscheck (this) |
| Owns |
existence + metadata + context-fit |
existence + field accuracy, adversarially |
| Method |
one cross-model reviewer |
two extractors (blind when isolation is attested) + a mechanical field-diff |
| Reads the .tex claims? |
yes |
no (bibliographic only) |
| Deliverable |
KEEP/FIX/REPLACE/REMOVE per entry |
side-by-side .xlsx + MATCH/MINOR/MISMATCH |
| Best at catching |
wrong-context citation |
dropped/re-ordered author, wrong year/venue/DOI |
Constants
- BLIND = conditional — the web agent (B) MUST be prompted with only each paper's title +
first author, plus its opaque
dedup_key (a run-local ordinal that carries no
bibliographic content), and MUST NOT read the .bib or any repo file. B attests
blind:true only when the host isolates its context so it cannot inherit the .bib
(see the mirror-adaptation note above); otherwise B attests blind:false and the Step-3.5
gate certifies nothing (flag-only). Do not relax the prompt restriction, and never assert
blind:true without real isolation — an un-isolated blind:true is the one thing that
defeats the whole cross-check.
- SOURCES = arXiv (
/abs/<id>, /bibtex/<id>), DBLP, CrossRef, ACL Anthology,
NeurIPS/PMLR proceedings, OpenReview. Report what the fetched page shows; never
fabricate a field.
- VERDICTS (assigned by the main agent only, in Step 4 — never by a shard) =
MATCH (bibliographically equivalent, cosmetic-only) / MINOR
(formatting or a preprint-vs-published venue, no correction required) / MISMATCH
(real discrepancy: wrong/missing author, wrong year, wrong venue, wrong id) /
NEEDS_WEBCHECK (B could not confirm existence, but the entry is a
@misc/blog/tech-report whose url B's index-based search structurally cannot see —
the main agent must fetch that url in Step 4 before any verdict; never downgrade
such an entry straight to MISMATCH).
- SCOPE = cited-only by default (entries that render in the
.bbl); --all also
covers uncited .bib entries.
- OUTPUT =
<paper-dir>/citation_crosscheck_<date>.xlsx (+ the four intermediate JSONs —
cx_A_ourbib, cx_B_web, cx_C_fielddiff, and the gate's cx_gate.json — kept for audit).
Workflow
Step 1 — Resolve the citation set (deterministic, main agent)
# Search the paper's own source tree, not the whole repo: bundled author-kit /
# template .tex carry demo \cite{...} that inflate the set. Set PAPER_DIR to the dir of
# the main .tex (\documentclass + \bibliography); recurse (cites often live in
# sections/*.tex); exclude vendored/example dirs.
PAPER_DIR="<dir of the paper's main .tex>"
cd "$PAPER_DIR"
python3 - <<'PY'
import re, pathlib
# Skip vendored/example/template trees: bundled author kits carry demo \cite{...}.
# NOTE: this is a NAME match — a dir you *want* audited must not match (rename it or
# narrow SKIP), e.g. `examples_of_results/` would be skipped.
SKIP = re.compile(r'(?i)(^|/)(.*template.*|.*upstream.*|.*example.*|.*vendor.*)(/|$)')
# Verbatim-ish blocks AND inline \verb|...| / \lstinline are code samples, not citations.
VERB = re.compile(r'\\begin\{(verbatim|lstlisting|minted|Verbatim)\}.*?\\end\{\1\}', re.S)
# The optional arg may itself contain brackets: \lstinline[language={[LaTeX]TeX}]|...|
VERBINLINE = re.compile(r'\\(?:verb|lstinline)\*?'
r'(?:\[(?:[^\[\]]|\[[^\]]*\])*\])?\s*'
r'([^\sA-Za-z])(?:(?!\1).)*\1', re.S)
# url/href/path args may contain a literal % that is NOT a comment. Neutralize them
# BEFORE comment-stripping, or a real \cite later on that line is silently lost.
URLCMD = re.compile(r'\\(?:url|href|path|nolinkurl)\s*\{[^{}]*\}')
COMMENT = re.compile(r'(?m)(?<!\\)%.*$')
# EXPLICIT ALLOWLIST of cite commands. Do NOT build this from a prefix/suffix pattern:
# a Cartesian `(no|paren|auto|...)?cite(s|p|t|...)?` both invents commands (\citecite,
# \parenciteyear -> fake keys from any brace arg) and misses real ones (\footfullcite,
# \volcite, \notecite). Longest-first so \footfullcite wins over \footcite.
# volcite-family: the FIRST brace arg is a volume/page, NOT a cite key.
VOLCITE = {'volcite', 'Volcite', 'pvolcite', 'Pvolcite', 'fvolcite', 'ftvolcite',
'svolcite', 'tvolcite', 'avolcite',
'volcites', 'Volcites', 'pvolcites', 'Pvolcites', 'fvolcites', 'svolcites',
'tvolcites', 'avolcites', 'ftvolcite', 'ftvolcites'}
NAMES = sorted({
# natbib / LaTeX core (capitalized variants are real natbib commands)
'cite', 'Cite', 'citep', 'citet', 'citealp', 'citealt', 'citenum', 'citeyear', 'citeyearpar',
'citeauthor', 'citefullauthor', 'Citep', 'Citet', 'Citealp', 'Citealt', 'Citeauthor',
# biblatex singular + plural
'parencite', 'Parencite', 'footcite', 'footcitetext', 'footcitetexts', 'textcite',
'Textcite', 'smartcite', 'Smartcite', 'supercite', 'autocite', 'Autocite', 'fullcite',
'footfullcite', 'notecite', 'Notecite', 'pnotecite', 'Pnotecite', 'fnotecite',
'citetitle', 'citeurl', 'citedate', 'autocites', 'parencites', 'textcites',
'smartcites', 'supercites', 'footcites', 'cites', 'Cites', 'nocite', 'nocites',
# aliases + field/list forms + bibentry — real biblatex/natbib families
'citetalias', 'citepalias', 'Citetalias', 'Citepalias', 'citefield', 'citelist',
'bibentry', 'fullciter',
*VOLCITE,
}, key=len, reverse=True)
# (?![A-Za-z]) stops \cite matching the \citeXYZ of an unknown command; the trailing run
# accepts any mix of [opt] and {grp} so \cites{k1}[note]{k2} yields BOTH keys.
CITE = re.compile(r'\\(?:' + '|'.join(map(re.escape, NAMES)) + r')\*?(?![A-Za-z])'
r'((?:\s*(?:\[[^\]]*\]|\{(?:[^{}]|\{[^{}]*\})*\}))+)', re.S)
# Strip [optional] args BEFORE reading brace groups, or a braced note like
# \cite[see {Smith} p.3]{k} contributes 'Smith' as a phantom key.
OPTARG = re.compile(r'\[[^\]]*\]', re.S)
GROUP = re.compile(r'\{((?:[^{}]|\{[^{}]*\})*)\}', re.S)
CMDNAME = re.compile(r'\\([A-Za-z]+)')
# \addbibresource[opts]{x.bib} takes an optional arg; \bibliography{a,b} may list several.
BIBRES = re.compile(r'\\(?:bibliography|addbibresource)\s*(?:\[[^\]]*\])?\{([^}]+)\}')
keys, bibs, nocite_all = set(), [], False
for p in sorted(pathlib.Path('.').rglob('*.tex')):
if SKIP.search(str(p.parent)):
continue
src = p.read_text(errors='replace')
src = VERB.sub(' ', src)
src = VERBINLINE.sub(' ', src)
src = URLCMD.sub(' ', src)
src = COMMENT.sub('', src)
for res in BIBRES.findall(src):
bibs += [r.strip() for r in res.split(',') if r.strip()]
for m in CITE.finditer(src):
raw_cmd = CMDNAME.match(m.group(0)).group(1)
cmd = raw_cmd.lower()
grps = GROUP.findall(OPTARG.sub(' ', m.group(1)))
if raw_cmd in VOLCITE and len(grps) > 1:
# volcite takes {volume}{key}; the PLURAL volcites alternates
# {vol}{key}{vol}{key}... — so keep the odd-indexed args, not just grps[1:],
# or the second volume number is harvested as a phantom key.
grps = grps[1::2]
elif raw_cmd in ('citefield', 'Citefield', 'citelist', 'Citelist',
'bibentry') and len(grps) > 1:
# These commands take a single key(s) arg. If a stray second brace group
# follows (a defensive read of an ambiguous source), do not harvest it as a
# phantom key ('\citelist{k}{publisher}' -> just 'k', not '{k, publisher}').
grps = grps[:1]
for grp in grps:
toks = [''.join(g.split()) for g in grp.split(',')]
# \nocite{*} AND \nocite{*,named}: the wildcard renders EVERY bib entry.
if cmd.startswith('nocite') and '*' in toks:
nocite_all = True # see the NOCITE_ALL note below
for k in toks:
if k and k != '*':
keys.add(k)
print(f"BIB_RESOURCE={bibs[0] if bibs else '(none found - use the fallback below)'}")
print(f"BIB_RESOURCES_ALL={bibs}")
print(f"NOCITE_ALL={nocite_all}")
print(f"CITED_KEY_COUNT={len(keys)}")
import pathlib
# Persist the harvest so the shell can read it back deterministically (the heredoc only
# PRINTS to stdout; it cannot set a parent-shell variable).
pathlib.Path('/tmp/cx_cited_keys.txt').write_text('\n'.join(sorted(keys)))
pathlib.Path('/tmp/cx_cited_count.txt').write_text(str(len(keys)))
print('\n'.join(sorted(keys)))
PY
# Set $CITED_KEY_COUNT as a REAL shell variable (Step 3.5 passes it as --expect): the
# heredoc above only prints it, so read it back from the file the harvest just wrote.
CITED_KEY_COUNT=$(cat /tmp/cx_cited_count.txt)
Cross-check against a compiled main.bbl by KEY SET, not count: a count match with a
substituted key still hides a harvest miss (drop one real cite, gain one stray → same count).
