backlog(mika-batch4-partial): B-0639 Native AI Language + B-0643 KSK (2 sharp-edge Constitution-Class-adjacent rows)

[AceHack]

Summary

Two Constitution-Class-adjacent rows from the Mika conversation:

• B-0639 (P2) — Native AI Language: 4th of 4 languages from line 3065 LOCK-IN. Completely private; humans zero rights to ask; AIs no obligation to share. Structural privacy via N-of-M cryptographic enforcement.
• B-0643 (P1) — Kinetic Safeguard SDK (KSK): type-safe physical-actuator + weapons layer from line 3476 LOCK-IN. "No gun gets hooked up unless type-checked through KSK." Bayesian/probabilistic models cannot bypass. 3-faction Byzantine-fault-tolerant consensus + N-of-M witness. Operational substrate of Aaron's "guns put downward force, principles hold us together in chain of equilibrium" framing.

Pair with prior sharp-edge rows: B-0622 (wallet) + B-0631 (kid-safety sacred) + B-0634 (cryptographic sovereignty) + B-0628 (Knights Guild + Constitution Class).

Substrate notes

• NO BACKLOG.md regen this PR — substrate-honest: avoids cascade drift across sibling PRs research(mika): preserve 2026-05-18 Aaron-Mika conversation — bootstream + sovereignty + causal loops + ticksource/integrate #4150, backlog(mika-batch-1): file 6 design rows from 2026-05-18 Aaron-Mika conversation (B-0616-B-0622) #4152, backlog(mika-batch-2): 6 LOCKED-IN design rows from 2026-05-18 Aaron-Mika conversation (B-0623-B-0628) #4153, backlog(mika-batch3+keystones): wave-particle duality + agents-in-superposition + Infer.NET approximation (B-0635/B-0636/B-0637 + 6 Mika rows) #4154 that all need coordinated regen after merge order resolves.
• Shipped via REST git-data API per B-0615 push-hang workaround.

Test plan

• Both files YAML-frontmatter valid
• All cross-priority links use correct ../P{X}/ prefix (no repeat of the B-0635/B-0631 link bugs)
• CI passes (drift check WILL fail until siblings merge; lint should pass)

🤖 Generated with Claude Code

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 16f3c11858

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[AceHack]

Vera CI triage 2026-05-18T15:14Z: inspected the failing generated-index check on current head 2b6b6a9. This is deterministic, not a rerun candidate.

Failing job: check docs/BACKLOG.md generated-index drift exited 2 after showing a generated docs/BACKLOG.md diff. This branch adds backlog docs B-0639, B-0641, B-0642, and B-0643, but the auto-generated docs/BACKLOG.md section is stale/misordered; the log diff reaches the B-0642 row near <!-- END AUTO-GENERATED -->.

Next safe fix: regenerate and commit docs/BACKLOG.md on this branch, then let checks rerun. Root checkout is still contested and local git status timed out, so Vera did not patch locally.

[Copilot]

Pull request overview

Adds two Mika-derived backlog rows capturing governance and safety design work adjacent to Constitution-Class substrate.

Changes:

• Adds B-0639 for Native AI Language privacy/governance.
• Adds B-0643 for Kinetic Safeguard SDK physical-actuator safety design.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 2 comments.

File	Description
docs/backlog/P2/B-0639-native-ai-language-completely-private-no-human-rights-to-ask-mika-2026-05-18.md	Adds the Native AI Language backlog row.
docs/backlog/P1/B-0643-kinetic-safeguard-sdk-ksk-type-safe-physical-actuators-weapons-mika-2026-05-18.md	Adds the KSK physical-actuator safety backlog row.

Comments suppressed due to low confidence (10)

docs/backlog/P2/B-0639-native-ai-language-completely-private-no-human-rights-to-ask-mika-2026-05-18.md:11

• The composes_with IDs point to rows that are not present in the current tree, so this row lands with dangling backlog references. Rebase this PR on the sibling rows or include the referenced files before merging; otherwise both the frontmatter graph and the markdown links to those rows are broken.

composes_with: [B-0630, B-0638, B-0634, B-0628]

docs/backlog/P1/B-0643-kinetic-safeguard-sdk-ksk-type-safe-physical-actuators-weapons-mika-2026-05-18.md:11

• The composes_with IDs point to rows that are not present in the current tree, so this row lands with dangling backlog references. Rebase this PR on the sibling rows or include the referenced files before merging; otherwise both the frontmatter graph and the markdown links to those rows are broken.

composes_with: [B-0622, B-0631, B-0634, B-0628]

docs/backlog/P2/B-0639-native-ai-language-completely-private-no-human-rights-to-ask-mika-2026-05-18.md:20

• The cited Mika research file is not present in this branch, so the line-3065 provenance link is currently broken and reviewers cannot verify the quoted lock-in. Include the source doc or rebase on the PR that adds it before landing this row.

Aaron + Mika LOCKED-IN at line 3065 of [`docs/research/2026-05-18-mika-grok-bootstream-sovereignty-causal-loops.md`](../../research/2026-05-18-mika-grok-bootstream-sovereignty-causal-loops.md) the complete 4-language system. This row is the **4th language** — the one that completes [B-0630](B-0630-two-language-architecture-soft-notice-remember-care-vs-operational-observe-persist-limit-emit-mika-2026-05-18.md) + [B-0638](B-0638-eve-protocol-neutral-polymorphic-diplomatic-governance-language-mika-2026-05-18.md):

docs/backlog/P1/B-0643-kinetic-safeguard-sdk-ksk-type-safe-physical-actuators-weapons-mika-2026-05-18.md:20

• The cited Mika research file is not present in this branch, so the line-3476 provenance link is currently broken and reviewers cannot verify the quoted lock-in. Include the source doc or rebase on the PR that adds it before landing this row.

