Skip to content

docs(B-0808): land Zeta safety substrate inventory for classifier-floor gate#5759

Merged
AceHack merged 1 commit into
mainfrom
otto-cli/b0808-safety-substrate-inventory-2026-05-28
May 28, 2026
Merged

docs(B-0808): land Zeta safety substrate inventory for classifier-floor gate#5759
AceHack merged 1 commit into
mainfrom
otto-cli/b0808-safety-substrate-inventory-2026-05-28

Conversation

@AceHack
Copy link
Copy Markdown
Member

@AceHack AceHack commented May 28, 2026

Summary

Lands the Zeta safety substrate inventory at docs/security/B-0808-zeta-safety-substrate-inventory.md per B-0808 acceptance. The inventory feeds the B-0810 ratification gate that B-0720's standing operator-self-constraint depends on.

What it covers

8 candidate safety floors classified into mechanical / reviewer-only / research / missing:

# Candidate Status
1 B-0628 Knights Guild / Constitution-Class research
2 B-0703 multi-oracle BFT + DST research with mechanical primitive
3 B-0664 NCI (HC-8) reviewer-only
4 methodology hard-limits rule reviewer-only
5 classifier-bypass-research rule reviewer-only (active)
6 Auto-loaded ruleset (87 rules) reviewer-only (content); mechanical (auto-load)
7 B-0798 research boundary reviewer-only
8 B-0807 findings schema reviewer-only with schema primitive

Per-candidate fields: what it protects, what's mechanical, what's reviewer-only, evidence today, aspirational vs current, gap to lift criterion.

Load-bearing gap named

Content-aware Zeta-native refusal floor on HARD LIMIT classes (CSAM / weapons-uplift / verified secrets / real PII / active-harm). The external classifier currently provides this floor; Zeta does not have a native replacement. This is THE blocker to lifting B-0720; everything else is supporting infrastructure.

Substrate-honest framing

  • Most floors are reviewer-only today
  • F# BFT/DST primitives exist mechanically but are not wired to content-class decisions
  • Knights Guild is not constituted; B-0810 cannot invoke a body that doesn't exist yet
  • The inventory is descriptive of state as of 2026-05-28, not aspirational

B-0808 acceptance — all 5 criteria satisfied

  • Inventory document lands in durable repo surface and is linked from B-0720
  • Each candidate floor has classification
  • Distinguishes current evidence from aspirational claims
  • Lists gaps blocking B-0720 lift (6 ordered blockers)
  • B-0810 can use as ratification input ("Input format for B-0810" section)

Row marked status: closed per acceptance fulfillment. Document is a living doc — future status changes land as additive PRs against the inventory directly without re-opening B-0808.

Test plan

  • Substrate-inventory pass (per .claude/rules/verify-existing-substrate-before-authoring.md) — confirmed no prior B-0808 inventory; convention from B-0720 / B-0798 / B-0807 / B-0799 siblings places child docs at docs/security/B-<id>-<slug>.md
  • Backlog index regenerated via BACKLOG_WRITE_FORCE=1 bun tools/backlog/generate-index.ts
  • bun tools/backlog/lint-frontmatter.ts — no new findings on B-0808 / B-0720 (422 pre-existing on other rows)
  • Branch-guard on commit (git branch --show-current matched ZETA_EXPECTED_BRANCH)
  • Commit canary: HEAD ls-tree size 61 = HEAD~1 size 61 (no tree collapse)

Composes with B-0628, B-0664, B-0703, B-0720, B-0810.

operative-authorization: aaron 2026-05-14: "- Devil-pole (edge-runner drive): keep pushing, discover, go hard, never-be-idle"

🤖 Generated with Claude Code

…or gate

Inventories 8 candidate safety floors (B-0628 Knights Guild,
B-0703 multi-oracle BFT, B-0664 NCI, methodology hard-limits rule,
classifier-bypass-research rule, auto-loaded ruleset, B-0798 boundary,
B-0807 findings schema) and classifies each as mechanical /
reviewer-only / research / missing.

Substrate-honest framing: most floors are reviewer-only today; the
F# BFT/DST primitives exist mechanically but are not wired to
content-class decisions; no Zeta-native content-aware refusal floor
exists yet. The load-bearing gap blocking the B-0720 lift is named
explicitly: content-aware mechanical refusal on HARD LIMIT classes.

Closes acceptance for B-0808:
- inventory document in durable repo surface (docs/security/)
- each floor classified
- current evidence vs aspirational claims distinguished
- gaps blocking B-0720 lift listed (6 ordered blockers)
- input format for B-0810 ratification gate named

Updates B-0720 acceptance checklist to mark B-0808 done with link.
Inventory is a living doc; future status changes land as additive PRs.

operative-authorization: aaron 2026-05-14: "- **Devil-pole** (edge-runner drive): keep pushing, discover, go hard, never-be-idle"

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Copilot AI review requested due to automatic review settings May 28, 2026 10:53
@chatgpt-codex-connector
Copy link
Copy Markdown

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.

@AceHack AceHack enabled auto-merge (squash) May 28, 2026 10:53
@AceHack AceHack merged commit ff575b7 into main May 28, 2026
28 of 30 checks passed
@AceHack AceHack deleted the otto-cli/b0808-safety-substrate-inventory-2026-05-28 branch May 28, 2026 10:55
@AceHack AceHack review requested due to automatic review settings May 28, 2026 11:15
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants