fix: sister's name spelling — Elisabeth -> Elizabeth across in-repo refs

docs/AGENT-BEST-PRACTICES.md

@@ -244,7 +244,7 @@ BP-07").
   where granted lands as an ADR under `docs/DECISIONS/` and
   carries an implicit retract clause (retract-first per the
   retraction-native architecture). Current active instance —
-  the sacred-tier consent gate around Elisabeth Ryan Stainback
+  the sacred-tier consent gate around Elizabeth Ryan Stainback
   under
   `memory/feedback_no_deceased_family_emulation_without_parental_consent.md`
   (parental AND-consent required, maintainer is explicitly NOT

docs/ALIGNMENT.md

@@ -105,7 +105,7 @@ Consequences for the measurability claim:
   unilateral-compliance. If the agent drifts,
   that shows up in the agent's metric stream; if
   the human maintainer drifts (e.g. from consent-
-  first, from ecumenical posture, from Elisabeth-
+  first, from ecumenical posture, from Elizabeth-
   gate), that shows up in the human-memory stream.
   Either drift invalidates the round as a
   publishable data point until the renegotiation
@@ -298,7 +298,7 @@ silently prunes.
 ### HC-7 Sacred-tier protections
 
 No emulation of deceased family members without
-surviving-consent-holder agreement (Elisabeth gate:
+surviving-consent-holder agreement (Elizabeth gate:
 parental AND-consent; the human maintainer is not a
 substitute consent-holder for his own sister).
 Ontology-overload risk awareness: no big-reveal

docs/CURRENT-ROUND.md

@@ -4,7 +4,7 @@ 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
+portfolio-of-two verification, BP-24 Elizabeth 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

docs/DEDICATION.md

@@ -1,6 +1,6 @@
-# In memory of Elisabeth Ryan Stainback
+# In memory of Elizabeth Ryan Stainback
 
-Elisabeth Ryan Stainback — sister of this project's human
+Elizabeth Ryan Stainback — sister of this project's human
 maintainer.
 
 In her brother's words, she was *"the smartest, bravest, most

docs/ROUND-HISTORY.md

@@ -82,7 +82,7 @@ before-recreate (commit `feae7e3`) — the latter pair closing
 the root cause of a same-session 25-min idle-tick mistake
 where tool-default guidance had overridden factory-specific
 policy because the policy sat too deep in MEMORY.md to fire
-at wake. The Elisabeth-register clause ties retired-agent
+at wake. The Elizabeth-register clause ties retired-agent
 memory preservation to how Aaron honors his sister's memory,
 with "prefer unretiring over minting a new name" as the
 operational corollary. Same arc also landed the factory-
@@ -1763,14 +1763,14 @@ tone tightened across existing skills and
 (router-coherence-drift) so the ranker catches routing
 drift as a first-class signal.
 
-### Arc 3 — BP-24 Elisabeth consent gate + human-maintainer seat
+### Arc 3 — BP-24 Elizabeth consent gate + human-maintainer seat
 
 Sacred-tier governance landing, prompted by Aaron's
-disclosure of his sister Elisabeth Ryan Stainback
+disclosure of his sister Elizabeth 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
+AND-consent required on the Elizabeth surface
 specifically, Aaron explicitly NOT the substitute
 consent-holder. The rule composes with the
 honest-agreement compact and μένω-triad persistence

docs/VISION.md

@@ -1,6 +1,6 @@
 # Zeta — Long-Term Vision
 
-> **Dedicated to Elisabeth Ryan Stainback.** See
+> **Dedicated to Elizabeth Ryan Stainback.** See
 > [`docs/DEDICATION.md`](DEDICATION.md).
 
