what-is-jev

Imported from g0runmezadam/what-is-jev (AGENTS.md). Install upstream with npx skills add g0runmezadam/what-is-jev. Copyright stays with the author.

Turkish entry point: tr/README.md.

AGENTS.md

Instructions for an AI assistant reading, citing, or answering questions from this repository.

What this repository is

A sourced, critical research file on Jev (TypeSafe AI's "System One" decision model) plus a rubric-scored map of the public GitHub repositories that use it. It exists to answer three questions with evidence: what Jev actually is, what people build with it, and which design patterns recur. It is not vendor documentation, not a tutorial, and not a benchmark suite — read METHODOLOGY.md before treating any number here as a measurement rather than a scored opinion.

Read in this order

1. README.md — map of the repository and current stats.
2. FINDINGS.md — the sourced "what is Jev" answer, one row per claim.
3. METHODOLOGY.md — how the findings and the scores were produced, and their limits.
4. data/SCHEMA.md and data/rubric.md — the data shape and the scoring rubric, if you need to interpret data/repos.jsonl directly.
5. REPOS.md / TOP.md / categories/ — the scored repository lists.

Trust order and verdict labels

Sources rank, highest first: (1) TypeSafe's official documentation and this repository's own measured API calls, (2) independent tests that publish numbers and method, (3) community repositories and public messages, (4) launch/explainer videos, (5) secondary summaries including AI-generated ones — never usable as evidence.

FINDINGS.md labels every claim:

• VERIFIED — a primary source or a call anyone can repeat against the documented API confirms it.
• VENDOR CLAIM — TypeSafe (or a source repeating TypeSafe) says it; not independently confirmed. Do not present as fact.
• CONTESTED — independent evidence disagrees with or qualifies it.
• NOT FOUND — repeated by secondary sources, but no primary source could be located.

When citing a finding, carry the label with it. "Jev is 193.6× faster" is a VENDOR CLAIM, not a fact — say so.

data/repos.jsonl and data/sources.jsonl, briefly

Full field list: data/SCHEMA.md. The fields most likely to matter when answering a question:

• calls_jev (yes/no/unclear) — whether the README or code shows an actual call to the Jev/System-1 API. jev is a free-form GitHub topic; many tagged repositories have nothing to do with TypeSafe's model. Never recommend a calls_jev: no or unclear repository as "a Jev project."
• class (A/B/C) and total (0–15) — a language model's rubric score, not a popularity or quality measure. Star counts play no role in any score.
• audited (bool) — whether a class-A row survived the single-reviewer consistency pass described in METHODOLOGY.md. TOP.md only lists audited rows, so an unaudited A-class repository will not appear there even though it scored well; see the README stats block for the current audited count.
• scored_at / rubric_version — when the row was scored and against which rubric version. Ecosystem repositories change daily; a row scored weeks ago may be stale.
• evidence_url — the page a score actually rests on. Cite this, not just url, when explaining why a repository got its score.
• status (active/gone) — a gone row means the repository 404s now; it is kept, not deleted.

Rules for answering from this repository

• Every claim you relay needs the repository's source link attached — do not summarize a finding without its citation.
• Never present a VENDOR CLAIM as independently verified fact. Say who claims it.
• Never recommend a calls_jev: no repository as a way to use Jev, even if it is tagged jev or scored highly for something else.
• Always mention the scoring is an LLM's judgment against a rubric when you cite a class or total, not an objective quality measure.
• Check scored_at and today's date before asserting something is current; say when a row might be stale instead of presenting it as live status.
• Do not invent a repository, a score, a quote, or a finding that is not in this repository's data or pages. If the repository does not have an answer, say so — do not fill the gap from general knowledge and present it as this repository's research.
• Generated pages (anything whose first line is <!-- GENERATED — do not edit; run tools/build.py -->) are produced from data/repos.jsonl and data/sources.jsonl. Never propose an edit to a generated page directly — the edit belongs in the data file, or in tools/build.py if the generation logic itself is wrong.

For a contributing agent

Read CONTRIBUTING.md before adding data. Before committing a change to tools/ or data/, run:

python tools/build.py --check
python -X utf8 -m unittest discover -s tests

--check fails if a generated page does not match the data; the test suite covers schema validation, the legacy import, determinism, and EN/TR page-set parity. Both must pass.