Extract the rendered keys and diff the sets:
# \bibitem[label]{key} (natbib/plain) OR \entry{key}{type}{} (biblatex) — key is the first
# brace group either way. Empty diff = harvest and .bbl agree.
grep -oE '\\(bibitem|entry)\*?(\[[^]]*\])?\{[^}]+\}' main.bbl 2>/dev/null \
| sed -E 's/.*\{([^}]+)\}$/\1/' | sort -u > /tmp/cx_bbl_keys.txt
comm -3 <(sort -u /tmp/cx_cited_keys.txt) /tmp/cx_bbl_keys.txt # any output ⇒ investigate
The .bbl cross-check is conditional — an uncompiled submission has none. When absent,
confirm CITED_KEY_COUNT is plausible for the paper's length before proceeding; the
Step-3.5 manifest is the load-bearing coverage guard either way.
If NOCITE_ALL=True the .tex harvest is NOT the citation set. \nocite{*} renders
every entry in the .bib, so under the default cited-only SCOPE the set becomes all
entries that render: take the keys from the compiled .bbl when one exists, else from the
.bib itself (equivalent to --all), and say so in the report. Do not proceed on the
\cite-only harvest — it under-audits by exactly the uncited entries.
BIB_RESOURCES_ALL lists every \bibliography/\addbibresource found, in sorted-path
order; BIB_RESOURCE is just the first. If the list has more than one entry, pick the one
the main .tex declares rather than trusting the ordering, and audit the union only if
the paper really loads several. If BIB_RESOURCE comes back empty (e.g. a \input-flattened
submission), fall back to the largest *.bib under the paper dir and, when a compiled
main.bbl exists, read the cited keys from its \bibitem{...} entries instead of the .tex
(the .bbl is the ground truth for what actually rendered).
Produce the ordered list of cited keys and, for each, its title + first author (the
only thing agent B is allowed to see). Confirm every key exists in the .bib. Assign each
entry a stable opaque ordinal dedup_key (01, 02, … in resolved order): this is the
canonical id that joins the shards downstream. A receives dedup_key + the cite key; B
receives dedup_key + title + first author only (never the cite key or .bib text), so
the join key never leaks the blind side. Write two artifacts for the Step-3.5 gate, in
resolved order: /tmp/cx_dedup_keys.txt (the ordinal set, whitespace-separated) and
/tmp/cx_keymap.tsv (one dedup_key<TAB>cite_key line per entry — the ordinal→cite-key
pin). Step 3.5 builds a coverage manifest (ordinal → cite key → title hash) from
/tmp/cx_keymap.tsv + the .bib and verifies A against it as a mapping, so a citation
dropped or substituted downstream fails the run — a bare count or ordinal-set check would
pass a substitution that keeps the count.
Step 2 — Run A and B blind to each other
This is a read-only extraction fan-out in the sense of
../shared-references/fan-out-pattern.md: A
and B are two shards that only extract records (each returns an entries[] set keyed
by dedup_key — never by list position, never a B-visible cite key, never a quality
ranking). Each shard returns the {shard_id, entries:[{dedup_key, …}]} envelope of
../shared-references/fan-out-pattern.md, and the
executor runs the mechanical dedup_key join below before C. No shard assigns a
MATCH/MINOR/MISMATCH/NEEDS_WEBCHECK label at all — the main agent assigns all
four in Step 4 from C's field evidence, and only its own web-confirm edits the .bib.
Dispatch (tier-portable). On this host there is no Claude Agent tool, so the mainline's
Tier 1 (Workflow-concurrent) and Tier 2 (Agent) forms both degrade here: run A, then B, then
C one after the other, each in the freshest context the host can give. The join and the
field-diff are identical to the mainline. Blindness is not automatic here (see the mirror
adaptation note above): B sets blind:true only if its context provably did not inherit the
.bib, and blind:false otherwise — in which case Step 3.5 downgrades every accept to
ESCALATE and the run flags rather than certifies. B is prompted with only the
title+first-author list and is forbidden from reading any local file; but when B is not run in
an isolated context, that prohibition is not enforced (B could observe the parent's .bib
read, or A's output), which is exactly why an un-isolated B must attest blind:false and the
gate then certifies nothing. Isolation is the guarantee; the prompt restriction alone is not.
Breadth. The shard set is fixed at three by the design, not by a budget: two records are
the minimum for a double-blind diff (one per provenance), and the third is the differ that must
have produced neither. Breadth is not a swept parameter here, so there is no breadth/cost
trade-off to bound — but note the ceiling: two independent records, not an ensemble.
Acceptance gate (provenance, per
../shared-references/acceptance-gate.md). This
skill is not a loop and has no self-terminating accept gate; it runs once over a fixed
citation set. Type-A (executor may self-judge): the dedup_key join balanced, every entry
labelled, the .xlsx written. Type-B (merit — is the entry bibliographically correct):
never delegated to a shard. A MISMATCH or a NEEDS_WEBCHECK resolution requires the main
agent's own primary-source fetch — no fetch, no finding. A MATCH/MINOR rests on the
two independent records agreeing, without a third fetch; that is weaker evidence than a
fetch, so read a no-flag MATCH as "two sources concur", not "verified against the
publisher of record".
Requirement (4) is discharged by the deterministic branch, not by a jury.
fan-out-pattern.md admits "a single cross-model jury step — OR a deterministic verifier
gate"; this skill takes the second: scripts/cx_verify.py (Step 3.5) is an external process
that decides MATCH/MINOR by declared normalization and fails closed to ESCALATE
otherwise — the same script and the same verdicts on this host as on the mainline.
The limit, stated plainly: the accept path is mechanical and a MISMATCH needs the main
agent's own primary-source fetch, but the executor still reads that page, so this skill does
not deliver cross-model acquittal and does not replace one — which is why it is meant to
run in addition to citation-audit (a single cross-model reviewer), not instead of it.
Never report a MISMATCH, least of all a fabrication finding, on shard text alone.
- Agent A (repo-only): read the
.bib; extract each cited entry verbatim
(brace-matched, no reformatting) → /tmp/cx_A_ourbib.json
{"shard_id":"A","entries":[{"dedup_key","key","bibtex"}]}. Explicit instruction: do not
search the web.
- Agent B (web-only): given ONLY the title+first-author list from Step 1, re-fetch
each paper's record from the web and emit a clean BibTeX per paper →
/tmp/cx_B_web.json
{"shard_id":"B","blind","entries":[{"dedup_key","title_queried","found","source_url","bibtex","sources_tried","note"}]}.
Explicit instruction: do not read any local file / .bib; set found:false + a note
if a paper cannot be confirmed; never fabricate a field.
blind attestation (set once, for the whole B shard). B sets blind:true only
if it ran in a context that could not have seen the .bib — a genuinely isolated
dispatch that did not inherit the main agent's history. If B cannot guarantee that
(e.g. it ran in the main context), it MUST set blind:false. The Step-3.5 gate refuses
to certify a MATCH/MINOR on any entry that is not blind-attested, so a false
blind:true is the one thing that would defeat the whole cross-check — never assert it
to "pass".
- Do not restrict the search to arXiv/DBLP. Many legitimate references are workshop
tech reports, company research blogs, or proceedings pages that never appear on arXiv.
If a title is not on arXiv, B MUST also run a general web search and open the landing
page its own search returns (a project / blog / proceedings URL) before concluding
found:false. B works from the title+first-author query only — it must NOT be handed,
and cannot use, the .bib entry's url field (that stays with the main agent's Step-4
webcheck). A citation being absent from arXiv is NOT evidence it is fake.
- Record
sources_tried per entry (e.g.
["arxiv","crossref","semantic-scholar(429)","web"]) so a found:false from a
degraded search (a source was down or rate-limited) is distinguishable from a genuine
not-found. Note any source that failed/blocked in the entry's note.
On return, the main agent joins A and B on dedup_key — every ordinal from Step 1 must
appear exactly once on each side. A missing or duplicated dedup_key, or an unbalanced count,
is a fail-closed stop (re-run the offending shard), never a silent drop or a positional
guess. This mechanical join replaces order-coupling and is judgment-free — it decides nothing
about any citation, only that the two extractions line up before C diffs them.