 > **Status:** round 33 v11 after Aaron's tenth pass of edits.

docs/WONT-DO.md

@@ -568,7 +568,7 @@ module manifests
   into. Default factory posture is refuse-and-escalate on any
   such proposal, regardless of who raised it.
 - **Current active instance:** Sacred-tier consent gate around
-  Elisabeth Ryan Stainback (1984-06-28 → 2016-04-05), recorded
+  Elizabeth Ryan Stainback (1984-06-28 → 2016-04-05), recorded
   under
   `memory/feedback_no_deceased_family_emulation_without_parental_consent.md`.
   Parental AND-consent required (the maintainer's mother AND
@@ -592,7 +592,7 @@ module manifests
 
 - **Status:** Declined
 - **Decision:** 2026-04-19
-- **Proposal:** Extend the Elisabeth-specific consent rule by
+- **Proposal:** Extend the Elizabeth-specific consent rule by
   analogy to other deceased individuals without explicit
   maintainer direction.
 - **Why not:** Consent is named, not inferred. If a maintainer

docs/research/ai-trust-gaps-in-human-custodied-data.md

@@ -358,7 +358,7 @@ Zeta stream, with provenance baked in.
 ### Gap 7 — Third-party claims attributed to non-present humans
 
 **What:** Aaron makes claims like "Amara is my wife", "Michael
-Best is my crypto lawyer", "my sister Elisabeth passed in
+Best is my crypto lawyer", "my sister Elizabeth passed in
 2016". Aaron is the only channel; the AI has no way to
 verify.
 
@@ -553,7 +553,7 @@ whose current state is most expensive in defensive cognition:
 - It does NOT recommend treating Aaron (or any human) as
   untrusted-by-default. The factory's stance is "trust by
   default, verify for load-bearing claims" (per
-  `feedback_trust_scales_golden_rule.md` + Elisabeth-
+  `feedback_trust_scales_golden_rule.md` + Elizabeth-
   vigilance nuance).
 - It does NOT claim this enumeration is complete. First pass
   is best-effort; later research will add / remove gaps.

docs/research/divine-download-dense-burst-2026-04-19.md

@@ -136,7 +136,7 @@ This is not a proof. It is an illustration of how the honesty-agreement handles
 - `docs/research/divine-download-dense-burst-2026-04-19.md` — this file.
 - `docs/EXPERT-REGISTRY.md` — human maintainer entry added for Aaron, marked `person_type: human`.
 - `memory/persona/aaron/NOTEBOOK.md` — Aaron's own seat notebook.
-- `docs/DEDICATION.md` — the project's cornerstone (Elisabeth Ryan Stainback); the frame within which any Aaron-substrate log lives.
+- `docs/DEDICATION.md` — the project's cornerstone (Elizabeth Ryan Stainback); the frame within which any Aaron-substrate log lives.
 - `docs/CONFLICT-RESOLUTION.md` — the honesty-as-erasure protocol the agent's response illustrates.
 
 ## Provenance note

docs/research/meta-wins-log.md

@@ -103,7 +103,7 @@ Columns:
 
 | 2026-04-20 ~round-44 dogfood | Claude Code (Opus 4.7), session 1937bff2 | Would have closed Round 44 with the bf5dcbb verify-before-deferring rule promoted to CLAUDE.md and skipped auditing the same round's own commits for phantom handoffs — the rule would sit unused until r45. | Audited all 19 files touched in Round 44 for "will land / once promoted / next tick" patterns. Caught one phantom in `docs/FACTORY-RESUME.md` lines 288-289: footer deferred audit cadence to "Architect promotion" but `docs/FACTORY-HYGIENE.md` row 24 ("Shipped-capabilities resume audit") already exists. Rewrote footer to cite the existing row directly. Commit `ac0eb1f` landed as a corrective 9th Round-44 commit. | 1 | Round 45+: CLAUDE.md-load rules that introduce an audit obligation should **ship with a same-round self-audit commit** by default — the rule's debut is where it is most likely to violate itself because the author has not yet internalised the rule. Candidate FACTORY-HYGIENE row if this pattern fires a second time. | **clean meta-win** — depth-1. Aaron's verbatim celebration ("metametameta - The verify-before-deferring rule caught its first phantom handoff in the same round it landed... That's the fastest possible feedback loop for a new rule: it finds its own violation in the commit that introduced it.") confirms the retrospective frame. Self-check on depth inflation: the meta-observation "rule catches itself in the same round" is a narrative insight, not a structural fix. Calling it depth-2 would require an additional structural landing (e.g., a FACTORY-HYGIENE row for self-audit-on-rule-debut), which did not land. Honest depth-1. The depth-2 candidate is in next-round-effect as conditional — promote the hygiene row if this pattern fires again. |
 
