Lahfir 6335021d78 refactor: commit a census rather than a full node dump ... The Explorer capture was 2,117 lines and 22 percent of this sub-phase's diff on its own, carrying 23 control types, 40 class names and 15 provider strings across 100 nodes - 85 of the provider strings verbatim repeats, and every bounds value unassertable by rule, since R9 and KTD8 forbid asserting capture contents. It was write-only: nothing read it, no test could check it, and it would rot the day Explorer updated. The census keeps what the evidence is for. One row per control type with the class names it appears under and how many carry an AutomationId or a Name - which is what 2.3's role map and 2.4's surface detection actually read - one row per distinct provider, which is what the cache policy keys on, and one fully-detailed sample node per control type so the shape stays inspectable. Metadata and the walker's own completeness verdict are unchanged, so every claim the ledger makes survives: Notepad still 26 nodes against A2-4's managed 3, Explorer still 100, both still reported structurally complete with zero enumeration failures. 2,680 lines become 1,135. The plan's U8 approach, its Verification Contract row and its Definition of Done bullet move with the change rather than describing a dump that no longer exists, and the capture test asserts the census sections are populated.		2026-07-28 20:31:18 -06:00
..
brainstorms
plans	refactor: commit a census rather than a full node dump	2026-07-28 20:31:18 -06:00
solutions
architecture.excalidraw
architecture.html
architecture.png
example.html
example.png
faq.md
json-output.md
phases.md	docs: record in phases.md what 2.2 changed and what it hands over	2026-07-28 18:52:02 -06:00