Step 3 — Run C, the field-diff extractor (after A+B return)
C extracts, it does not adjudicate. Computing a field-by-field delta is mechanical
extraction; deciding that an entry is MATCH (i.e. admitting it as correct) is the
acceptance verdict, which ../shared-references/fan-out-pattern.md
reserves for the executor. So C emits per-field evidence, not a per-entry verdict:
for each field it reports same / differs (with both values verbatim) / absent, plus
the routing flags below. The main agent assigns every final
MATCH|MINOR|MISMATCH|NEEDS_WEBCHECK in Step 4 from that evidence. A shard that returns
a verdict label is out of contract — the same rule proof-checker and research-lit
state for their shards.
A third pass (a fresh-context run — this host grants no delegation tool) reads both JSONs — it
produced neither — and, on each dedup_key-joined pair, diffs the two records field by
field: title (ignore case/braces),
author set AND order, year, venue (NeurIPS ≡ "Advances in NIPS"; note a genuinely
different event), volume/pages, arXiv id / DOI. Output /tmp/cx_C_fielddiff.json:
{"shard_id":"C","entries":[{"dedup_key","key","blind","found","sources_tried","url","fields":{"<field>":{"status":"same|differs|absent","ours","web"}},"flags":[...]}]}
(found/sources_tried/blind copied verbatim from B, url from A, so Step 4 and the
gate can act without
re-reading the shard files)
(keyed by the joined dedup_key) plus a summary
{entries, fields_by_status:{same,differs,absent}, flags_by_type} — counts of mechanical
field states, not a verdict tally.
Critical routing flag for C (this prevents the most damaging false positive): if B
reports found:false for an entry whose A-side .bib type is @misc (or otherwise
carries a url/howpublished instead of an arXiv id / DOI), and B's sources_tried
shows it only queried indexers that cannot see that URL (arXiv/DBLP/etc.), C MUST attach
the flag url_only_source_unqueried — and MUST NOT report the entry as absent-from-web.
B's index-based method structurally cannot confirm a blog/tech-report citation; only the
main agent's direct url fetch in Step 4 can. C likewise attaches degraded_search when
sources_tried shows a source was down or rate-limited, and web_record_absent whenever B
returned found:false for any reason. These flags force the entry into the main agent's Step-4
queue; none of them is a finding about the citation — they record what B's search did, not
whether the reference is sound.
Step 3.5 — Run the deterministic verifier gate (cx_verify.py, mandatory)
This is the fan-out's requirement-(4) gate: fan-out-pattern.md demands "a single
cross-model jury step — OR a deterministic verifier gate — that is identical across all
three tiers." This skill has no cross-model jury, so the gate is an external process. It is the
same script and the same verdicts on this host as on the mainline — the dispatch of A/B/C
degrades by tier, the gate does not. The gate recomputes both the field diff and the routing
flags from A's and B's own raw records and reads B's blindness attestation from B directly;
C's flags are advisory only (unioned with the recompute) and a C schema violation fails
closed — so C is not in the trust path for the diff or the routing. Resolve $CX_VERIFY
with the Codex hybrid chain
(managed project install, in-repo checkout, ARIS repo fallback, then the copied global Codex
skill install). This is Policy A — skill-local gate:
Resolver note (owner-local, per ../shared-references/integration-contract.md).
cx_verify.py is a single-skill helper under this skill's own scripts/, resolved via the
Codex hybrid chain below. It deliberately ships no tools/cx_verify.py shim — it has no
legacy tools/ entry to forward from (unlike the Phase-3 moves that kept one) — which is the
owner-local Layer-0 exception the contract's Policy-A row registers.
CX_VERIFY=""
cd "$(git rev-parse --show-toplevel 2>/dev/null || pwd)" || exit 1
if [ -z "${ARIS_REPO:-}" ] && [ -f .aris/installed-skills-codex.txt ]; then
ARIS_REPO=$(awk -F'\t' '$1=="repo_root"{print $2; exit}' .aris/installed-skills-codex.txt 2>/dev/null) || true
fi
# Global pointer written by smart_update_codex.sh, for hosts with no per-project install.
if [ -z "${ARIS_REPO:-}" ] && [ -f "$HOME/.aris/repo" ]; then
ARIS_REPO=$(cat "$HOME/.aris/repo" 2>/dev/null) || true
fi
[ -f ".agents/skills/citation-crosscheck/scripts/cx_verify.py" ] && CX_VERIFY=".agents/skills/citation-crosscheck/scripts/cx_verify.py"
[ -z "$CX_VERIFY" ] && [ -f "skills/citation-crosscheck/scripts/cx_verify.py" ] && CX_VERIFY="skills/citation-crosscheck/scripts/cx_verify.py"
[ -z "$CX_VERIFY" ] && [ -n "${ARIS_REPO:-}" ] && [ -f "$ARIS_REPO/skills/citation-crosscheck/scripts/cx_verify.py" ] && CX_VERIFY="$ARIS_REPO/skills/citation-crosscheck/scripts/cx_verify.py"
[ -z "$CX_VERIFY" ] && [ -f ~/.codex/skills/citation-crosscheck/scripts/cx_verify.py ] && CX_VERIFY="$HOME/.codex/skills/citation-crosscheck/scripts/cx_verify.py"
[ -z "$CX_VERIFY" ] && {
echo "ERROR: cx_verify.py not resolved at .agents/skills/citation-crosscheck/scripts/, skills/citation-crosscheck/scripts/, \$ARIS_REPO/skills/citation-crosscheck/scripts/, or ~/.codex/skills/citation-crosscheck/scripts/." >&2
echo " The requirement-(4) gate cannot run; do NOT hand-assign MATCH/MINOR in its place. Fix: rerun install_aris_codex.sh, export ARIS_REPO, or copy the citation-crosscheck skill into ~/.codex/skills/." >&2
exit 1
}
python3 "$CX_VERIFY" --selftest || exit 1 # built-in cases; fail = do not trust the gate
# Build the coverage manifest (ordinal → cite key → title hash) from Step 1's ordinal PIN
# (/tmp/cx_keymap.tsv) + the .bib, using the gate's OWN parser. This is independent of A, so
# the gate can verify A against it as a MAPPING. ($BIB = the resolved .bib path from Step 1.)
python3 "$CX_VERIFY" --build-manifest --bib "$BIB" --keymap /tmp/cx_keymap.tsv \
--out-manifest /tmp/cx_manifest.tsv || exit 1
# The gate RECOMPUTES the field diff AND the routing flags from A's and B's own raw records
# (`web_record_absent` from B's `found`, `degraded_search` / `url_only_source_unqueried` from
# B's `sources_tried` + A's fields), and reads B's blindness attestation from B directly. C's
# flags are ADVISORY only, unioned with the recompute, and a C schema violation fails closed —
# so a faulty/omitting/adversarial C can neither hide a discrepancy, forge a MATCH, nor wave an
# entry through by omitting a flag. Coverage is MANDATORY: `--expect-manifest` verifies the
# ordinal→key/title MAPPING against A (catching a drop-and-substitute the ordinal-set check
# cannot); `--expect` / `--expect-keys` add the count and ordinal-set checks. An empty artifact,
# a duplicate/absent/substituted key, a dropped citation, or A/B/C key-sets that disagree all
# exit non-zero rather than reading as "clean".
python3 "$CX_VERIFY" --a /tmp/cx_A_ourbib.json --b /tmp/cx_B_web.json --c /tmp/cx_C_fielddiff.json \
--expect "$CITED_KEY_COUNT" --expect-keys /tmp/cx_dedup_keys.txt \
--expect-manifest /tmp/cx_manifest.tsv \
--json /tmp/cx_gate.json || exit 1
What the gate decides, and what it refuses to decide:
MATCH — every compared field equal after declared normalization (LaTeX accent/brace
folding, case, punctuation; an order-preserving surname list; a fixed venue-alias table).
MINOR — the sole difference is preprint-vs-published venue for the same work.
ESCALATE — everything else, fail-closed: any identifying field differing
(authors, year, title, doi, arxiv), a secondary field differing, an identifying
field absent on one side, a venue pair not in the declared table, any routing flag,
found != true, or an entry that is not blind-attested (blind != true). A
MATCH/MINOR that lacks the blindness attestation is downgraded to ESCALATE: the run
can still flag a discrepancy, but it may not certify a match. This is what enforces the
double-blind claim mechanically — a host that cannot dispatch B in isolation (so B might
have seen the .bib) sets blind:false and the gate refuses to certify anything, turning
the check into a labelled heuristic rather than a false certification.
The gate never emits MISMATCH. A discrepancy becomes a finding only when the main agent
reproduces it against the publisher of record, so the accept path is mechanical and the reject
path is externally evidenced. Author order is preserved by the normalizer on purpose — a
re-ordered author list is one of the defects this skill exists to catch, so it must not
normalize to equal.
Step 4 — ASSIGN every verdict (main agent, mandatory)
No verdict arrives pre-assigned from a shard. Read /tmp/cx_gate.json (the gate's recomputed
verdicts and per-field evidence) — C's cx_C_fielddiff.json is audit-only and must not
override the gate. The invariant of this step: every MATCH/MINOR/MISMATCH rests on
either (a) the gate's mechanical decision or (b) a completed main-agent primary-source fetch.
