Test the untested decision logic: tracker finalization, read-cache staleness, budget gating

[egorfedorov]

Problem

The highest-risk modules have zero tests on their core paths (per architecture audit):

• tracker.js — session finalization, usefulness recompute, pattern/co-occurrence aggregation, auto-template creation. High risk, untested.
• read-cache.js — checkStaleness() (token/file/time thresholds) and the big-file map-then-load nudge. High risk, untested.
• budget.js — warning-emission gating at 50/70/85/95% and the compact recommendation.

Current tests/test.js covers pure helpers (estimateTokens, isRangeCovered, globToRegex, analyzePrompt, etc.) but not the stateful pipelines that actually decide what gets blocked and what "savings" are reported.

Why it matters

A product marketed as honest needs its decision logic under test, not just its formatters.

Proposal

• Add fixture-based tests: feed a synthetic session → assert finalized usefulness/wasted classification + pattern aggregation.
• Staleness: table-driven tests over (elapsedTokens, elapsedFiles, elapsedTime) → allow/block.
• Budget gating: assert each threshold emits exactly once.
• Refactor any process.exit(0) out of the testable core (see read-cache allow()), so tests don't terminate the runner.

Acceptance

• tracker finalization, read-cache staleness, budget gating each covered.
• npm test count meaningfully increases; no path calls process.exit during tests.