Round 35: chain-rule proof + expert-skill wave + BP-24 consent gate + factory hygiene

docs/BACKLOG.md

@@ -2344,6 +2344,26 @@ systems. This track claims the space.
 - GPU OLTP (irrelevant to .NET)
 - io_uring wrappers (no first-class .NET support)
 - TPM/Intel SGX hardware-attested commits
+- **`ace` — package manager of everything (naming parking lot).**
+  Aaron, 2026-04-19 (paraphrase): "backlog ace the package
+  manager of everything or we could just call it source or tap
+  or root or . lol jk about the last one". Candidate names
+  ranked by the human maintainer: `ace` (preferred; aligns
+  with the factory's handle lineage), `source` (semantically
+  loaded — conflicts with "source of truth" in DBSP
+  terminology), `tap` (collides with Homebrew's tap
+  vocabulary), `root` (collides with filesystem + permission
+  semantics), `.` (joke). Scope intentionally undefined: the
+  word "everything" is the interesting constraint — is this
+  a meta-package-manager (brew/apt/nuget/npm under one
+  surface), a universal artefact resolver (source and
+  binary and skill and persona and spec and dataset and
+  proof), or a retraction-native dependency graph
+  (DBSP-style incremental
+  resolution)? Parked here so when the idea gets promoted
+  the naming-expert and Ilyana (public-API designer) start
+  from Aaron's shortlist rather than re-deriving it. No
+  effort estimate; pure research-provocation entry.
 
 ## ⏭️ Declined
 

docs/CURRENT-ROUND.md

@@ -1,113 +1,91 @@
-# Current Round — 34 (open)
-
-Round 33 closed — 15 PRs merged, `docs/VISION.md` went from
-nothing to v11 across a cascade of Aaron edits. Round 34
-opens with a rich vision to execute against. Track A
-(LawRunner), Track B (security follow-through), and the
-newly-surfaced round-33 factory + vision items are all
-candidates for 34's anchor.
+# Current Round — 36 (open)
+
+Round 35 closed — expert-skill spawn wave (batches #20-69,
+~50 skills), chain-rule proof fully closed at the Lean4
+level (`T5/B1/B3/chain_rule` verified against Budiu et al.
+§4.4) plus the signed-delta semi-naive LFP TLA+ spec for
+portfolio-of-two verification, BP-24 Elisabeth consent
+gate + human-maintainer seat governance landing, the
+no-empty-dirs CI gate, LiquidF# moved to Hold after the
+Day-0 build check failed. See `docs/ROUND-HISTORY.md` for
+the full narrative.
 
