agents: membrane-tier doctrine + T1/T2 and T2/T3 wardens - #1174
Merged
Conversation
Formalises the 3+ layer abstraction law the workspace already runs piecemeal (the ndarray polyfill at T0/T1, the lance-graph-java mask-native invariant at T2/T3) as ONE pattern, inherited by every consumer. - .claude/knowledge/membrane-tiers.md — the doctrine: five tiers (T0 substrate / T1 primitive / T2 selection / T3 intent / R2IL), the rule "a tier may only know the vocabulary of the membrane beneath it; nothing crosses except by NAME", the compile-through/entropy rule (old code is re-admitted only by compiling THROUGH a membrane, not around it), the agent->membrane map, and an append-only T2->T3 leak ledger (L1..L7) with the honest statement of what the gate cannot catch (int classid vs int facet are the same type; the warden reviews what reflection can't). - kernel-membrane-warden (T1/T2, opus): NAMED / HAND-COMPOSED / GEOMETRY-LEAK. Catches T2 spelling a T1 op out of smaller ops (the two-AND vs one AND3 ternlog) or computing geometry T2 should read from a lane accessor. - bbb-warden (T2/T3, opus): HANDLE-CLEAN / BYTE-POSITION / UNNAMED-BREACH. The blood-brain barrier: names cross, byte positions never do. - simd-savant gains its T0/T1 membrane section (POLYFILL-CLEAN / RAW-INTRINSIC / SHADOW-KERNEL); preflight-drift-auditor gains a spec-vs-membrane drift check (a brief must not hand a worker wrong-tier vocabulary for the file it edits). - CLAUDE.md knowledge-base list points at the doctrine. Markdown only; no code. Provenance: the 7.5->1.1 ms lgj_hop (T0/T1) and the ternlog conjunction (T1/T2) landed the same day are the membranes working. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_016WkNBjHc2e3zuyz9i8qJEv
|
Important Draft PR not reviewedDraft PRs are not automatically reviewed by default.
To automatically review draft PRs, update your CodeRabbit configuration: reviews:
auto_review:
drafts: trueComment |
Bugbot couldn't run - usage limit reachedBugbot is counted against Cursor usage for this user or team, and this run hit a usage or spend limit. A user or team admin can review and increase usage limits in the Cursor dashboard. (requestId: serverGenReqId_852adfe9-a2a8-482b-8e6f-aac97935875f) |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What
Formalises the 3+ layer abstraction law the workspace already runs piecemeal — the ndarray polyfill at T0/T1, the lance-graph-java mask-native invariant at T2/T3 — as one pattern, inherited by every consumer of the fork. Markdown only, no code.
The principle: a tier may only know the vocabulary of the membrane directly beneath it; nothing crosses a membrane except by name.
ndarray::simd,kernels.rs)&[u64],(offset,stride),IMMsimd-savant)where/hop)kernel-membrane-warden)bbb-warden+ApiSurfaceTest)Files
.claude/knowledge/membrane-tiers.md— the doctrine, the compile-through/entropy rule (old code is re-admitted only by compiling through a membrane, not around it), the agent→membrane map, and an append-only T2→T3 leak ledger (L1–L7). Includes an honest statement of what the gate cannot catch (int classidvsint facetare the same type — reflection can't tell them apart; the warden reviews what the gate can't).kernel-membrane-warden(T1/T2, opus) —NAMED / HAND-COMPOSED / GEOMETRY-LEAK. Catches T2 spelling a T1 op out of smaller ops (twomask_andvs oneAND3ternlog) or computing geometry it should read from a lane accessor.bbb-warden(T2/T3, opus) —HANDLE-CLEAN / BYTE-POSITION / UNNAMED-BREACH. The blood-brain barrier.simd-savantgains its T0/T1 membrane section;preflight-drift-auditorgains a spec-vs-membrane drift check.CLAUDE.mdknowledge-base list points at the doctrine.Provenance
The two fixes landed the same day are the membranes working: the 7.5→1.1 ms
lgj_hop(T0/T1 — a primitive gathered a contiguous lane) and the ternlog conjunction (T1/T2 — T2 hand-composed a T1 op). Both in ndarray#301 + lance-graph-java#73.Paired enforcement PR: AdaWorldAPI/lance-graph-java# (ApiSurfaceTest fence).
🤖 Generated with Claude Code
https://claude.ai/code/session_016WkNBjHc2e3zuyz9i8qJEv
Generated by Claude Code