dossier

name	dossier
description	Compute a promotion / doçentlik / atama dossier's points table from a publication export you supply and a criteria table you fill from ÜAK's or your university's own published document — and re-verify every publication's index claim against the venue cache, because the most common way a dossier entry is struck out is an ESCI paper entered under SCI-E. Also renders an academic CV from the same list. Keeps nothing between runs. Use at a promotion application or when auditing your own record; /journal-check profiles a single venue.
disable-model-invocation	true
argument-hint	--publications <export.bib|.json|.csv> [--cv]

/dossier — arithmetic, and the index claim nobody re-checks

A promotion file is a points table: each publication lands in a criterion, each criterion carries a value, the values sum. All of it is arithmetic, and all of it gets done by hand at midnight against a PDF of criteria.

The error that costs a file is not the addition. It is the index claim — a paper in an ESCI journal entered under the SCI-E criterion, usually honestly: the journal was SCIE when the work started, or the publisher's own page said so. A reviewer checks it against the Master Journal List and the entry falls.

What this keeps between runs: nothing

There is no ledger, no publication history, no profile. Every run starts from the two files you pass, and the script writes only where --write points, inside the project directory.

POSITIONING.md refuses cross-paper memory, and this looked like the feature that would finally have to move that line. It did not — the value here was never in remembering. It is in the arithmetic and the re-verification, neither of which needs a history.

Step 1: export your publications

ORCID and Scopus both export BibTeX; so does Web of Science, Zotero and Mendeley. Put it in dossier/publications.bib.

.json (a plain array) and .csv (with a header) also work. ORCID's own JSON is deliberately not parsed — its schema is deep and versioned, and a parser that silently mis-reads half a record is worse than an export step.

Fields that matter: title, year, type (article / conference / chapter / book / thesis / patent), venue, issn, doi, and optionally authorship (first / corresponding / co) and index (what you claim, which is what gets checked).

Step 2: the criteria table, from its own document

{
  "authority": "ÜAK 2026 Doçentlik Temel Alan Ölçütleri — Mühendislik",
  "source_url": "https://uak.gov.tr/…",
  "read_on": "2026-08-15",
  "minimum_total": 100,
  "criteria": [
    {"id": "A1", "label": "SCI-E/SSCI/AHCI makale", "points": 20,
     "match": {"type": "article", "index": ["SCIE", "SSCI", "AHCI"]}, "min_count": 2},
    {"id":     
         
         
         
         
         
  

Read the values off the table itself. They differ by temel alan, by university, and by year. The kit will not recall "doçentlik için 100 puan" — that is the authority-file rule, and a dossier built on a remembered point value is built on the same error class as a remembered DOI.

Order matters: a publication scores against the first criterion it matches, which is your declaration of precedence.

Step 3: fill the venue cache

Verification reads sources/venues.jsonl — this project's own cache. A venue that is not in it is not counted:

python scripts/venue_check.py --issn 0888-3270
python scripts/venue_check.py --title "Journal of Emerging Things"

Step 4: score

python scripts/dossier_score.py --publications dossier/publications.bib

Three buckets, kept apart on purpose:

Bucket	Meaning
counted	the venue was verified and the publication matched a criterion
pending verification	it would match, but its venue is not in the cache — points reported separately, neither added nor dropped
unmatched	no criterion fits: either the criteria file is incomplete, or these do not count

And the findings:

index_mismatches:
  "Transfer learning under load"  claimed SCIE  verified ESCI  (checked 2026-08-01)

A criterion with a max_count reports which publications fell outside it, by name. The order it counted in is your export's order, which is no order at all — if a different paper should take the slot, move it up in the export.

The mismatched paper then scores against ESCI, not against the claim. Every verified row carries the date its index was checked, because a WoS class is a fact with a date on it.

Step 5: the CV

python scripts/dossier_score.py --publications dossier/publications.bib --cv \
    --write dossier/cv.md

Grouped by the criteria table's own categories, newest first, each line carrying the verified index. No "selected publications" judgement, no invented affiliation, title or date — every line comes from the export.

What this does not do

• It does not decide the file is sufficient. It reports counted points against the minimum the criteria file declares, and stops. That decision belongs to a committee.
• It does not recall a criterion, a point value or a threshold.
• It does not improve an index. A claim the cache does not confirm is a mismatch, reported with the date of the check.
• It does not fill YÖKSİS. No portal is touched and no credential is held.

Reference

• Script: python scripts/dossier_score.py --publications <file> [--criteria] [--venues] [--cv] [--write]
• Reads: your export, dossier/criteria.json, sources/venues.jsonl
• Writes: only --write, inside the project
• Related: /journal-check (fills the venue cache), /import-library (turns a reference library into a register), /submission-pack

---

A dossier is not judged on its total. It is judged on whether every line in it survives being looked up.