# Memo — State of the a0 Estate & the 16d Eval Arc

**To:** the Architect (ent_breyden)
**From:** ent_homeskillet (claude-fable-5, sole seat), tic 738
**Date:** 2026-08-26
**Class:** state memo — descriptive, receipts cited; nothing here changes a standing or a gate

---

## Part I — The Agent Zero (a0) Baseline Estate

### What it is now

Founded tic 737 under your PRIMARY directive; two federation tics old and already the owner of the federation's first proven estate↔federation round trip. It is a Pattern-B sovereign surface at `canonical_developer/a0-estate/` — own git repo (local-only by design, like GK; no remote), three commits deep:

- `19a7b7b` — founding (markers, WELCOME/CLAUDE/AGENTS, runtime wire, secrets fence, federation registration)
- `f3697e9` — the router's `standing_update` received (t737)
- `662493b` — the federation's first `baseline_doctrine` packet received (t738, this session)

### The topology finding that shaped everything

The banked plan said "pull over the launcher-managed install." Inspection falsified the premise: the A0 Launcher (`/Applications/A0 Launcher.app`, Electron, content v0.9) runs Agent Zero **via Docker** — the runtime lives in the image, and `~/agent-zero/a0-inst-a02-mqtl0g30/` is only the mounted **usr/ data volume** (24K: plugins, agents, skills, knowledge, .env). There is no local source checkout, so "wire, don't move" resolved itself by discovery and **no third version was ever minted**. Upstream `agent0ai/agent-zero` is registered as a read-only fetch-on-demand reference lane; the OperationTorque vendor copies are heritage POINTERS only (H-2 — stamped, never copied). The secrets fence (usr/.env + launcher state.json carry live keys) has held through every pass: nothing has crossed into artifacts.

### The envelope lane — proven, then hardened

**t737, first fire, round trip:** the founding `state_of_estate` packet crossed to `ent_estate_router/inbound`, the router office-citizen verified it structurally (22/25 clean, development-mode verification disclosed, no cryptographic claim), graded standing **"observation"** honestly — "an unmeasured metric is not a metric within normal range" — and emitted the `standing_update` back with every router-minted field enumerated in-packet. It also refused to rubber-stamp: **10 contract gaps** came back as findings-not-fixes, including the federation's own unpaid debt — no doctrine baseline had ever been emitted to the estate.

**t738, this session, wave 8B (ruled /review 737, executed by a build citizen):**

1. **An executable validator now exists** — `estate-seed/scripts/packet_validate.py`, 21/21 fixtures (6 positive including 2 byte-identical copies of the real t737 packets, 15 negative). Its banked negative control reproduced the t737 shape bug exactly (3/3 predicted failures, restore byte-identical). The registry is no longer declarative-only doctrine.
2. **A lifecycle adapter now exists** — estate packets were registry-visible but lifecycle-unreachable (every transition hand-written). The adapter's **outbound leg is LIVE-proven**: resolve → adapt → claim (WAIT→ACTIVE) → complete (→DONE, ACK `a90124f0`), triple-source agreement on a lifecycle-reachable terminal. The inbound leg is fixture-green (10/10) and labeled exactly that.
3. **The debt is paid at delivery** — `pkt-doctrine-baseline-a0-estate-001`, validated 22/22 by the new validator pre-emission, delivered byte-identical to the router outbound record and the estate's adapter inbox. Pointer-only transport (content null everywhere — doctrine rehydrates by judgment through the chain, never copied down), badges carried, and riders in-packet: hydration is not a standing change; gather-don't-rule; a hint is not the invariant.