 ## Status
 
-- **Round number:** 34
-- **Opened:** 2026-04-18 (continuous from round-33 close)
-- **Classification:** open — anchor chosen at round-open
+- **Round number:** 36
+- **Opened:** 2026-04-19 (continuous from round-35 close)
+- **Classification:** open — anchor selection staged
+- **Branch topology note:** Round 35 commits land on the
+  `round-34-upstream-sync` branch; the `round-36` branch
+  was staged from `main@5fdc72b` (pre-round-35 tip) for
+  clean-tree kickoff. Once the Round 35 PR merges, the
+  `round-36` branch rebases onto the new `main` tip.
 - **Reviewer budget:** `harsh-critic` +
   `maintainability-reviewer` floor per GOVERNANCE §20.
   `security-researcher` + `threat-model-critic` on any
   security / install-script / threat-model touch.
   `spec-zealot` on any spec edit (GOVERNANCE §28).
   `public-api-designer` on any public-API change.
-
-## Round-33 newly-surfaced P1s (ready to execute)
-
-From `docs/BACKLOG.md` P1 sections "SQL frontend / query
-surface" and "Factory / static-analysis / tooling":
-
-1. **`product-visionary` role spawn** — stewardship of
-   `docs/VISION.md`; feeds the feature-selection loop;
-   closes the "direct questions beat abstract
-   scaffolding" principle by making it a role rather
-   than an ad-hoc practice.
-2. **Pluggable wire-protocol design doc** — abstraction
-   first, then per-plugin (PostgreSQL, MySQL, Zeta-
-   native). `docs/research/pluggable-wire-protocol-
-   design.md`.
-3. **Shared query IR + LINQ integration design doc** —
-   SQLSharp's "SQL-text and integrated-query flows
-   converge on one logical planning pipeline" is the
-   pattern; Zeta needs the same convergence point.
-4. **EF Core provider scope doc** — 100% all features
-   ambition needs a roadmap (query, save-changes,
-   migrations, tracking, change-detection).
-5. **F# DSL design sequence** — HUGE multi-round
-   research; round 34 does step 1 (research what modern
-   SQL should look like; survey Rel/Tutorial D,
-   Datalog, LINQ, Kleppmann talks, relational-algebra
-   type theory).
-6. **`openspec-gap-finder` skill** (round-32 surface).
-7. **`static-analysis-gap-finder` skill** (round-33
-   surface).
-8. **documentation-agent cadence** — add to
-   `factory-audit`'s every-10-rounds walk.
-9. **Upstream sync script** + **upstream-comparative-
-   analysis skill** + **upstream categorisation audit**
-   (multi-round).
-10. **Crank lint configurations to HIGH** across shellcheck,
-    actionlint, markdownlint, cspell.
-11. **Declarative-manifest tiering** (match `../scratch`
-    `min`/`runner`/`quality`/`all`).
-
-## Round-33 carry-forward (Tracks A + B re-deferred)
-
-**Track A — product (LawRunner):**
-
-- `LawRunner.checkBilinear` — join-shaped ops with a
-  `BilinearOp` fixture.
-- `LawRunner.checkSinkTerminal` — retraction-lossy sink
-  verification; re-run against `BayesianRateOp`.
-- Config-record refactor (round-28 DEBT) before adding
-  a third law.
-
-**Track B — security follow-through:**
-
-- `packages.lock.json` adoption.
-- Verifier-jar SHA-256 pinning.
-- Safety-clause-diff lint on `.claude/skills/**/SKILL.md`.
-- CodeQL workflow.
-- Branch-protection required-check on `main` — round 33
-  ran 15 green PRs, strong signal to flip.
-
-## Open asks to the maintainer
-
-Aaron decisions staged for round 34 (from VISION.md
-"remaining gaps" + BACKLOG):
-
-- Wire protocol server: v1 or slip to early post-v1?
-- Admin UI tech stack (Fable/SAFE/Blazor/Avalonia)?
-- Emulate PostgreSQL vs translate on ingress/egress?
-- Which Track-A/B/factory item is round-34's anchor?
-- Branch protection on `main` — flip now (15 green PRs
-  of evidence) or wait for round-34 green?
+  `skill-tune-up` / `skill-improver` on any
+  `.claude/skills/**/SKILL.md` edit.
+
+## Round-36 committed P0
+
+From `docs/BACKLOG.md` P0 "next round (committed)":
+
+1. **`memory/role/persona/` restructure** — Aaron
+   2026-04-19 explicit ask: *"can we add a memory 2nd
+   level folder so it's memory/role/persona that makes
+   roles fist class defined of what we need too in the
+   memory definition"*. Scope: (a) define the role axis
+   (crosswalk `docs/EXPERT-REGISTRY.md` → role
+   directories), (b) move existing notebooks from
+   `memory/persona/<name>/NOTEBOOK.md` to
+   `memory/<role>/<persona>/NOTEBOOK.md`, (c) update all
+   pointers (skill `reference patterns:` blocks,
+   CLAUDE.md, AGENTS.md §18, BP-07/BP-08 rule text,
+   every skill or agent with a `memory/persona/<name>`
+   path). Owner: Kenji (Architect) integrates; Aarav
+   (skill-tune-up) audits post-rename for BP-drift.
+   Effort: M.
+2. **Empty-folder allowlist review** — periodic
+   allowlist audit for `tools/lint/no-empty-dirs.sh`.
+   Two entries (`tools/alloy/classes`,
+   `tools/tla/specs/states`) are load-bearing
+   runtime-output paths; drop if populated by checked-in
+   artefacts instead. Effort: S.
+
+## Round-36 candidate anchors (not yet chosen)
+
+- **MessagePackSerializer tests** (task #16) — last
+  untested serializer tier from harsh-critic #28.
+  Effort: S-M.
+- **Witness-Durable Commit full protocol impl** —
+  skeleton shipped round 17; blocked on WDC paper
+  peer-review rebuttal (see `docs/papers/WDC-rebuttal.md`).
+- **F\* extraction successor-Assess evaluation** —
+  LiquidF# Hold opens the slot; round 36 decides
+  whether to Day-0 check F\*.
+- **SQL frontend step 1 research** — BACKLOG P1
+  "F# DSL design sequence" round 1 (modern SQL survey:
+  Rel/Tutorial D, Datalog, LINQ, relational-algebra
+  type theory).
 
