A deterministic, ML-free churn-shape metric; measurements across seven codebases showing a gradient from hand-maintained (0.57) to agent-built (0.998); its limits; and how to reproduce every figure with git and arithmetic.
Conventional churn analysis counts lines added, deleted, and changed between two snapshots of a codebase. This paper examines the shape of that churn rather than its volume. We define REP_CHURN = (ADD + DEL) / CRN, bounded 0–1: the share of all change activity that was wholesale replacement rather than in-place modification. The hypothesis is that AI-assisted development exhibits a distinctive replacement-dominant signature — generated code is accepted or regenerated, rarely hand-edited — whereas traditional maintenance spends a large fraction of its churn reworking existing statements. Measurements across seven codebases support the hypothesis: repositories with near-zero AI-agent commit attribution cluster at lower REP_CHURN with substantial in-place modification (a hand-maintained JavaScript framework: REP_CHURN 0.57, with 43% of churn modifying existing statements), while an intensively agent-assisted codebase measures REP_CHURN 0.998, with 0.19% of churn modifying existing statements. The metric is deterministic diff arithmetic — no machine learning, no stylistic inference — and every figure is reproducible from public history. We state its failure modes explicitly: bulk file moves, vendoring, and greenfield growth also produce high values, so REP_CHURN indicates a development process, not the authorship of any individual change.
Given the standard two-snapshot churn quantities over logical statements (LLOC) — CHG (statements modified in place), DEL (statements removed), ADD (statements introduced), and total churn CRN = CHG + DEL + ADD — define:
REP_CHURN = (ADD + DEL) / CRN
Its complement, CHG / CRN (the rework share), is often the more intuitive reading: the fraction of all change activity spent editing statements that already existed. REP_CHURN = 1 describes change composed purely of insertions and removals; REP_CHURN = 0 would describe change composed purely of in-place edits. The classification of statements into CHG/DEL/ADD follows the two-pass sequence-alignment method described in Paper A; REP_CHURN adds no new measurement machinery, only a ratio over quantities already computed.
Human maintenance of source code is dominated by editing: renaming, adjusting conditions, threading a parameter, tightening an expression. Each such act modifies existing statements in place and is classified as CHG. Development assisted by code-generating AI proceeds differently as a matter of workflow mechanics: the tool produces a block; the human accepts, rejects, or asks for regeneration. Unsatisfactory code tends to be replaced by a new generation rather than repaired line by line. At the diff level this produces DEL + ADD patterns rather than CHG, regardless of the style or quality of the generated text.
This suggests a process signature that is invisible to stylistic authorship detection but visible to plain churn accounting: as AI assistance increases, the rework share of churn should collapse. Because the signature arises from the workflow rather than the text, it cannot be prompt-engineered away, and it does not degrade as generation quality improves — if anything, better generators make replacement cheaper and the signature stronger.
Seven codebases were measured. For six public repositories, two snapshots were taken (1 January 2026 and 2 July 2026, the nearest commits preceding each date) and compared with a two-pass LLOC differ (Paper A). The seventh subject is the authors’ own agent-assisted commercial codebase, measured over one month of development (1 June – 3 July 2026) using the same engine in git-native mode.
As independent ground truth for the degree of AI involvement, each repository’s commit history was scanned for machine-readable AI co-authorship attributions — Co-Authored-By trailers and equivalent markers left by named AI coding agents (GitHub Copilot’s agent, Claude Code, Cursor, and eight others). Such attribution is voluntary and strippable, so measured rates are floors, not totals. Matching is exact-pattern against agent-specific strings; dependency bots and other scripted automation cannot match.
| Codebase | Character | Agent-signed commits | Rework share (CHG/CRN) | REP_CHURN |
|---|---|---|---|---|
| Express | JavaScript framework, est. 2010, hand-maintained | 0.02% | 42.9% | 0.57 |
| LangChain | AI-infrastructure library | 0.3% | 10.3% | 0.90 |
| FastAPI | Python web framework | 0.03% | 8.6% | 0.91 |
| Redis | C database, est. 2009 | 0.1% | 6.6% | 0.93 |
| CrewAI | AI-native agents library | 3.9% | 2.7% | 0.97 |
| GitHub CLI * | Go tool; 248 Copilot-co-authored commits in the period | 2.2% | 0.9% | 0.99 * |
| Authors’ codebase | agent-assisted commercial product (one month) | 65.9% | 0.19% | 0.998 |
The rework-share column tells the story most plainly. The hand-maintained end of the table spends over forty percent of its churn editing what exists; the agent-assisted end spends a fifth of one percent. The gradient tracks the independent attribution ground truth throughout — with one instructive exception, marked *, discussed below. In absolute terms, the agent-assisted subject also produced churn at roughly six times the monthly rate of the fastest-moving public repository in the sample — replacement-dominant development is fast as well as differently shaped.
* The GitHub CLI figure includes a large repository restructure in the measurement window: bulk file moves classify as DEL + ADD and inflate REP_CHURN. This is the metric’s principal failure mode, and the reason the value carries an asterisk rather than a conclusion.
The defensible claim, stated once and precisely: replacement-dominant churn is characteristic of AI-assisted development; it is not proof of it.
Every figure above can be reproduced without specialised tooling:
git rev-list -1 --before=<date> HEAD, then git worktree add).git log scan for agent co-author trailers.The proposed next study is longitudinal: the same repositories sliced year by year, 2019–2026, testing whether the rework share exhibits an inflection as AI coding assistance became widespread. The method above applies unchanged.