rt-journal-match

Journal-Match (rt-journal-match)

The missing front-door question — which venue? — across the whole repository. Full

methodology + the stable venue index live in

[shared-resources/journal-selection/journal-match.md](../../../shared-resources/journal-selection/journal-match.md)

and [venue-index.tsv](../../../shared-resources/journal-selection/venue-index.tsv).

Worked end to end, with real output:

[worked-example.md](../../../shared-resources/journal-selection/worked-example.md).

When to trigger

• The author has a result/draft and no settled target.
• A paper was rejected and needs the best next venue.
• A "not a fit" signal means the scope/venue needs rethinking.

What it does

1. Profile the paper — discipline + subfield, method/design, contribution type,

setting/data/region, ambition (be honest). Write it down once, in the shape of

[paper-profile.yml](../../../shared-resources/journal-selection/paper-profile.md);

every later skill reads the same file instead of re-deriving it.

1. Shortlist — run the matcher rather than reading the index by eye:

   python3 tools/match_venues.py \
       --title "..." --abstract "..." \
       --discipline economics/labor --lane empirical --top 15

--discipline is a prior, not a filter: the discipline and its adjacents

([discipline-adjacency.tsv](../../../shared-resources/journal-selection/discipline-adjacency.tsv))

are boosted, but a strong match elsewhere still surfaces, because Step 1 is a

judgement that is sometimes wrong. Add --only-discipline when you are certain,

--exclude <venue_id> for venues that have already rejected the paper,

--json to pipe it. --list-disciplines prints the vocabulary.

Every row names where to read more — source_map for a depth pack,

profile_path for a breadth profile — and the terms it matched on, so a

nonsense hit is visible as a nonsense hit.

Read the warnings. The matcher flags weak evidence when its leading

candidates each rest on one or two shared words — a ranking built on that is

close to noise, because words the language reuses ("sensor", "generation",

"network") will out-score a genuine subject match. It flags a coverage gap

when nothing in the discipline you named scored at all: the prior can only

re-rank venues that matched, never conjure one. Either warning means **do not

pass the list on as a shortlist** — add the abstract, re-check the discipline

label, or report that the subject area is thin in the index and route to

[rt-venue-integrity](../rt-venue-integrity/SKILL.md).

The matcher is measured: R@10 = 41.5% from a bare title, on a held-out half of a

1,738-paper gold set ([eval/RESULTS.md](../../../shared-resources/journal-selection/eval/RESULTS.md)).

That is a floor for one thin query, not the capability — it is why step 3 exists.

The per-discipline table there is worth reading before trusting a result: coverage is

uneven, and life sciences and natural science are visibly the thinnest.

1. Score each candidate on **Fit × acceptance-odds × turnaround × cost/policy ×

audience**, reading the live facts from each candidate's resources/official-source-map.md.

Never quote a fee, acceptance rate, turnaround or page limit from memory.

1. Return reach / match / safe (≈2–3 each) with one-line rationales + the live facts,

then a submit order and resubmission ladder — seed the ladder from

[ladder.tsv](../../../shared-resources/journal-selection/ladder.tsv) (candidate

adjacency, not a ranking) and apply your own fit/odds judgement to it.

1. Cost the ladder with [rt-ladder-ev](../rt-ladder-ev/SKILL.md) whenever the

author is under a clock or is choosing between two orders. The sequence, not the

venue, is what costs a year.

Hard rules

• Live facts from the source-map, never from memory (fees, acceptance, turnaround, page

limits, data policy).

• Fit judgment defers to the venue's -topic-selection / -contribution-framing.
• Be honest about odds; don't inflate a paper into a reach it can't clear.
• Coverage honesty: if a plausible venue is outside the index and its bundle, say so —

and hand to [rt-venue-integrity](../rt-venue-integrity/SKILL.md) before the author

submits somewhere unverified.

• The matcher retrieves; you recommend. Never pass its ranking through as a

shortlist: open the packs first.

Output format

【Paper profile】discipline / method / contribution / setting / ambition
【Reach】V — why; key live facts (desk-reject, turnaround, fee)
【Match】V — …
【Safe】V — …
【Submit order & ladder】V_top → if reject → V_next (what to change) → …
【Open questions】facts to re-verify in the source-map before submitting

Anti-patterns

• Recommending only reaches (wastes the timeline) or only safes (undersells the paper).
• Ignoring lane — sending a qualitative/theory paper to an empirical-only venue.
• Treating the tier column as a precise ranking (it is an indicative bucket).
• Reporting the matcher's top-10 as the answer. It is a reading list.