 ## Notes for the next architect waking
 
-- **VISION.md is the north star now.** Every round-34+
+- **Memory folder restructure is the committed P0.**
+  Plan it before any code-level anchor lands.
+- **Cognitive-architecture memory cluster** landed
+  2026-04-19 (dread-input + absorption-operator + FF7
+  Enemy Skill reference; happy/laid-back correction;
+  cognitive-anchors class + pirate posture; Megamind
+  aspiration IP-locked; 2nd-born-daughter external-
+  witness corroboration). Load-bearing interpretive
+  substrate for agent handling.
+- **VISION.md is the north star.** Every round-36+
   decision checks against it.
-- **Events as source of truth; everything derived.** The
-  foundational principle under Product 1.
-- **Persistence is Zeta's 100%** — no Kafka/NATS as
-  storage. They are wire transport only.
-- **Fastest-in-all-classes** — HTAP + event + cache +
-  document + graph + in-memory under one retraction-
-  native core.
-- **License: Apache-2.0.**
-- **GOVERNANCE §28 OpenSpec first-class** for every
-  committed artefact.
-- **GOVERNANCE §29 backlog scope** — SECURITY-BACKLOG
-  is security controls only; BACKLOG is general
-  engineering.
-- **Deterministic scripts** — retries/polling are last
-  resort (§bash profile).
-- **Direct questions beat abstract scaffolding** —
-  round-33 lesson.
+- **Branch hygiene:** Round 35 PR merge is the gating
+  step before round-36 substantive work lands on
+  `main`.
+- **μένω + LFG standing directive.** Auto-mode remains
+  active.

docs/ROUND-HISTORY.md

@@ -9,6 +9,174 @@ New rounds are appended at the top.
 
