docs: FABRIC-3 XLVIII.26 -- mechanism A, and the protocol passing on edbb79e2

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
This commit is contained in:
Claude
2026-11-17 23:36:21 +00:00
parent 9f172ea977
commit f93cde222d
+21
View File
@@ -6760,3 +6760,24 @@ Pushed at Captain Bob's word (8660ab77).
**Next:** mechanism A, test first, measured by five alternating runs, then the protocol.
### 26. Mechanism A: the fabric step walks only the places that hold a node (2026-11-17)
Captain Bob: "Go". The design and its evidence are in `docs/v4.0.0/plans/2026-11-17-the-fabric-step-cost.md`
section 7.
**The change** (edbb79e2):
- The fabric keeps its occupied places in a list, in ascending order. The step's nine passes walk that
list, in the same order as before.
- A node removed while a pass is at it keeps its link, so the walk goes on to the next node.
**Test first:** `test_fabric_walk.c`, 13 checks. Every existing test passes with the same checks.
**The cost:** the host mesh boot's median fell from 11.695 s to 7.999 s, 32% less, over five alternating
runs (`logs/20261117-mechanism-a/timing.txt`).
**The protocol on edbb79e2** (`logs/20261117-222402/PROTOCOL.md`): **PASS**.
- All four boots run POST 990 of 990 with one `dict_hash`, `0x627f392ccbaa96f1`.
- aarch64 is inside its 570 s. It had stopped in POST since 37469f6c.
- test-32, the proofs, the images and the docs pass.
- The docs stage runs longer than one foreground command; it ran in the background at Captain Bob's ruling.