Tech debt, worked as a repeatable pass instead of a vibe. "The whole
codebase" is not actionable; this makes it a ranked, sized list.

STEP 1 — INVENTORY, BY TYPE (not one bucket)
  ROT      — dead code, unused deps, stale config
  FRICTION — the files everyone dreads touching (ask the team; they know)
  RISK     — no tests on critical paths, security shortcuts
  DESIGN   — the abstraction that fights every new feature

STEP 2 — SCORE EACH: PAIN x FREQUENCY
The key move. A gnarly module nobody touches is low priority; a
mildly-annoying one edited weekly is high. Debt cost = how bad x how
often you hit it. This ranking is what stops you from gold-plating a
dead corner while the daily-friction file festers.

STEP 3 — SEPARATE "HURTS" FROM "UGLY"
Ugly-but-stable code that's rarely touched: leave it. Debt only counts
if it's slowing you down or risking an incident. Half of most "tech
debt" is just aesthetic and safe to ignore.

STEP 4 — THE PLAN
  Top 3 by score → scheduled, sized, owned.
  The rest → a "boy scout" rule (fix debt in files you're already
  editing), so it drains without a dedicated project.

STEP 5 — STOP THE BLEEDING
One prevention change (a lint rule, a review gate) so the same debt
class doesn't refill behind you.

WHAT NOT TO DO
A three-month "pay down all debt" project. It stops feature work, and
the low-frequency items you'd fix weren't costing anything. Rank, fix
the top 3, boy-scout the rest.

Point me at the repo or name the files people dread, and I'll build the
ranked list.