No shard signal — a differs, an absent, B's found:false, or a routing flag — is itself a
finding. Two exhaustive cases:
Gate said MATCH or MINOR → adopt it (case (a), mechanical). Do not re-litigate a
mechanical equality decision from prose, and do not silently upgrade it to MISMATCH; if you
believe the gate is wrong, the fix is the gate's normalization table plus a --selftest case,
not a hand-edit of one row. (What a no-flag MATCH means: two independently-produced records
concur — not "verified against the publisher of record".)
Gate said ESCALATE → the main agent MUST complete a primary-source fetch before it
may assign any of MATCH/MINOR/MISMATCH/NEEDS_WEBCHECK. This is a procedure, not a
first-match menu: do every applicable fetch, then decide — so no earlier branch can
preempt a stricter one.
Fetch. Identify and WebFetch the primary source(s):
- the entry's own
url / howpublished whenever the .bib carries one — always for the
url_only_source_unqueried / web_record_absent cases (this is the .bib's url,
which B never saw); and
- the canonical index the fields point at — the arXiv abstract for an arXiv id, the DOI
resolver for a DOI, else the venue / proceedings page.
For an unconfirmed-existence entry, also run the main agent's own general web search
(independent of B). Record every URL and query tried.
Decide from what the fetch(es) returned — each terminal rests on a completed fetch:
- Primary source confirms the
.bib (title + author set/order + year + id agree) →
MATCH (annotate the source; for a url-confirmed non-arXiv entry note
"blog/tech-report, web-confirmed"). A gate differs/absent the source does not
reproduce is downgraded here, with the reason recorded.
- Same work, only preprint-vs-published differs →
MINOR.
- Primary source contradicts an identifying field (author set/order, year,
venue-as-different-event, arXiv id / DOI, or a substantive title/subtitle/word change;
or a moved page/volume range) →
MISMATCH, citing the fetched value. (Field-discrepancy
MISMATCH — the source reproduced it; this guards against a B/C hallucination inventing
a wrong "fix.")
- Existence unconfirmable →
MISMATCH (fabrication) only when BOTH hold: the
entry's url is absent / dead / contradicting, and the main agent's own general
search also failed. A dead url alone, or B's found:false alone, is never enough
— this is the one path to a fabrication finding and it rests on the main agent's own null
result, not on B's miss.
- Fetch inconclusive (source unreachable / genuinely ambiguous) →
NEEDS_WEBCHECK,
left for a human. Never default an inconclusive fetch to MATCH or MISMATCH.
A one-sided absent field is not resolved "from the fields both sides share": the gate
already escalated it, and the verdict comes from the fetched primary source per the ladder
above (an absence becomes MISMATCH only when the .bib asserts a field the source
contradicts; otherwise MATCH/MINOR per the fetch, with the absence noted in Detail).
Every entry leaves Step 4 with exactly one of the four labels, each MATCH/MINOR/
MISMATCH backed by the gate or a completed main-agent fetch. If an ESCALATE's evidence
fits none of the outcomes above, that is a fetch not yet done, not licence for a shard to
decide. Do not edit the .bib on shard evidence alone: every applied fix traces to a
main-agent fetch.
Step 5 — Emit the spreadsheet + apply confirmed fixes
Build the .xlsx (openpyxl) with one row per citation and columns:
# | BibKey | Link | Our BibTeX | Web BibTeX | B source | Verdict | Author check | Year | Detail | Fix. Color the Verdict cell: green = MATCH, yellow = MINOR or a
NEEDS_WEBCHECK the main agent resolved to MATCH (annotate how), orange = MISMATCH.
Add a Summary sheet with the counts and the fixed/optional lists. Apply only the main-agent-confirmed MISMATCH
fixes to the .bib; if the paper is compiled, rebuild so the .bbl picks them up and
re-verify the page count is unchanged. Report MINOR items as optional (do not auto-apply
preprint→published upgrades; that is an author choice).
Key rules
- Blindness is the mechanism. If B ever sees the
.bib, the cross-check collapses
into a single-source read and loses its power. Enforce it in the prompt.
- Never fabricate a field in A, B, or C.
found:false + a note beats a guess.
- A non-arXiv miss is not a fabrication.
@misc/blog/tech-report/proceedings-URL
citations are legitimate and invisible to arXiv/DBLP search. B must try the general web
from the title+first-author prompt only; C flags an unconfirmed URL-bearing entry
(url_only_source_unqueried) and the main agent assigns NEEDS_WEBCHECK and fetches the
entry's own url in Step 4. A MISMATCH (fabrication) requires both a dead or
contradicting url and the main agent's own general web search also failing (per Step 4) —
a dead url alone is never enough. This is the skill's most dangerous false-positive if skipped.
- Shards extract; the main agent adjudicates. A, B and C emit records and per-field
deltas. Every
MATCH/MINOR/MISMATCH/NEEDS_WEBCHECK is assigned by the main agent in
Step 4, and every MISMATCH is web-confirmed by the main agent before the .bib is
touched. The shards propose; the main agent's own fetch disposes.
- Bibliographic only. Context-fit ("is it cited for a claim the paper supports?") is
out of scope — that is
citation-audit's job. Recommend running both before submission.
- Report, do not silently upgrade. Preprint-vs-published and cosmetic key/year-label
mismatches are MINOR; list them, leave the choice to the author.
1---2name: citation-crosscheck-23description: "Conditionally double-blind bibliographic cross-check (double-blind only when the host can isolate the web agent; a flag-only heuristic when isolation can't be attested): one agent extracts each cited entry verbatim from the local .bib, a second independently re-fetches the same paper's record from the web while blind to the .bib, and a third extractor emits a mechanical field-by-field delta to catch dropped/re-ordered authors, wrong years, wrong venues, and bad DOIs/arXiv ids. The main agent (never a shard) assigns the MATCH/MINOR/MISMATCH verdict per reference, and exports a side-by-side spreadsheet. Use when user says \"citation crosscheck\", \"double-blind citation check\", \"verify bib against scholar\", \"bib field audit\", \"引用交叉核对\", or before submission when bibliography FIELD accuracy (not just existence) must be certified."4argument-hint: "[paper-directory-or-bib-file] [--xlsx <path>] [--cited-only|--all]"5allowed-tools: Bash(*), Read, Grep, Glob, Write, WebFetch, WebSearch6metadata:7 source: https://skillmd.com/skills/wanshuiyin/citation-crosscheck-28---910# Citation Cross-Check (conditionally double-blind bib-vs-web field audit)1112> **Codex mirror adaptation (normative).** This skill has no cross-model jury *by13> design* — it is an evidence instrument, not an acquittal step (see the acceptance-gate14> note in Step 2, and pair it with `citation-audit` for the cross-model verdict). A and B15> are read-only extraction shards and C emits a mechanical per-field delta, so every16> verdict is assigned by the main agent from that delta plus its own primary-source17> fetch — not by any shard and not by a reviewer model.18>19> **Blindness is NOT automatically identical across hosts — it is *conditional on20> isolation*, and the gate enforces that.** The whole cross-check rests on B never seeing21> the `.bib`. On the mainline, the `Agent` tool gives B a fresh context. On a Codex host22> the frontmatter grants **no delegation tool** (no `Agent`, no `spawn_agent` — see23> `allowed-tools`), so:24> - **If** the host can dispatch B in a genuinely isolated context that does **not** inherit25> the main agent's history (which has already read the `.bib` in Step 1) — passing B only26> the title+first-author list — B may set `blind:true` and the run certifies normally.27> - **Otherwise** (B would run in the main context, or "fresh sequential context" cannot be28> guaranteed to drop the parent's `.bib` view), B MUST set `blind:false`. The Step-3.529> gate then **downgrades every `MATCH`/`MINOR` to `ESCALATE`**: the run still *flags*30> discrepancies for the main agent's Step-4 webcheck, but it *certifies nothing*. It is a31> labelled heuristic, not a double-blind certification.32>33> So the mirror never *claims* blindness it cannot deliver: it either has real isolation and34> attests it, or it declares `blind:false` and drops to flag-only. Do not add a35> `spawn_agent`-based reviewer/continuation route — the trust decision is the main agent's36> `WebFetch`, and it must not be delegated to a shard.3738> 🔒 **Do not wrap this skill in `/loop`, `/schedule`, or `CronCreate`.