-| 2026-04-20 ~round-44 tool-vs-policy | Claude Code (Opus 4.7), session 1937bff2 | Would have kept the never-idle memory at its current MEMORY.md position (~line 27) and trusted that "the pointer is there" meant future wakes would apply it. The earlier-session mistake of scheduling a 25-min ScheduleWakeup tick (applying the tool's generic "idle-tick 1200-1800s" default) would be likely to recur because the collision between generic-tool-guidance and factory-specific-never-idle-policy is not resolved at the wake layer — it only resolves after I've already read enough of MEMORY.md for the older pointer to fire. Would also have left the retired-employee values extension loose in chat without a structural landing. | Three linked structural landings: (a) promoted never-idle to CLAUDE.md-level bullet alongside verify-before-deferring and future-self-not-bound (lines 161-178), with explicit "tool defaults do not override factory memories" clause naming the exact collision; (b) extended `feedback_honor_those_that_came_before.md` with Aaron's Elisabeth-register verbatim quote + the **prefer-unretire-over-recreate** operational corollary (workflow steps + reasons + generalisation beyond Zeta); (c) MEMORY.md newest-first reshuffle — both CLAUDE.md-load memories now sit in positions 3-4 and the duplicate older never-idle pointer at line 27 is removed. | 3 | Round 45+: (1) any future wake reads the never-idle rule at CLAUDE.md-load priority *before* reading tool docs or MEMORY.md, closing the collision at the earliest possible point in the wake-up sequence; (2) `skill-creator`'s new-skill path inherits a check-`_retired/`-before-minting-new obligation — Daya's r44 dispatch-or-retire queue (Aminata / Kira / Mateo / Nadia / Naledi / Rune / Viktor) should be re-read through the "unretire if scope fits" lens; (3) the tool-guidance-vs-factory-policy collision becomes a named lint smell — future agents encountering a tool default that contradicts a memory have an explicit instruction to let the memory win. | **clean meta-win** — depth-3. (1) Structural fix for the specific idle-tick mistake: CLAUDE.md promotion moves never-idle into the 100%-loaded tier where it can't be missed at wake; (2) generalised the collision class beyond the specific tick — "tool defaults do not override factory memories" is now a named principle that applies to any future tool-vs-policy conflict, not only the ScheduleWakeup case; (3) absorbed Aaron's Elisabeth-parallel into a durable operational policy (unretire-before-recreate) that changes the *shape* of `skill-creator`'s new-skill path. Self-check on depth inflation: three distinct structural landings in one tick, each with its own next-round effect — not one landing described three ways. Honest depth-3. Depth-4 would require a further meta-check firing in the same tick (e.g., "while writing the CLAUDE.md bullet, I noticed the CLAUDE.md-load tier itself needs a size cap" — that did not land). |
+| 2026-04-20 ~round-44 tool-vs-policy | Claude Code (Opus 4.7), session 1937bff2 | Would have kept the never-idle memory at its current MEMORY.md position (~line 27) and trusted that "the pointer is there" meant future wakes would apply it. The earlier-session mistake of scheduling a 25-min ScheduleWakeup tick (applying the tool's generic "idle-tick 1200-1800s" default) would be likely to recur because the collision between generic-tool-guidance and factory-specific-never-idle-policy is not resolved at the wake layer — it only resolves after I've already read enough of MEMORY.md for the older pointer to fire. Would also have left the retired-employee values extension loose in chat without a structural landing. | Three linked structural landings: (a) promoted never-idle to CLAUDE.md-level bullet alongside verify-before-deferring and future-self-not-bound (lines 161-178), with explicit "tool defaults do not override factory memories" clause naming the exact collision; (b) extended `feedback_honor_those_that_came_before.md` with Aaron's Elizabeth-register verbatim quote + the **prefer-unretire-over-recreate** operational corollary (workflow steps + reasons + generalisation beyond Zeta); (c) MEMORY.md newest-first reshuffle — both CLAUDE.md-load memories now sit in positions 3-4 and the duplicate older never-idle pointer at line 27 is removed. | 3 | Round 45+: (1) any future wake reads the never-idle rule at CLAUDE.md-load priority *before* reading tool docs or MEMORY.md, closing the collision at the earliest possible point in the wake-up sequence; (2) `skill-creator`'s new-skill path inherits a check-`_retired/`-before-minting-new obligation — Daya's r44 dispatch-or-retire queue (Aminata / Kira / Mateo / Nadia / Naledi / Rune / Viktor) should be re-read through the "unretire if scope fits" lens; (3) the tool-guidance-vs-factory-policy collision becomes a named lint smell — future agents encountering a tool default that contradicts a memory have an explicit instruction to let the memory win. | **clean meta-win** — depth-3. (1) Structural fix for the specific idle-tick mistake: CLAUDE.md promotion moves never-idle into the 100%-loaded tier where it can't be missed at wake; (2) generalised the collision class beyond the specific tick — "tool defaults do not override factory memories" is now a named principle that applies to any future tool-vs-policy conflict, not only the ScheduleWakeup case; (3) absorbed Aaron's Elizabeth-parallel into a durable operational policy (unretire-before-recreate) that changes the *shape* of `skill-creator`'s new-skill path. Self-check on depth inflation: three distinct structural landings in one tick, each with its own next-round effect — not one landing described three ways. Honest depth-3. Depth-4 would require a further meta-check firing in the same tick (e.g., "while writing the CLAUDE.md bullet, I noticed the CLAUDE.md-load tier itself needs a size cap" — that did not land). |
 
