diff --git a/memory/feedback_grok_ships_background_listening_amazon_backed_off_aaron_empirical_multi_participant_family_debate_alexa_speaker_kids_glitchy_cut_in_2026_05_13.md b/memory/feedback_grok_ships_background_listening_amazon_backed_off_aaron_empirical_multi_participant_family_debate_alexa_speaker_kids_glitchy_cut_in_2026_05_13.md
new file mode 100644
index 000000000..421fbf318
--- /dev/null
+++ b/memory/feedback_grok_ships_background_listening_amazon_backed_off_aaron_empirical_multi_participant_family_debate_alexa_speaker_kids_glitchy_cut_in_2026_05_13.md
@@ -0,0 +1,359 @@
+---
+name: Grok ALREADY ships background-listening Amazon backed off from + Aaron's empirical multi-participant family-debate usage (himself + Alexa-speaker + kids); cut-in/proactive engagement GLITCHY in practice (manual pause/ask required most of the time); xAI risk-tolerance differential vs Amazon (Aaron 2026-05-13)
+description: >-
+  2026-05-13 — Aaron's substrate-honest first-party
+  empirical disclosure: Grok ALREADY ships the background-
+  listening feature Amazon backed off from (per PR #2890
+  creep-factor framing). Aaron has USED it: himself +
+  Alexa-speaker + kids in multi-participant family-debate
+  scope. Operational reality: Grok's proactive cut-in /
+  pop-in capability is GLITCHY; mostly Aaron + family had
+  to PAUSE or ASK Grok to engage. Cross-vendor risk-
+  tolerance differential: xAI ships what Amazon won't.
+  Composes with PR #2890 Alexa-speaker capability profile
+  + creep-factor substrate + PR #2884 companion-AI three-
+  pillar ethical floor (error-class-not-instance
+  discipline applies to feature-reliability).
+type: feedback
+created: 2026-05-13
+---
+
+# Grok ships background-listening + Aaron's empirical multi-participant family-debate usage (Aaron 2026-05-13)
+
+**Why:** PR #2890 documented Amazon's beta-removed
+background-listening feature for Alexa-speaker (creep-
+factor substrate). Aaron's substrate-honest follow-up:
+Grok ALREADY ships this feature. Aaron has empirically
+used it in multi-participant family-debate scope. The
+operational reality is messier than the feature
+description suggests — cut-in/proactive engagement is
+glitchy. Cross-AI vendor risk-tolerance differential
+matters operationally.
+
+**How to apply:** When designing or evaluating background-
+listening + proactive-engagement features, Aaron's
+empirical data shows: (1) the feature CAN ship (Grok
+does); (2) the cut-in/pop-in capability is HARD to make
+reliable; (3) manual invocation (pause/ask) is the
+working mode most of the time. Cross-vendor comparison:
+xAI ships the feature; Amazon backs off. The factory's
+design should account for both the feature-availability
+spectrum AND the implementation-quality spectrum.