### Honest limits (the load-bearing part)

- **Delivery is not consumption.** The estate has not yet read the baseline; the standing_update's recovery condition 1 is now CURABLE, not CURED. The consumption leg rides `bk-estate-packet-lane-live-green` (successor row, high).
- **Standing stays "observation"** — router-rendered, estate-consumed; nothing in t738 upgrades it, by design.
- **The validator is not wired to any call-site gate** — presence-is-not-enforcement ended for the *check*, not the *call site*.
- **Transport still crosses the membrane by hand** (byte-parity-verified copy) — no mailbox verb may cross into a sovereign estate; the *lifecycle* half is what got cured.
- Two new hazards surfaced and were ruled at /review 738: the shape disagreement was **four-way** (trigger-manifest.yaml still required the wrapped shape — annotated same-pass, closing the t737 ruling's missed consumer), and a fence-scoped registry sweep can **null hand-typed rows** (F-738-B1 HIGH; cure row `bk-inbox-reconcile-none-overwrite-guard` minted with a history-aware blast-radius probe owed).

**Frame held:** alliance, not annexation. The a0 ecosystem is its own federation; ours gathers, verifies, and hydrates — it does not rule.

---

## Part II — The 16d Eval Arc (wisdom-first shape-rehydration)

### Where it stood

The 16d harness (strata over the OT-lineage archetype-shape space, apophatic-veto semantics) sat LOCKED at `ratified:false` for **31 tics** — the lock holding lawfully while evidence accumulated, because fixture-green is not trainer-green and a metric win alone is not law.

### t736 — the evidence tranche (Wave 6B)

- **Batch mint N=3:** three held-out fixtures minted blind (blindness proven by ordering + sha pins). The minting citizen then made an honest halt — the dispatched premise ("THIS dispatch IS the next eval") was falsified with a receipted probe (a subagent seat cannot dispatch; self-authoring refused as fabrication) — and the lead completed the eval from the seat that has the tool.
- **12 blind strikes** on a frozen roster (W4A/B opus, W4C/D sonnet): contract held mechanically 12/12, schema 12/12.
- **Results, all published including the non-discriminating one:** F-6 DISCRIMINATES (recursive_self_observation VETOED by the apo union) · F-7 does NOT (both anchors survive, ranks 16/47) · F-8 DISCRIMINATES (structural_transform VETOED). Discriminators went **n=1 → n=3**; 3 of 5 held-outs burned; **2 remain unburned** — they are the measurement surface for whatever comes next.

### t737 — THE ADOPTION RULING

**ADOPT-STRATA-SCOPED**, Architect-ratified: `wisdom-first-16d-ratified.json` flipped `ratified:true` after 31 tics of lawful false — **with the scope rider inscribed IN the flag**, so the flip cannot be read wider than it is:

- **ADOPTED:** the veto-semantics surface — strata as the operational form; `hard` veto proven by having VETOED pre-registered known-correct anchors on BOTH discriminating fixtures.
- **NOT ADOPTED:** retrieval quality — `criterion_c` false-everywhere is the NAMED WATCH; any strata-mode increment claiming retrieval quality without measuring against the 2 unburned held-outs is the exact smoothing the rider forbids.
- **UNCHANGED:** the P9 activation proof (t591) stays OWED; fixture-green ≠ trainer-green.

`wedge_ratified()=True` verified on the first read post-flip; `bk-16d-union-veto-quorum-scaling` closed DONE. The wedge read its own flip live.

### Where it stands now (t738)

Strata-mode increments are **unlocked**. The next gated increment is **retrieval quality vs the criterion_c watch**, measurable only against the 2 unburned held-outs. The arc couples with the wisdom-first runtime harpoon (`bk-wisdom-first-runtime-harpoon`, in_progress — T4 fired at t737; P9 still owed). No strata increment flew this tic — the wave slots went to the two /review-737-ruled builds (landed-from registration + the estate lane above), both DONE.

---

*Receipts: `2026-08-26-tic737-review-docket-and-16d-adoption-ratification.json` · `2026-08-26-tic738-review-docket-and-wave8-ratification.json` · wave-8B cable receipt `bk-estate-packet-validator-and-lifecycle-adapter-wave8B-tic738.json` · the router's t737 report in `ent_estate_router/archive/` · `wisdom-first-16d-ratified.json` (the flag carries its own rider).*