Measured trust, turning static analysis into a product.
CUPID scans an entire repository and returns a prioritised, line-located,
prescriptive list of improvements: where the problem is, why it carries a cost and how to
fix it, evidenced by real fixes from respected open-source projects.
CUPID ranks only what it can explain, measure, and defend.
BET 01 · RULE-FIRST, TOKEN-FREE
Zero LLM, from detection to wording
Conclusions, ranking, even the report wording all come from static
rules, the IR, dependency graphs and version history; no model call sits anywhere on the
product path. Scans are cheap, reproducible run to run, and the source never leaves the
machine.
BET 02 · HISTORY-VALIDATED
Trajectories, not snapshots
A single snapshot cannot tell essential complexity from accidental
complexity. CUPID uses cross-release trajectories to prove which problems really get
fixed, and cites how others fixed them.
BET 03 · ONE NEUTRAL IR
Languages and rules fully decoupled
Every language first lowers into one neutral fact model, and rules
read only facts. A new language is one new frontend; the product grows by addition,
never by rewrite.
localized · pinned to file / function / lineprescriptive · says how to fix, not just labelsgrounded · cites real gold-repo fixesrule-first · zero tokens at scan scalehistory-validated · backed by 15 open-source histories
CUPID is customer-facing: it judges a repo's present and
prescribes maintainability / performance / readability / architecture fixes ·
JaredMind is the internal sister system: it analyses authors and commit diffs;
a clean division of labour · Python / C / C++ · 2026-07