Retrofit effort

How much agent-assisted work would bring every applicable dimension up to level 3 ("solid")? Dimensions that are n/a for the project are left out. The integrability index (the mean level) also covers only the applicable dimensions, and the report prints how many that is.

effort = Σ base[d] × (3 − level[d]) × legibility_multiplier × size_factor

Term	Value	Why
base[d]	person-days per level per dimension at ~250k tokens: interface 3, legibility 4, events/identity/portability/extensibility 2, ergonomics 1.5, stability/embeddability/observability 1	Adding a contract or a safety net touches more than adding a health endpoint.
size_factor	√(tokens / 250k), clamped to 0.5–4	Retrofits work at boundaries, so cost grows sub-linearly with size.
legibility_multiplier	2.5, 1.6, 1.0, 0.75, 0.5 for legibility 0–4	Hard-to-change code makes every other fix more expensive. It isn't applied to legibility itself.

The report also gives wrappability:

Effort	Verdict
index already ≥ 3	already integrable
≤ 5 days	cheap to wrap
≤ 20 days	moderate retrofit
≤ 60 days	expensive retrofit
more	replace or wrap externally

This is the answer that matters most when re-evaluating older applications. A legacy system with no API but a clean CLI, tests and a readable schema is often cheap to wrap, and the per-dimension breakdown shows where the days would go.

All values are in neosloc/estimate.py and are uncalibrated. See Calibration.