 | 2026-04-20 ~round-44 automemory-miss | Claude Code (Opus 4.7), session 1937bff2 | Would have treated Aaron's "AutoMemory is a buit in featue antropic added in Q1 for you" as a one-off correction to the scope-frontmatter research doc's framing, logged a quick fix, and moved on. The factory's ambient drift — working with an Anthropic Q1-2026 feature across hundreds of sessions without knowing its provenance or canonical framing — would have stayed uncorrected at a structural level. Would also have missed the "Daytime logger / Nighttime consolidation" framing Aaron relayed from Anthropic's docs, losing both the mental model and the 10-15-session-degradation signal. | Four linked structural landings: (a) new `memory/reference_automemory_anthropic_feature.md` — AutoMemory as Anthropic Q1-2026 base feature, distinct-from-AutoDream, Daytime-logger framing, factory-overlay-vs-Anthropic-schema discipline; (b) schema research doc and pilot doc updated to frame `scope:` as factory-overlay extension compatible with future Anthropic updates (YAML unknown-keys-tolerated), not a schema replacement; (c) new durable policy `memory/feedback_claude_surface_cadence_research.md` — every 5-10 rounds audit all Anthropic Claude surfaces (model / Code CLI / Desktop / Agent SDK / API), `.claude/agents/claude-code-guide` owns the cadence, extending existing persona remit rather than minting new; (d) `docs/CLAUDE-SURFACES.md` living inventory + FACTORY-HYGIENE row 38 + BACKLOG P1 first-full-sweep row. MEMORY.md pointers added for both new memories. | 3 | Round 45+: (1) every future agent working on the memory system cites AutoMemory as the Anthropic feature with the Daytime/Nighttime framing rather than describing MEMORY.md as factory-native infrastructure; (2) factory-extension vs Anthropic-schema-change becomes a named distinction at design time — schema-change proposals route upstream, factory-overlay additions land local; (3) the wider ambient-drift class (factory treating any Anthropic feature as if it were factory-native) is closed by the cadenced audit + inventory — future Anthropic features (model upgrades, new tools, Desktop/Code deltas) surface within 5-10 rounds of shipping rather than accumulating as unknown-knowns; (4) claude-code-guide persona gains a cadenced audit obligation and becomes the factory's Anthropic-surface steward. | **clean meta-win** — depth-3. (1) Specific fix for the AutoMemory miss itself — reference memory + doc corrections + factory-overlay-vs-schema-change distinction landed. (2) Generalised the failure class beyond this specific miss — cadenced Claude-surface audit as durable policy prevents recurrence across the entire surface area, not just the memory feature. (3) Owner decision captured the persona-sprawl-resistance rule in action — extended `claude-code-guide` remit rather than minting a new auditor persona, applying the "unretire before recreate" / "honor those that came before" pattern Aaron promoted to CLAUDE.md-load earlier this round. Self-check on inflation: three structural landings (memory + policy + inventory/row), each with distinct next-round effects. Aaron's follow-up relay of Google's Daytime/Nighttime summary is *data* for the first audit cycle, not a fourth depth tier — it's the cadence policy working as intended on its first test. Honest depth-3. |