docs: dict_hash 0x847bd8ef42f53e14 in README and CLAUDE.md; FABRIC-3 XLVIII.28 closes the capsule-blocks work

Captain Bob ruled: the line for new work is drawn with a merge, then a tag.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
This commit is contained in:
Claude
2026-11-18 21:43:57 +00:00
parent 49c6ec82ee
commit e29a00f717
3 changed files with 34 additions and 2 deletions
+31
View File
@@ -6816,3 +6816,34 @@ What the record asks beyond POST, each gap put to Captain Bob one at a time:
- **Ruled:** a long stage runs in the background, but only alone: "as long as there is no other
background process running including docs. In fact that might generalize too" (`.claude/memory/rules.md`).
### 28. Capsule blocks: no claims, placed in RAM, the lowest loaded (2026-11-18)
**The talk** (Captain Bob): "Capsules are ephemeral." Its rulings, one at a time, are in
`docs/v4.0.0/plans/2026-11-18-capsule-blocks.md`:
- `Block n` is copied down by R, the RAM block count; the fixed 2048 was "THE BUG".
- A later write overwrites: "Deterministic and final", within a capsule too.
- Where a block lands, and any chaining, is the developer's responsibility.
- The block map is "a convenience", and the range tables go.
**Built, test first, each step verified:**
| Step | Commit | What |
|---|---|---|
| 1 | 6f9900f8 | `BLK_RAM_USER_BLOCKS` is a Kconfig knob (default 2048). The kernel's block layout, `mkcapsule` and `mkartemis` follow it. |
| 2 | f30005e1 | `mkcapsule` makes no claims. The collision gate, `capsule-reserved.txt` and the resolver go ("All of it goes"). |
| 3 + 4, v4 | 6dcfbe67 | Every v4 loader places a capsule's blocks in RAM at `n - R`, then interprets `<lowest> LOAD`. The v4 capsules are renumbered to land at 1900 and up, and chained with `-->`. |
**Kept as is:**
- v3's `INIT` ("Yeah, keep").
- The hosted product, only kept green for the gates that run it: "Nobody cares about hosted anymore".
**`(-` as a build marker per block:** an open stub, until the SDK.
**The protocol on 53b1e98e** (`logs/20261118-204012/PROTOCOL.md`): **PASS**.
- POST is 990/990 on all four boots.
- `dict_hash` is `0x847bd8ef42f53e14` on all three ISAs; it changed with the block path.
**The line drawn here** (Captain Bob: "I want to make a delineation for new work"; ruled "Merge, then a
tag"): PR #197 merges once its Acceptance passes, and a tag on `master` marks the line. New work starts
after it.