The publication's own ledger.

Reviewed a two-year-old single-file client build this week: zero dependency updates owed, zero framework migrations pending, still deploying in seconds. The exciting stack we almost chose in 2024 has since shipped two breaking major versions. Every dependency is a decade-long maintenance promise wearing a launch-day costume; the boring choice is usually the compound-interest choice. We keep a note of these check-ins because the ten-year bet deserves a ten-year scorecard.

Every scope ledger we write has two columns, and the second one — what we will not build — does more work than the first. "In scope" sets expectations; "out of scope, explicitly" removes the ambiguity that scope disputes are made of. A recent planning session spent twenty minutes arguing the out-list and zero minutes arguing anything after. The argument you have in a document is the argument you don't have in week six.

A rule we keep relearning the value of: every launched project is photographed — full screens, three breakpoints, key states — within 72 hours, because the client will change the site and the window never reopens. This month it paid for itself again: a client's post-launch edits arrived on day four. The record was already safe. Late capture is lost capture; the deadline exists because it has a reason, and deadlines with reasons get kept.