Three snapshots can produce a misleading clean regression diagnosis #80

The stored-history analysis accepts three snapshots, but three snapshots provide only two interval measurements. The sustained-shift detector needs three measurements. The CLI can therefore claim nothing got measurably worse without enough history to detect a shift at all.

Reproduced on upstream 282ffe8fdd7acdd7b6e1dbd8cf04c155401d3a92 through why.Analyze and the real CLI backed by a temporary SQLite store. Three snapshots emit no insufficient-history guidance. Four snapshots with interval means 8, 26 and 26 correctly expose the smallest sufficient sustained slowdown.

I would like to fix the minimum consistently across analysis, offline/live rendering, wider-window guidance and the MCP tool description. This is not an algorithm change or a requirement that every four-snapshot history guarantees usable data: gaps and resets can still limit evidence.

Base regressions fail, affected why/CLI race suites pass, and the committed-HEAD gate, all six release-target builds and full race suite pass. Existing report JSON field shapes are unchanged; no new SQL, dependencies or migrations are needed.