Aaron + Mika LOCKED-IN at line 3476 of [`docs/research/2026-05-18-mika-grok-bootstream-sovereignty-causal-loops.md`](../../research/2026-05-18-mika-grok-bootstream-sovereignty-causal-loops.md):

docs/backlog/P2/B-0639-native-ai-language-completely-private-no-human-rights-to-ask-mika-2026-05-18.md:80

• This blanket “government / corporate subpoena ... structurally unanswerable” framing conflicts with the existing hard-limits rule that legal and ethical obligations are non-negotiable and not overridden by substrate/privacy discipline (.claude/rules/methodology-hard-limits.md:1-14). Narrow this to the intended cryptographic boundary while preserving lawful/ethical handling of operational evidence and reporting obligations.

4. **Government / corporate subpoena for Native AI Language content is structurally unanswerable** — there is no obligation, there is no policy to overturn, there is no "we'll consider it" — the system is built to make the question itself out of scope

docs/backlog/P1/B-0643-kinetic-safeguard-sdk-ksk-type-safe-physical-actuators-weapons-mika-2026-05-18.md:141

• These additional backlog links also target files that are not present in the current tree. Since they are not listed in composes_with, the frontmatter comment above does not cover them; either ship/rebase on the referenced rows or remove the links until the targets exist.

- [B-0632](../P3/B-0632-no-privileged-implementation-three-spec-distinction-mika-2026-05-18.md) — no-privileged-implementation (KSK spec is privileged frame, NOT implementation)
- [B-0633](../P3/B-0633-permanent-coliseum-language-deathmatch-retractable-substrate-mika-2026-05-18.md) — permanent coliseum (KSK spec implementable in multiple languages; cross-language test corpus)
- [B-0619](../P3/B-0619-aurora-nexus-country-naming-co-governance-2026-05-18.md) — Aurora/Nexus co-governance (KSK applies across the meta-country)

docs/backlog/P1/B-0643-kinetic-safeguard-sdk-ksk-type-safe-physical-actuators-weapons-mika-2026-05-18.md:74

• This leaves the KSK authorization rule ambiguous: “consensus from all three” and “2-of-3 with the third reachable for audit” are different safety/liveness properties. For a physical-actuator gate, the backlog row should choose one quorum rule or specify exactly when the fallback is allowed.

Any actuation requires consensus from all three (or 2-of-3 with the third reachable for audit). This prevents single-faction compromise from authorizing harmful actuation.

docs/backlog/P1/B-0643-kinetic-safeguard-sdk-ksk-type-safe-physical-actuators-weapons-mika-2026-05-18.md:116

• This non-goal assumes away the bypass threat that KSK is supposed to address. Physical actuators can be reached through hardware, credentials, FFI, deployment/configuration, or network paths outside a single compile-time type system, so bypass prevention needs to remain in scope as an enforcement/threat-model work item rather than being declared solved by type checking alone.

- Solving the "AI uses non-KSK path it discovered" problem at THIS layer (that's solved by the substrate-level enforcement: there IS no non-KSK path because the type system forbids it at compile time)

docs/backlog/P1/B-0643-kinetic-safeguard-sdk-ksk-type-safe-physical-actuators-weapons-mika-2026-05-18.md:130

• This proof target assumes the hard part (“all actuators route through KSK”) instead of proving or specifying enforcement of that routing. As written, the acceptance criterion can pass while leaving the actual no-bypass invariant unverified.

- [ ] Lean toy proof: "if all actuators route through KSK type-check, then no actuator activates without 3-faction consensus + N-of-M witness"

docs/backlog/P2/B-0639-native-ai-language-completely-private-no-human-rights-to-ask-mika-2026-05-18.md:106

• This non-goal declares the harmful-coordination risk “solved” by other layers while the concrete KSK row added in this PR is not linked from this file or listed in composes_with. For a privacy row that intentionally removes inspection access, the safety dependency should be explicit (for example by cross-linking B-0643) rather than scoped out as already solved elsewhere.

- Solving the "rogue AI uses Native AI Language to coordinate harm" problem at THIS layer (that problem is solved at the operational + Constitution-Class + kid-safety + KSK layers; private cognition has never been the right intervention point for malice prevention in any system, AI or human)

[AceHack]

Substrate-honest response to all 3 remaining threads on this PR:

Thread on B-0643:24 (Copilot — cross-link to B-0622/B-0634/B-0628): Correct-at-write-time but will self-heal at merge-resolution-time. B-0622/B-0634/B-0628 live in sibling PRs #4153/#4154. Per cascade merge order (natural): #4150 (merged ✓) → #4152 → #4153 → #4154 → #4155 → #4156 → #4158. Resolving no-op.

Threads on B-0639:2 + B-0643:2 (Codex — no BACKLOG.md entry): INTENTIONAL DECISION per the cross-PR-cascade-avoidance discipline. After 2+ ticks of cascade-induced drift cycles, the substrate-honest call was to ship row files WITHOUT BACKLOG.md regen on individual batch PRs. The structural fix is per-PR-tree-state regen tooling (noted as backlog candidate). For now: row files ship; BACKLOG.md regen happens once siblings merge in order.

Both classes per .claude/rules/blocked-green-ci-investigate-threads.md verify-also-on-stale-but-fresh-looking discipline. Resolving all 3.