Overruling precedent is a force push and should require gated conditions.

In version control, a force push rewrites shared history. It is sometimes necessary, and every serious team gates it, because it destroys the record other people built on. Overruling precedent does the same to a legal order: it rewrites the history on which litigants, lower courts and legislatures relied.

The Gibsonian canons gate it. Precedent binds unless manifestly erroneous, and it may be overturned only on a showing that the prior interpretation was manifestly erroneous under the canons, not merely disfavored; that reliance interests built on it can be fairly addressed; and that the overruling is itself reasoned under the canons rather than by appeal to changing values, preferences or political circumstances. In the version control reading, a legitimate overruling names the violated canon, supplies the canonical re-reading, addresses reliance and reasons textually, and an overruling that fails those conditions is itself void as precedent. Disagreement with an outcome is not manifest error.

Two consequences follow. First, the conditions make overruling auditable. Whether a decision met them is a question about its reasoning that anyone can check, not a question about whether one likes its result. The Version Control for Law essay reports that, tested against landmark cases, the framework did not track left or right; it tracked whether the opinion did its textual work. Second, the overruled line is not erased. Like a dissent, it remains a preserved branch in the record.

For legal machines the thesis is practical. A system should represent overruling as an event with a date, a scope and a stated ground, not as a flag that silently swaps one rule for another.