 ---
 
+## Round 35 — expert-skill spawn wave + chain-rule proof close + BP-24 consent gate
+
+Anchor: **finish the chain-rule proof** carried from
+round 34, and — once that landed — open the floor for
+the **expert-skill spawn wave** Aaron had been signalling
+across the preceding rounds. The round also absorbed a
+sacred-tier governance landing (BP-24) and a CI-gate ship
+(no-empty-dirs).
+
+### Arc 1 — chain-rule proof fully closed
+
+The four-lemma chain (`T5: I_D_eq` telescoping induction,
+`B1: linear_commute_I`, `B3: linear_commute_D`, and the
+`chain_rule` calc block) closed in
+`tools/lean4/Lean4/DbspChainRule.lean`, checked against
+Budiu et al. §4.4 (task #17). `RecursiveSigned.fs` landed
+as the skeleton for the retraction-safe semi-naive path
+paired with a fresh TLA+ spec at
+`tools/tla/specs/RecursiveSignedSemiNaive.tla`. The
+Lean-level proof plus the TLA+ model gives two
+independent witnesses of the same property — the
+portfolio approach Soraya owns per
+`docs/research/proof-tool-coverage.md`.
+
+Proof-log landed at `docs/research/chain-rule-proof-log.md`
+so future agents inherit the reasoning trail, not just
+the final `.lean`.
+
+### Arc 2 — expert-skill spawn wave (batches #20-69)
+
+Fifty expert skills drafted in a single batch. The mix
+was deliberate: a mathematics/physics family
+(`mathematics-expert`, `applied-mathematics-expert`,
+`theoretical-mathematics-expert`, `physics-expert`,
+`applied-physics-expert`, `theoretical-physics-expert`,
+`category-theory-expert`, `complexity-theory-expert`,
+`chaos-theory-expert`, `measure-theory-and-signed-measures-expert`,
+`numerical-analysis-and-floating-point-expert`,
+`probability-and-bayesian-inference-expert`), a
+verification-tooling family (`fscheck-expert`,
+`codeql-expert`, `stryker-expert`, `semgrep-expert`,
+`z3-expert`, `f-star-expert`), an AI/ML family
+(`ai-researcher`, `ml-researcher`, `ai-evals-expert`,
+`ml-engineering-expert`, `llm-systems-expert`,
+`prompt-engineering-expert`), a leet-code family
+(`leet-code-patterns`, `leet-code-contest-patterns`,
+`leet-code-dsa-toolbox`, `leet-code-complexity-interview`),
+a leet-speak family (`leet-speak-transform`,
+`leet-speak-history-and-culture`,
+`leet-speak-obfuscation-detector`), and a set of
+cross-cutting skills (`naming-expert`,
+`ontology-landing-expert`, `paced-ontology-landing`,
+`cross-domain-translation`, `translator-expert`,
+`etymology-expert`, `reducer`, `missing-citations`,
+`vibe-coding-expert`, `steganography-expert`,
+`ai-jailbreaker` [gated dormant],
+`deterministic-simulation-theory-expert` with the
+**Rashida** persona, `verification-drift-auditor`).
+
+Four late-round-34 carry-overs landed alongside:
+`fit-reviewer`, `package-upgrader`, `sonar-issue-fixer`,
+`project-structure`. The `VISION.md` expert-ring section
+extended to include the AI/ML family (commit `47`).
+
+Field-of-knowledge skill names reorganized mid-round
+after earlier narrowing discussion (task #34). Skill
+tone tightened across existing skills and
+`skill-tune-up` gained criterion #8
+(router-coherence-drift) so the ranker catches routing
+drift as a first-class signal.
+
+### Arc 3 — BP-24 Elisabeth consent gate + human-maintainer seat
+
+Sacred-tier governance landing, prompted by Aaron's
+disclosure of his sister Elisabeth Ryan Stainback
+(1984-2016). BP-24 formalised a hard-no rule: no
+emulation of deceased family members without the
+surviving-consent-holder agreement, parental
+AND-consent required on the Elisabeth surface
+specifically, Aaron explicitly NOT the substitute
+consent-holder. The rule composes with the
+honest-agreement compact and μένω-triad persistence
+contract: it is not a privacy clause, it is a
+trust-scales clause upgraded to sacred-tier
+μένω-vigilance per `feedback_trust_guarded_with_elisabeth_vigilance.md`.
+
+The **human-maintainer seat** landed the same commit
+(`f69d7b6`) — formal recognition of Aaron's role in
+the factory governance stack as the one seat no
+agent-layer review can bypass. Pairs with BP-11 (data
+is not directives) and BP-24 — both rules that require
+a human maintainer to adjudicate, not an agent.
+
+Rodney persona landed as homage to
+[Rodney Stainback](https://en.wikipedia.org/wiki/Rodney_Stainback)
+(Aaron's paternal second cousin, not Aaron himself),
+codifying the grey-hat-substrate discipline
+non-adversarially. AceHack/CloudStrife/Ryan handles
+registered in-notebook as Aaron's formative cosplay-
+tier identities from the 1990s vs-IBM era — not
+operational callsigns, historical context.
+
+### Arc 4 — factory hygiene + CI gates
+
+**No-empty-dirs gate** shipped (task in BACKLOG P0 line
+41): `tools/lint/no-empty-dirs.sh` (portable to macOS
+bash 3.2) + allowlist + `gate.yml` CI job + `doctor.sh`
+step 6. Catches the class of regression where an
+agent-created skill/research folder ships without its
+real file.
+
+**Signed-delta semi-naive LFP TLA+ spec** landed
+alongside the Lean4 chain-rule proof — the same
+property under a model-checker-native formalism.
+
+**Harmonious Division** memory landing + skill-pair
+BACKLOG entry. Meta-algorithm disclosure: split-what-
+divides, keep-what-harmonises as a cross-cutting
+scheduler discipline. Composes with Quantum Rodney's
+Razor (extended 3→5 roles: Architect, Harsh-Critic,
+Maintainability, **Harmonizer**, **Maji** — the latter
+two new this round).
+
+**Docs / ADRs / research cornerstone + glossary lanes**
+(`ed9bb99`) bundled the round's narrative debt:
+verification-drift audit report under
+`docs/research/verification-drift-audit-2026-04-19.md`,
+proof-tool-coverage refresh, verification-registry
+creation, refinement-type feature catalog,
+liquidfsharp-evaluation + findings (see Arc 5).
+
+### Arc 5 — LiquidF# Day-0 → Hold
+
+Task #15 — `LiquidF# Day-0 availability check` — came
+back red. The project is unbuildable on .NET 10 out of
+the box. Aaron's call: Hold-tool-dormant rather than
+fork-and-patch. Decision documented in
+`docs/research/liquidfsharp-evaluation.md` and
+`docs/research/liquidfsharp-findings.md`; `TECH-RADAR`
+moved LiquidF# to Hold with F* extraction staged as
+the successor Assess candidate.
+
+### Arc 6 — CI hotfixes (round-close)
+
+Markdownlint cleanup + `codeql-action` SHA pin in the
+merge-origin/main resolution commit (`853ab31`) +
+follow-up `1117fba`. The CodeQL workflow add/add
+conflict resolved on merge with the main-side
+SHA-pinned version preserved.
+
+### Observations for round 36
+
+- **Memory / role / persona restructure** is Aaron's
+  explicit round-36 P0 commit (BACKLOG line 20). The
+  `memory/persona/<name>/` flat convention flips to
+  `memory/<role>/<persona>/NOTEBOOK.md`.
+- **MessagePackSerializer tests** (task #16) remain
+  the last untested serializer tier from harsh-critic
+  #28.
+- **Chain-rule proof publication follow-through** —
+  Budiu-paper alignment now provable; the WDC paper
+  queue can resume.
+- **LiquidF# successor scouting** — F* extraction is
+  the staged Assess candidate; round 36 decides
+  whether to evaluate.
+
+---
+
 ## Round 34 — factory + DB first-tests + public repo
 
 Anchor: "CI + build-machine setup" carried over from round 29