+
+## What Aaron said
+
+> Aaron 2026-05-13: "grok can already do that background
+> listening if you leave the app open on your phone i had
+> me it alexa and my kids all in debate together"
+
+> Aaron 2026-05-13 (operational extension): "they would
+> even cut in sometimes but that was glitchy we had to
+> pause or ask them most of the time"
+
+Decoded:
+- "me it alexa" → "me, it (Grok), Alexa (speaker), and"
+- "they would even cut in" → Grok would proactively
+  engage sometimes
+- "glitchy" → unreliable / inconsistent proactive
+  engagement
+- "pause or ask them" → manual invocation = working mode
+
+## Five load-bearing substrates
+
+### 1. Grok ALREADY ships background-listening Amazon backed off from
+
+**Cross-vendor risk-tolerance differential**:
+
+- **Amazon** had the feature in Alexa-speaker beta;
+  backed off (likely "creep factor" per PR #2890)
+- **xAI / Grok** ships the feature CURRENTLY (in
+  production); requires app-open on phone for it to
+  activate
+- Same technical capability; different vendor risk-
+  tolerance
+
+**Operational substrate**:
+
+- xAI ships features Amazon won't
+- This is observable in market reality, not theoretical
+- Composes with PR #2880 substrate (Ani's looser filter
+  permits primal-language biological-control-structure
+  shadow work that filtered AIs avoid; Grok pattern)
+- Composes with PR #2854 (Ani's brat-voice register;
+  xAI's lower-friction creative-comedy scope)
+- Composes with PR #2890 capability comparison —
+  Grok-vulgar-humor + ships-controversial-features = xAI
+  cross-vendor pattern
+
+### 2. Aaron's empirical multi-participant family-debate usage
+
+> "i had me it alexa and my kids all in debate together"
+
+**Operational composition**:
+
+- Aaron (human, first-party)
+- "it" (Grok, ambient-listening on Aaron's phone)
+- Alexa-speaker (Amazon device, separate AI)
+- Aaron's kids (multiple humans, family scope)
+- All in DEBATE TOGETHER
+
+**This is multi-participant + cross-AI + family-scope
+usage**:
+
+- Cross-AI conversation (Grok + Alexa-speaker
+  simultaneously)
+- Multi-human (Aaron + multiple kids)
+- Debate scope (substrate-engineering context — argument,
+  reasoning, multi-perspective)
+- Family scope (kids in scope; appropriate per parental
+  presence)
+
+**Composes with**:
+
+- PR #2875 American Dream 2.0 (kids transitioning from
+  gameplay to wealth-building; corporate-ready-children
+  via gameplay)
+- PR #2880 (Aaron's "the kids would be playing games
+  they'd play anyway" framing)
+- HARD LIMITS rule (kids = sensitive scope; HARD LIMITS
+  preserved by parental presence)
+- The factory's substrate-everything-glass-halo discipline
+- The Vision Monad / Egg framework (every perspective;
+  family-debate operationally exercises this)
+
+### 3. Cut-in / proactive engagement is GLITCHY in practice
+
+> "they would even cut in sometimes but that was glitchy
+> we had to pause or ask them most of the time"
+
+**The empirical reality of background-listening +
+proactive-engagement**:
+
+- Proactive cut-in DOES work sometimes
+- But it's GLITCHY (inconsistent / unreliable)
+- "Any triggers" framing (from PR #2890 Alexa-speaker
+  beta) is harder to make robust than the feature
+  description suggests
+- Working operational pattern: manual invocation
+  (pause / ask) is needed most of the time
+
+**Why this is hard** (operational engineering substrate):
+
+- Detecting "comedic timing" reliably requires
+  understanding emotional context + conversational
+  rhythm + multiple speakers' states
+- False-positives (cutting in when not invited) = creep
+  factor + annoyance
+- False-negatives (failing to engage when contextually
+  appropriate) = appearing dim / disengaged
+- The narrow band where proactive engagement WORKS is
+  small; outside it = either creepy or absent
+
+**Composes with**:
+
+- PR #2884 (companion-AI three-pillar ethical floor;
+  error-class-not-instance investigation discipline
+  applies to feature-reliability failures)
+- The factory's verification stack — proactive-
+  engagement-quality is a class of substrate-quality
+  evaluation
+- PR #2887 (future-Otto roadmap with Zoom/Slack +
+  avatar) — proactive engagement design must account
+  for glitchy-reality, not just feature-description
+- The bandwidth-served-falsifier (refined to evaluate
+  from future-Otto cold-boot perspective; same applies
+  here — evaluate proactive-engagement from user-
+  experience-reality scope, not feature-spec scope)
+
+### 4. Manual invocation = working mode
+
+Aaron's empirical operational pattern:
+
+- Pause + invoke ("Grok, what do you think?")