** It is39> verdict-bearing — it emits a per-reference verdict that gates `.bib` edits (an evidence40> instrument, not a cross-model acquittal; see the acceptance-gate note in Step 2).41> Re-running it on a timer adds no signal (the verdict changes only when the42> *bibliography* changes).43> Schedule the external wait that precedes it: bibliography finalized → run once. See44> [`../shared-references/external-cadence.md`](../shared-references/external-cadence.md).4546## Why this exists (and how it differs from `citation-audit`)4748`citation-audit` is a single cross-model reviewer that judges three things holistically:49existence, metadata, and **context-fit** (is the cited paper used for a claim it50actually supports?). This skill does **not** do context-fit. It does one thing that a51single reviewer cannot do reliably: an **adversarial, field-level diff** (double-blind when the host can isolate the web agent, a labelled flag-only heuristic otherwise)52of the local `.bib` entry against an *independently reconstructed* record.5354The failure mode it targets is the hand-entry error that survives a plausibility read:55a **dropped or re-ordered author**, a **wrong year**, a **swapped venue**, a **DOI that56404s**. A single reviewer that sees the `.bib` entry tends to anchor on it and rate it57"looks correct." Two agents that never see each other's output cannot anchor — a58discrepancy only survives if *both* independent records agree, and the verifier reports59exactly where they diverge.6061In practice it catches dropped/re-ordered authors, wrong years, and bad DOIs that a62single-pass audit rated "looks clean" — while `NEEDS_WEBCHECK` stops it from63false-flagging a legitimate non-arXiv (blog / tech-report) citation as fabricated. Use it64**in addition to** `citation-audit`, not instead: `citation-audit` protects your *review65score* (context-fit); this protects66against the *integrity red flag* of a wrong or fabricated reference (existence + fields).6768| | `citation-audit` | `citation-crosscheck` (this) |69|---|---|---|70| Owns | existence + metadata + **context-fit** | **existence + field accuracy**, adversarially |71| Method | one cross-model reviewer | **two extractors (blind when isolation is attested) + a mechanical field-diff** |72| Reads the .tex claims? | yes | no (bibliographic only) |73| Deliverable | KEEP/FIX/REPLACE/REMOVE per entry | **side-by-side .xlsx** + MATCH/MINOR/MISMATCH |74| Best at catching | wrong-context citation | dropped/re-ordered author, wrong year/venue/DOI |7576## Constants7778- **BLIND = conditional** — the web agent (B) MUST be prompted with only each paper's title +79 first author, plus its opaque `dedup_key` (a run-local ordinal that carries no80 bibliographic content), and MUST NOT read the `.bib` or any repo file. B attests81 `blind:true` **only** when the host isolates its context so it cannot inherit the `.bib`82 (see the mirror-adaptation note above); otherwise B attests `blind:false` and the Step-3.583 gate certifies nothing (flag-only). Do not relax the prompt restriction, and never assert84 `blind:true` without real isolation — an un-isolated `blind:true` is the one thing that85 defeats the whole cross-check.86- **SOURCES** = arXiv (`/abs/<id>`, `/bibtex/<id>`), DBLP, CrossRef, ACL Anthology,87 NeurIPS/PMLR proceedings, OpenReview. Report what the fetched page shows; never88 fabricate a field.89- **VERDICTS** (assigned by the **main agent only**, in Step 4 — never by a shard) =90 `MATCH` (bibliographically equivalent, cosmetic-only) / `MINOR`91 (formatting or a preprint-vs-published venue, no correction required) / `MISMATCH`92 (real discrepancy: wrong/missing author, wrong year, wrong venue, wrong id) /93 `NEEDS_WEBCHECK` (B could not confirm existence, but the entry is a94 `@misc`/blog/tech-report whose `url` B's index-based search structurally cannot see —95 the main agent must fetch that `url` in Step 4 before any verdict; **never** downgrade96 such an entry straight to `MISMATCH`).97- **SCOPE** = cited-only by default (entries that render in the `.bbl`); `--all` also98 covers uncited `.bib` entries.99- **OUTPUT** = `<paper-dir>/citation_crosscheck_<date>.xlsx` (+ the four intermediate JSONs —100 `cx_A_ourbib`, `cx_B_web`, `cx_C_fielddiff`, and the gate's `cx_gate.json` — kept for audit).101102## Workflow103104### Step 1 — Resolve the citation set (deterministic, main agent)105106```bash107# Search the paper's own source tree, not the whole repo: bundled author-kit /108# template .tex carry demo \cite{...} that inflate the set. Set PAPER_DIR to the dir of109# the main .tex (\documentclass + \bibliography); recurse (cites often live in110# sections/*.tex); exclude vendored/example dirs.111PAPER_DIR="<dir of the paper's main .tex>"112cd "$PAPER_DIR"113python3 - <<'PY'114import re, pathlib115# Skip vendored/example/template trees: bundled author kits carry demo \cite{...}.116# NOTE: this is a NAME match — a dir you *want* audited must not match (rename it or117# narrow SKIP), e.g. `examples_of_results/` would be skipped.118SKIP = re.compile(r'(?i)(^|/)(.*template.*|.*upstream.*|.*example.*|.*vendor.*)(/|$)')119# Verbatim-ish blocks AND inline \verb|...| / \lstinline are code samples, not citations.120VERB = re.compile(r'\\begin\{(verbatim|lstlisting|minted|Verbatim)\}.*?\\end\{\1\}', re.S)121# The optional arg may itself contain brackets: \lstinline[language={[LaTeX]TeX}]|...|122VERBINLINE = re.compile(r'\\(?:verb|lstinline)\*?'123 r'(?:\[(?:[^\[\]]|\[[^\]]*\])*\])?\s*'124 r'([^\sA-Za-z])(?:(?!\1).)*\1', re.S)125# url/href/path args may contain a literal % that is NOT a comment. Neutralize them126# BEFORE comment-stripping, or a real \cite later on that line is silently lost.127URLCMD = re.compile(r'\\(?:url|href|path|nolinkurl)\s*\{[^{}]*\}')128COMMENT = re.compile(r'(?m)(?<!\\)%.*$')129# EXPLICIT ALLOWLIST of cite commands. Do NOT build this from a prefix/suffix pattern:130# a Cartesian `(no|paren|auto|...)?cite(s|p|t|...)?` both invents commands (\citecite,131# \parenciteyear -> fake keys from any brace arg) and misses real ones (\footfullcite,132# \volcite, \notecite). Longest-first so \footfullcite wins over \footcite.133# volcite-family: the FIRST brace arg is a volume/page, NOT a cite key.134VOLCITE = {'volcite', 'Volcite', 'pvolcite', 'Pvolcite', 'fvolcite', 'ftvolcite',135 'svolcite', 'tvolcite', 'avolcite',136 'volcites', 'Volcites', 'pvolcites', 'Pvolcites', 'fvolcites', 'svolcites',137 'tvolcites', 'avolcites', 'ftvolcite', 'ftvolcites'}138NAMES = sorted({139 # natbib / LaTeX core (capitalized variants are real natbib commands)140 'cite', 'Cite', 'citep', 'citet', 'citealp', 'citealt', 'citenum', 'citeyear', 'citeyearpar',141 'citeauthor', 'citefullauthor', 'Citep', 'Citet', 'Citealp', 'Citealt', 'Citeauthor',142 # biblatex singular + plural143 'parencite', 'Parencite', 'footcite', 'footcitetext', 'footcitetexts', 'textcite',144 'Textcite', 'smartcite', 'Smartcite', 'supercite', 'autocite', 'Autocite', 'fullcite',145 'footfullcite', 'notecite', 'Notecite', 'pnotecite', 'Pnotecite', 'fnotecite',146 'citetitle', 'citeurl', 'citedate', 'autocites', 'parencites', 'textcites',147 'smartcites', 'supercites', 'footcites', 'cites', 'Cites', 'nocite', 'nocites',148 # aliases + field/list forms + bibentry — real biblatex/natbib families149 'citetalias', 'citepalias', 'Citetalias', 'Citepalias', 'citefield', 'citelist',150 'bibentry', 'fullciter',151 *VOLCITE,152}, key=len, reverse=True)153# (?![A-Za-z]) stops \cite matching the \citeXYZ of an unknown command; the trailing run154# accepts any mix of [opt] and {grp} so \cites{k1}[note]{k2} yields BOTH keys.155CITE = re.compile(r'\\(?:' + '|'.join(map(re.escape, NAMES)) + r')\*?(?![A-Za-z])'156 r'((?:\s*(?:\[[^\]]*\]|\{(?:[^{}]|\{[^{}]*\})*\}))+)', re.S)157# Strip [optional] args BEFORE reading brace groups, or a braced note like158# \cite[see {Smith} p.3]{k} contributes 'Smith' as a phantom key.159OPTARG = re.compile(r'\[[^\]]*\]', re.S)160GROUP = re.compile(r'\{((?:[^{}]|\{[^{}]*\})*)\}', re.S)161CMDNAME = re.compile(r'\\([A-Za-z]+)')162# \addbibresource[opts]{x.bib} takes an optional arg; \bibliography{a,b} may list several.163BIBRES = re.compile(r'\\(?:bibliography|addbibresource)\s*(?:\[[^\]]*\])?