What is this repo actually made of, and how much would it cost to develop?

COCOMO — the Constructive Cost Model — is a software cost-estimation algorithm developed by Barry Boehm in 1981 and calibrated against 63 industrial software projects. It translates codebase size (in thousands of lines of code) into estimates of effort (person-months), schedule (calendar months), and team size, using empirically-fit power-law equations.

This site applies COCOMO's organic mode — the calibration for small-to-medium projects built by experienced developers in a familiar environment:

• effort = 2.4 · KSLOC1.05 person-months
• schedule = 2.5 · effort0.38 calendar months
• people = effort / schedule
• cost = effort × (salary / 12) × 2.4 overhead — default salary $110,000/yr (US median dev base)

These are model-driven estimates from an industry-standard formula — not a quote, not a contract bid. They're useful for comparing repos and for ballparking "how much engineering work is sitting in this codebase."

LOCOMO — the LLM Output COst MOdel — is a heuristic from scc that estimates what it would cost to regenerate a codebase with a large language model. It's the AI-era counterpart to COCOMO: where COCOMO prices the human effort to build the code, LOCOMO prices the tokens to have a model write it again.

It works per file, scaling the work up with complexity — branchy code takes more prompt context and more back-and-forth iterations than boilerplate:

• density = complexity / code
• cycles = 1.5 + √density × 2 — the retry-and-iterate factor
• inputFactor = 1 + √density × 5 — how much prompt/context it takes
• output tokens = code × 10 × cycles; input tokens = code × 20 × inputFactor × cycles
• cost = (input × $/Mtok + output × $/Mtok) ÷ 1,000,000

The headline number uses a medium model tier (~$3 in / $15 out per million tokens); the report page lets you switch between large, medium, small, and local-model tiers. Like COCOMO, treat it as a conversation-starter, not a quote — it ignores shared context across files, the gap between boilerplate and hard algorithmic code, LLM blind spots, and the cost of writing tests and reviewing the output.