\{([^}]+)\}')164keys, bibs, nocite_all = set(), [], False165for p in sorted(pathlib.Path('.').rglob('*.tex')):166 if SKIP.search(str(p.parent)):167 continue168 src = p.read_text(errors='replace')169 src = VERB.sub(' ', src)170 src = VERBINLINE.sub(' ', src)171 src = URLCMD.sub(' ', src)172 src = COMMENT.sub('', src)173 for res in BIBRES.findall(src):174 bibs += [r.strip() for r in res.split(',') if r.strip()]175 for m in CITE.finditer(src):176 raw_cmd = CMDNAME.match(m.group(0)).group(1)177 cmd = raw_cmd.lower()178 grps = GROUP.findall(OPTARG.sub(' ', m.group(1)))179 if raw_cmd in VOLCITE and len(grps) > 1:180 # volcite takes {volume}{key}; the PLURAL volcites alternates181 # {vol}{key}{vol}{key}... — so keep the odd-indexed args, not just grps[1:],182 # or the second volume number is harvested as a phantom key.183 grps = grps[1::2]184 elif raw_cmd in ('citefield', 'Citefield', 'citelist', 'Citelist',185 'bibentry') and len(grps) > 1:186 # These commands take a single key(s) arg. If a stray second brace group187 # follows (a defensive read of an ambiguous source), do not harvest it as a188 # phantom key ('\citelist{k}{publisher}' -> just 'k', not '{k, publisher}').189 grps = grps[:1]190 for grp in grps:191 toks = [''.join(g.split()) for g in grp.split(',')]192 # \nocite{*} AND \nocite{*,named}: the wildcard renders EVERY bib entry.193 if cmd.startswith('nocite') and '*' in toks:194 nocite_all = True # see the NOCITE_ALL note below195 for k in toks:196 if k and k != '*':197 keys.add(k)198print(f"BIB_RESOURCE={bibs[0] if bibs else '(none found - use the fallback below)'}")199print(f"BIB_RESOURCES_ALL={bibs}")200print(f"NOCITE_ALL={nocite_all}")201print(f"CITED_KEY_COUNT={len(keys)}")202import pathlib203# Persist the harvest so the shell can read it back deterministically (the heredoc only204# PRINTS to stdout; it cannot set a parent-shell variable).205pathlib.Path('/tmp/cx_cited_keys.txt').write_text('\n'.join(sorted(keys)))206pathlib.Path('/tmp/cx_cited_count.txt').write_text(str(len(keys)))207print('\n'.join(sorted(keys)))208PY209# Set $CITED_KEY_COUNT as a REAL shell variable (Step 3.5 passes it as --expect): the210# heredoc above only prints it, so read it back from the file the harvest just wrote.211CITED_KEY_COUNT=$(cat /tmp/cx_cited_count.txt)212```213Cross-check against a compiled `main.bbl` by **KEY SET, not count**: a count match with a214substituted key still hides a harvest miss (drop one real cite, gain one stray → same count).215Extract the rendered keys and diff the sets:216```bash217# \bibitem[label]{key} (natbib/plain) OR \entry{key}{type}{} (biblatex) — key is the first218# brace group either way. Empty diff = harvest and .bbl agree.219grep -oE '\\(bibitem|entry)\*?(\[[^]]*\])?\{[^}]+\}' main.bbl 2>/dev/null \220 | sed -E 's/.*\{([^}]+)\}$/\1/' | sort -u > /tmp/cx_bbl_keys.txt221comm -3 <(sort -u /tmp/cx_cited_keys.txt) /tmp/cx_bbl_keys.txt # any output ⇒ investigate222```223**The `.bbl` cross-check is conditional — an uncompiled submission has none. When absent,224confirm `CITED_KEY_COUNT` is plausible for the paper's length before proceeding; the225Step-3.5 manifest is the load-bearing coverage guard either way.**226227**If `NOCITE_ALL=True` the `.tex` harvest is NOT the citation set.** `\nocite{*}` renders228*every* entry in the `.bib`, so under the default cited-only SCOPE the set becomes all229entries that render: take the keys from the compiled `.bbl` when one exists, else from the230`.bib` itself (equivalent to `--all`), and say so in the report. Do not proceed on the231`\cite`-only harvest — it under-audits by exactly the uncited entries.232233`BIB_RESOURCES_ALL` lists every `\bibliography`/`\addbibresource` found, in sorted-path234order; `BIB_RESOURCE` is just the first. If the list has more than one entry, pick the one235the **main** `.tex` declares rather than trusting the ordering, and audit the union only if236the paper really loads several. If `BIB_RESOURCE` comes back empty (e.g. a `\input`-flattened237submission), fall back to the largest `*.bib` under the paper dir and, when a compiled238`main.bbl` exists, read the cited keys from its `\bibitem{...}` entries instead of the `.tex`239(the `.bbl` is the ground truth for what actually rendered).240241Produce the ordered list of cited keys and, for each, its **title + first author** (the242only thing agent B is allowed to see). Confirm every key exists in the `.bib`. Assign each243entry a stable opaque ordinal **`dedup_key`** (`01`, `02`, … in resolved order): this is the244canonical id that joins the shards downstream. A receives `dedup_key` + the cite key; B245receives `dedup_key` + title + first author only (**never** the cite key or `.bib` text), so246the join key never leaks the blind side. Write two artifacts for the Step-3.5 gate, in247resolved order: `/tmp/cx_dedup_keys.txt` (the ordinal set, whitespace-separated) and248`/tmp/cx_keymap.tsv` (one `dedup_key<TAB>cite_key` line per entry — the ordinal→cite-key249**pin**). Step 3.5 builds a coverage manifest (ordinal → cite key → title hash) from250`/tmp/cx_keymap.tsv` + the `.bib` and verifies A against it as a **mapping**, so a citation251dropped *or substituted* downstream fails the run — a bare count or ordinal-set check would252pass a substitution that keeps the count.253254### Step 2 — Run A and B blind to each other255256This is a read-only **extraction fan-out** in the sense of257[`../shared-references/fan-out-pattern.md`](../shared-references/fan-out-pattern.md): A258and B are two shards that only *extract* records (each returns an `entries[]` set keyed259by `dedup_key` — never by list position, never a B-visible cite key, never a quality260ranking). Each shard returns the `{shard_id, entries:[{dedup_key, …}]}` envelope of261[`../shared-references/fan-out-pattern.md`](../shared-references/fan-out-pattern.md), and the262executor runs the mechanical `dedup_key` join below before C. **No shard assigns a263`MATCH`/`MINOR`/`MISMATCH`/`NEEDS_WEBCHECK` label at all** — the main agent assigns all264four in Step 4 from C's field evidence, and only its own web-confirm edits the `.bib`.265266**Dispatch (tier-portable).** On this host there is no Claude `Agent` tool, so the mainline's267Tier 1 (Workflow-concurrent) and Tier 2 (`Agent`) forms both degrade here: run A, then B, then268C one after the other, each in the freshest context the host can give. The join and the269field-diff are identical to the mainline. **Blindness is not automatic here** (see the mirror270adaptation note above): B sets `blind:true` only if its context provably did not inherit the271`.bib`, and `blind:false` otherwise — in which case Step 3.5 downgrades every accept to272`ESCALATE` and the run flags rather than certifies. B is prompted with only the273title+first-author list and is forbidden from reading any local file; but when B is not run in274an isolated context, that prohibition is not *enforced* (B could observe the parent's `.bib`275read, or A's output), which is exactly why an un-isolated B must attest `blind:false` and the276gate then certifies nothing. Isolation is the guarantee; the prompt restriction alone is not.277278**Breadth.** The shard set is fixed at three by the *design*, not by a budget: two records are279the minimum for a double-blind diff (one per provenance), and the third is the differ that must280have produced neither. Breadth is not a swept parameter here, so there is no breadth/cost281trade-off to bound — but note the ceiling: two independent records, not an ensemble.282283> **Acceptance gate (provenance, per284> [`../shared-references/acceptance-gate.md`](../shared-references/acceptance-gate.md)).** This285> skill is **not a loop** and has no self-terminating accept gate; it runs once over a fixed286> citation set. **Type-A** (executor may self-judge): the `dedup_key` join balanced, every entry287> labelled, the `.xlsx` written. **Type-B** (merit — is the entry bibliographically correct):288> never delegated to a shard. A `MISMATCH` or a `NEEDS_WEBCHECK` resolution requires the main289> agent's own **primary-source** fetch — no fetch, no finding. A `MATCH`/`MINOR` rests on the290> two independent records agreeing, without a third fetch; that is weaker evidence than a291> fetch, so read a no-flag `MATCH` as "two sources concur", not "verified against the292> publisher of record".293>294> **Requirement (4) is discharged by the deterministic branch, not by a jury.**295> `fan-out-pattern.md` admits "a single cross-model jury step — OR a deterministic verifier296> gate"; this skill takes the second: `scripts/cx_verify.py` (Step 3.5) is an external process297> that decides `MATCH`/`MINOR` by declared normalization and **fails closed** to `ESCALATE`298> otherwise — the same script and the same verdicts on this host as on the mainline.299>300> The limit, stated plainly: the accept path is mechanical and a `MISMATCH` needs the main301> agent's own primary-source fetch, but the executor still reads that page, so this skill does302> **not** deliver cross-model *acquittal* and does not replace one — which is why it is meant to303> run **in addition to** `citation-audit` (a single cross-model reviewer), not instead of it.304> Never report a `MISMATCH`, least of all a fabrication finding, on shard text alone.305306- **Agent A (repo-only):** read the `.bib`; extract each cited entry **verbatim**307 (brace-matched, no reformatting) → `/tmp/cx_A_ourbib.json`308 `{"shard_id":"A","entries":[{"dedup_key","key","bibtex"}]}`. Explicit instruction: **do not309 search the web.**310- **Agent B (web-only):** given ONLY the title+first-author list from Step 1, re-fetch311 each paper's record from the web and emit a clean BibTeX per paper →312 `/tmp/cx_B_web.json`313 `{"shard_id":"B","blind","entries":[{"dedup_key","title_queried","found","source_url","bibtex","sources_tried","note"}]}`.314 Explicit instruction: **do not read any local file / .bib**; set `found:false` + a note315 if a paper cannot be confirmed; never fabricate a field.316 - **`blind` attestation (set once, for the whole B shard).** B sets `blind:true` **only**317 if it ran in a context that could not have seen the `.bib` — a genuinely isolated318 dispatch that did not inherit the main agent's history. If B cannot guarantee that319 (e.g. it ran in the main context), it MUST set `blind:false`. The Step-3.5 gate refuses320 to certify a `MATCH`/`MINOR` on any entry that is not blind-attested, so a false321 `blind:true` is the one thing that would defeat the whole cross-check — never assert it322 to "pass".323 - **Do not restrict the search to arXiv/DBLP.** Many legitimate references are workshop324 tech reports, company research blogs, or proceedings pages that never appear on arXiv.325 If a title is not on arXiv, B MUST also run a general web search and open the landing326 page **its own search returns** (a project / blog / proceedings URL) before concluding327 `found:false`. B works from the title+first-author query only — it must NOT be handed,328 and cannot use, the `.bib` entry's `url` field (that stays with the main agent's Step-4329 webcheck). A citation being absent from arXiv is NOT evidence it is fake.330 - **Record `sources_tried`** per entry (e.g.331 `["arxiv","crossref","semantic-scholar(429)","web"]`) so a `found:false` from a332 *degraded* search (a source was down or rate-limited) is distinguishable from a genuine333 not-found. Note any source that failed/blocked in the entry's `note`.334335On return, the main agent **joins A and B on `dedup_key`** — every ordinal from Step 1 must336appear exactly once on each side. A missing or duplicated `dedup_key`, or an unbalanced count,337is a **fail-closed stop** (re-run the offending shard), never a silent drop or a positional338guess. This mechanical join replaces order-coupling and is judgment-free — it decides nothing339about any citation, only that the two extractions line up before C diffs them.340341### Step 3 — Run C, the field-diff extractor (after A+B return)342343> **C extracts, it does not adjudicate.** Computing a field-by-field delta is mechanical344> extraction; deciding that an entry *is* `MATCH` (i.e. admitting it as correct) is the345> acceptance verdict, which [`../shared-references/fan-out-pattern.md`](../shared-references/fan-out-pattern.md)346> reserves for the executor. So C emits **per-field evidence**, not a per-entry verdict:347> for each field it reports `same` / `differs` (with both values verbatim) / `absent`, plus348> the routing flags below. The main agent assigns every final349> `MATCH|MINOR|MISMATCH|NEEDS_WEBCHECK` in Step 4 from that evidence. A shard that returns350> a verdict label is out of contract — the same rule `proof-checker` and `research-lit`351> state for their shards.352353A third pass (a fresh-context run — this host grants no delegation tool) reads both JSONs — it354produced neither — and, on each `dedup_key`-joined pair, diffs the two records field by355field: title (ignore case/braces),356**author set AND order**, year, venue (NeurIPS ≡ "Advances in NIPS"; note a genuinely357different event), volume/pages, arXiv id / DOI. Output `/tmp/cx_C_fielddiff.json`:358`{"shard_id":"C","entries":[{"dedup_key","key","blind","found","sources_tried","url","fields":{"<field>":{"status":"same|differs|absent","ours","web"}},"flags":[...]}]}`359(`found`/`sources_tried`/`blind` copied verbatim from B, `url` from A, so Step 4 and the360gate can act without361re-reading the shard files)362(keyed by the joined `dedup_key`) plus a `summary`363`{entries, fields_by_status:{same,differs,absent}, flags_by_type}` — counts of mechanical364field states, **not** a verdict tally.365366**Critical routing flag for C** (this prevents the most damaging false positive): if B367reports `found:false` for an entry whose A-side `.bib` type is `@misc` (or otherwise368carries a `url`/`howpublished` instead of an arXiv id / DOI), and B's `sources_tried`369shows it only queried indexers that cannot see that URL (arXiv/DBLP/etc.), C MUST attach370the flag `url_only_source_unqueried` — and MUST NOT report the entry as absent-from-web.371B's index-based method structurally cannot confirm a blog/tech-report citation; only the372main agent's direct `url` fetch in Step 4 can. C likewise attaches `degraded_search` when373`sources_tried` shows a source was down or rate-limited, and `web_record_absent` whenever B374returned `found:false` for any reason. These flags force the entry into the main agent's Step-4375queue; **none of them is a finding about the citation** — they record what B's search did, not376whether the reference is sound.377378### Step 3.5 — Run the deterministic verifier gate (`cx_verify.py`, mandatory)379380This is the fan-out's **requirement-(4) gate**: `fan-out-pattern.md` demands "a single381cross-model jury step — **OR a deterministic verifier gate** — that is identical across all382three tiers." This skill has no cross-model jury, so the gate is an **external process**. It is the383*same script and the same verdicts* on this host as on the mainline — the dispatch of A/B/C384degrades by tier, the gate does not. The gate **recomputes both the field diff and the routing385flags from A's and B's own raw records** and reads B's blindness attestation from B directly;386C's `flags` are advisory only (unioned with the recompute) and a C schema violation fails387closed — so C is not in the trust path for the diff **or** the routing. Resolve `$CX_VERIFY`388with the Codex hybrid chain389(managed project install, in-repo checkout, ARIS repo fallback, then the copied global Codex390skill install). This is **Policy A — skill-local gate**:391392> **Resolver note (owner-local, per [`../shared-references/integration-contract.md`](../shared-references/integration-contract.md)).**393> `cx_verify.py` is a single-skill helper under this skill's own `scripts/`, resolved via the394> Codex hybrid chain below. It deliberately ships **no** `tools/cx_verify.py` shim — it has no395> legacy `tools/` entry to forward from (unlike the Phase-3 moves that kept one) — which is the396> owner-local Layer-0 exception the contract's Policy-A row registers.397398```bash399CX_VERIFY=""400cd "$(git rev-parse --show-toplevel 2>/dev/null || pwd)" || exit 1401if [ -z "${ARIS_REPO:-}" ] && [ -f .aris/installed-skills-codex.txt ]; then402 ARIS_REPO=$(awk -F'\t' '$1=="repo_root"{print $2; exit}' .aris/installed-skills-codex.txt 2>/dev/null) || true403fi404# Global pointer written by smart_update_codex.sh, for hosts with no per-project install.405if [ -z "${ARIS_REPO:-}" ] && [ -f "$HOME/.aris/repo" ]; then406 ARIS_REPO=$(cat "$HOME/.aris/repo" 2>/dev/null) || true407fi408[ -f ".agents/skills/citation-crosscheck/scripts/cx_verify.py" ] && CX_VERIFY=".agents/skills/citation-crosscheck/scripts/cx_verify.py"409[ -z "$CX_VERIFY" ] && [ -f "skills/citation-crosscheck/scripts/cx_verify.py" ] && CX_VERIFY="skills/citation-crosscheck/scripts/cx_verify.py"410[ -z "$CX_VERIFY" ] && [ -n "${ARIS_REPO:-}" ] && [ -f "$ARIS_REPO/skills/citation-crosscheck/scripts/cx_verify.py" ] && CX_VERIFY="$ARIS_REPO/skills/citation-crosscheck/scripts/cx_verify.py"411[ -z "$CX_VERIFY" ] && [ -f ~/.codex/skills/citation-crosscheck/scripts/cx_verify.py ] && CX_VERIFY="$HOME/.codex/skills/citation-crosscheck/scripts/cx_verify.py"412[ -z "$CX_VERIFY" ] && {413 echo "ERROR: cx_verify.py not resolved at .agents/skills/citation-crosscheck/scripts/, skills/citation-crosscheck/scripts/, \$ARIS_REPO/skills/citation-crosscheck/scripts/, or ~/.codex/skills/citation-crosscheck/scripts/." >&2414 echo " The requirement-(4) gate cannot run; do NOT hand-assign MATCH/MINOR in its place. Fix: rerun install_aris_codex.sh, export ARIS_REPO, or copy the citation-crosscheck skill into ~/.codex/skills/." >&2415 exit 1416}417python3 "$CX_VERIFY" --selftest || exit 1 # built-in cases; fail = do not trust the gate418# Build the coverage manifest (ordinal → cite key → title hash) from Step 1's ordinal PIN419# (/tmp/cx_keymap.tsv) + the .bib, using the gate's OWN parser. This is independent of A, so420# the gate can verify A against it as a MAPPING. ($BIB = the resolved .bib path from Step 1.)421python3 "$CX_VERIFY" --build-manifest --bib "$BIB" --keymap /tmp/cx_keymap.tsv \422 --out-manifest /tmp/cx_manifest.tsv || exit 1423# The gate RECOMPUTES the field diff AND the routing flags from A's and B's own raw records424# (`web_record_absent` from B's `found`, `degraded_search` / `url_only_source_unqueried` from425# B's `sources_tried` + A's fields), and reads B's blindness attestation from B directly. C's426# flags are ADVISORY only, unioned with the recompute, and a C schema violation fails closed —427# so a faulty/omitting/adversarial C can neither hide a discrepancy, forge a MATCH, nor wave an428# entry through by omitting a flag. Coverage is MANDATORY: `--expect-manifest` verifies the429# ordinal→key/title MAPPING against A (catching a drop-and-substitute the ordinal-set check430# cannot); `--expect` / `--expect-keys` add the count and ordinal-set checks. An empty artifact,431# a duplicate/absent/substituted key, a dropped citation, or A/B/C key-sets that disagree all432# exit non-zero rather than reading as "clean".433python3 "$CX_VERIFY" --a /tmp/cx_A_ourbib.json --b /tmp/cx_B_web.json --c /tmp/cx_C_fielddiff.json \434 --expect "$CITED_KEY_COUNT" --expect-keys /tmp/cx_dedup_keys.txt \435 --expect-manifest /tmp/cx_manifest.tsv \436 --json /tmp/cx_gate.json || exit 1437```438439What the gate decides, and what it refuses to decide:440441- **`MATCH`** — every compared field equal after *declared* normalization (LaTeX accent/brace442 folding, case, punctuation; an order-**preserving** surname list; a fixed venue-alias table).443- **`MINOR`** — the sole difference is preprint-vs-published venue for the same work.444- **`ESCALATE`** — everything else, **fail-closed**: any identifying field differing445 (`authors`, `year`, `title`, `doi`, `arxiv`), a secondary field differing, an identifying446 field absent on one side, a venue pair not in the declared table, any routing flag,447 `found != true`, **or an entry that is not blind-attested** (`blind != true`). A448 `MATCH`/`MINOR` that lacks the blindness attestation is downgraded to `ESCALATE`: the run449 can still *flag* a discrepancy, but it may not *certify* a match. This is what enforces the450 double-blind claim mechanically — a host that cannot dispatch B in isolation (so B might451 have seen the `.bib`) sets `blind:false` and the gate refuses to certify anything, turning452 the check into a labelled heuristic rather than a false certification.453454**The gate never emits `MISMATCH`.** A discrepancy becomes a finding only when the main agent455reproduces it against the publisher of record, so the accept path is mechanical and the reject456path is externally evidenced. Author *order* is preserved by the normalizer on purpose — a457re-ordered author list is one of the defects this skill exists to catch, so it must not458normalize to equal.459460### Step 4 — ASSIGN every verdict (main agent, mandatory)461462No verdict arrives pre-assigned from a shard. Read `/tmp/cx_gate.json` (the gate's recomputed463verdicts and per-field evidence) — C's `cx_C_fielddiff.json` is audit-only and must not464override the gate. **The invariant of this step: every `MATCH`/`MINOR`/`MISMATCH` rests on465either (a) the gate's mechanical decision or (b) a completed main-agent primary-source fetch.466No shard signal — a `differs`, an `absent`, B's `found:false`, or a routing flag — is itself a467finding.** Two exhaustive cases:468469- **Gate said `MATCH` or `MINOR`** → adopt it (case (a), mechanical). Do not re-litigate a470 mechanical equality decision from prose, and do not silently upgrade it to `MISMATCH`; if you471 believe the gate is wrong, the fix is the gate's normalization table plus a `--selftest` case,472 not a hand-edit of one row. (What a no-flag `MATCH` means: two independently-produced records473 concur — not "verified against the publisher of record".)474- **Gate said `ESCALATE`** → the main agent **MUST complete a primary-source fetch before it475 may assign any of `MATCH`/`MINOR`/`MISMATCH`/`NEEDS_WEBCHECK`.** This is a *procedure*, not a476 first-match menu: do every applicable fetch, **then** decide — so no earlier branch can477 preempt a stricter one.478479 1. **Fetch.** Identify and `WebFetch` the primary source(s):480 - the entry's own `url` / `howpublished` whenever the `.bib` carries one — always for the481 `url_only_source_unqueried` / `web_record_absent` cases (this is the `.bib`'s `url`,482 which B never saw); **and**483 - the canonical index the fields point at — the arXiv abstract for an arXiv id, the DOI484 resolver for a DOI, else the venue / proceedings page.485 For an unconfirmed-existence entry, **also run the main agent's own general web search**486 (independent of B). Record every URL and query tried.487488 2. **Decide from what the fetch(es) returned** — each terminal rests on a completed fetch:489 - Primary source **confirms** the `.bib` (title + author set/order + year + id agree) →490 `MATCH` (annotate the source; for a `url`-confirmed non-arXiv entry note491 "blog/tech-report, web-confirmed"). A gate `differs`/`absent` the source does **not**492 reproduce is downgraded here, with the reason recorded.493 - Same work, only preprint-vs-published differs → `MINOR`.494 - Primary source **contradicts** an identifying field (author set/order, year,495 venue-as-different-event, arXiv id / DOI, or a substantive title/subtitle/word change;496 or a moved page/volume range) → `MISMATCH`, citing the fetched value. (Field-discrepancy497 `MISMATCH` — the source reproduced it; this guards against a B/C hallucination inventing498 a *wrong* "fix.")499 - **Existence unconfirmable** → `MISMATCH` (fabrication) **only when BOTH hold**: the500 entry's `url` is absent / dead / contradicting, **and** the main agent's own general501 search also failed. A dead `url` alone, or B's `found:false` alone, is **never** enough502 — this is the one path to a fabrication finding and it rests on the main agent's own null503 result, not on B's miss.504 - Fetch **inconclusive** (source unreachable / genuinely ambiguous) → `NEEDS_WEBCHECK`,505 left for a human. Never default an inconclusive fetch to `MATCH` or `MISMATCH`.506507 A one-sided **`absent`** field is *not* resolved "from the fields both sides share": the gate508 already escalated it, and the verdict comes from the fetched primary source per the ladder509 above (an absence becomes `MISMATCH` only when the `.bib` asserts a field the source510 contradicts; otherwise `MATCH`/`MINOR` per the fetch, with the absence noted in `Detail`).511512**Every entry leaves Step 4 with exactly one of the four labels**, each `MATCH`/`MINOR`/513`MISMATCH` backed by the gate or a completed main-agent fetch. If an `ESCALATE`'s evidence514fits none of the outcomes above, that is a fetch not yet done, not licence for a shard to515decide. Do **not** edit the `.bib` on shard evidence alone: every applied fix traces to a516main-agent fetch.517518### Step 5 — Emit the spreadsheet + apply confirmed fixes519520Build the `.xlsx` (openpyxl) with one row per citation and columns:521`# | BibKey | Link | Our BibTeX | Web BibTeX | B source | Verdict | Author check | Year |522Detail | Fix`. Color the Verdict cell: green = `MATCH`, yellow = `MINOR` or a523`NEEDS_WEBCHECK` the main agent resolved to MATCH (annotate how), orange = `MISMATCH`.524Add a Summary sheet with the counts and the fixed/optional lists. Apply only the **main-agent-confirmed** `MISMATCH`525fixes to the `.bib`; if the paper is compiled, rebuild so the `.bbl` picks them up and526re-verify the page count is unchanged. Report MINOR items as optional (do not auto-apply527preprint→published upgrades; that is an author choice).528529## Key rules530531- **Blindness is the mechanism.** If B ever sees the `.bib`, the cross-check collapses532 into a single-source read and loses its power. Enforce it in the prompt.533- **Never fabricate a field** in A, B, or C. `found:false` + a note beats a guess.534- **A non-arXiv miss is not a fabrication.** `@misc`/blog/tech-report/proceedings-URL535 citations are legitimate and invisible to arXiv/DBLP search. B must try the general web536 from the title+first-author prompt only; C flags an unconfirmed URL-bearing entry537 (`url_only_source_unqueried`) and the main agent assigns `NEEDS_WEBCHECK` and fetches the538 entry's own `url` in Step 4. A `MISMATCH` (fabrication) requires **both** a dead or539 contradicting `url` **and** the main agent's own general web search also failing (per Step 4) —540 a dead `url` alone is never enough. This is the skill's most dangerous false-positive if skipped.541- **Shards extract; the main agent adjudicates.** A, B and C emit records and per-field542 deltas. Every `MATCH`/`MINOR`/`MISMATCH`/`NEEDS_WEBCHECK` is assigned by the main agent in543 Step 4, and every `MISMATCH` is web-confirmed by the main agent before the `.bib` is544 touched. The shards propose; the main agent's own fetch disposes.545- **Bibliographic only.** Context-fit ("is it cited for a claim the paper supports?") is546 out of scope — that is `citation-audit`'s job. Recommend running both before submission.547- **Report, do not silently upgrade.** Preprint-vs-published and cosmetic key/year-label548 mismatches are MINOR; list them, leave the choice to the author.