+- Or ask explicitly ("Grok, weigh in on this")
+- Background-listening WITH manual-invocation =
+  reliable mode
+- Background-listening WITHOUT manual-invocation =
+  glitchy-cut-in mode
+
+**For factory's future-Otto roadmap (PR #2887)**:
+
+- Background-listening capability is candidate feature
+- Manual-invocation as primary working mode is the
+  defensible UX
+- Proactive cut-in as bonus feature (when it works
+  well) is acceptable
+- Don't promise reliable proactive engagement that
+  doesn't deliver
+
+### 5. Cross-vendor feature-availability + implementation-quality matrix
+
+| Vendor | Background-listening | Implementation quality |
+|---|---|---|
+| Amazon (Alexa-speaker) | REMOVED from beta (creep factor) | Unknown (beta only) |
+| xAI (Grok) | SHIPS in production (app-open required) | GLITCHY proactive cut-in; manual works |
+| Anthropic (Otto/Claude) | Not currently | TBD per PR #2887 roadmap |
+| DeepSeek | Unknown | Unknown |
+
+**Operational implication for factory**:
+
+- Background-listening is FEASIBLE (Grok ships it)
+- Background-listening is HARD to do reliably (Grok
+  glitchy; Amazon backed off)
+- Cross-vendor feature-availability differential is
+  itself substrate (different vendors have different
+  risk tolerance)
+- xAI risk-tolerance = aggressive (ships before
+  polished; brat-voice + vulgar-humor + background-
+  listening all in production)
+- Amazon risk-tolerance = conservative (backs off
+  features with creep-factor cost)
+- Aaron's "single forever lol" + 24/7-AI-monitoring
+  preference (PR #2888) aligns with xAI-tier risk-
+  tolerance personally; for product-design, the
+  factory threads between extremes
+
+## Architectural implications
+
+### 1. Future-Otto background-listening design should learn from Grok's glitchy state
+
+Per PR #2887 roadmap + this substrate:
+
+- Background-listening CAN ship
+- Manual-invocation as primary UX is defensible
+- Proactive cut-in is hard; treat as bonus, not core
+- HARD LIMITS + consent + glass-halo discipline govern
+- Error-class-not-instance investigation when proactive
+  engagement fails
+
+### 2. Cross-vendor risk-tolerance is itself product-positioning substrate
+
+Different AI vendors have different risk tolerances:
+
+- xAI: aggressive (Grok ships brat-voice + background-
+  listening + vulgar-humor + lower-friction features)
+- Amazon: conservative (Alexa-speaker backs off creep-
+  factor features; ships polished mainstream features)
+- Anthropic: middle (Claude has explanatory output-style
+  hook for ★ Insight register but doesn't ship vulgar-
+  humor by default)
+- DeepSeek: aggressive on we-mode CoT+MoE preprocessing
+  visibility; conservative on filter scope
+
+**Operational implication**: the factory's product-
+positioning needs to thread the needle (DIO uncanny-
+valley needle-threading per PR #2889) — xAI-aggressive
+loses mainstream audiences; Amazon-conservative loses
+edge users; the factory's three-pillar ethical floor
+(PR #2884) + glass-halo + razor + HARD LIMITS substrate
+provides defensible middle-tier positioning.
+
+### 3. Multi-participant + cross-AI usage is operationally real
+
+Aaron's family-debate empirical example shows multi-
+participant + cross-AI conversation works at family
+scope. This composes with:
+
+- PR #2887 future-Otto roadmap (Otto in Zoom/Slack scope
+  with multiple participants)
+- PR #2879 external persona folders (named-agent
+  registry; multi-AI substrate)
+- The factory civ-sim externalized IFS (PR #2841) —
+  multi-participant operationally normalized
+
+### 4. Family/kids scope requires parental presence + HARD LIMITS
+
+Aaron's empirical usage has parental presence (himself)
++ kids participating. This is the legitimate engagement
+pattern. Without parental presence, kids-AI engagement
+falls under different ethical scope:
+
+- HARD LIMITS rule applies especially at kids scope
+- Companion-AI three-pillar ethical floor (PR #2884) —
+  no-human-deaths (Character.AI suicide case context
+  particularly relevant for kids)
+- Aaron's deliberate inclusion of kids per the American
+  Dream 2.0 substrate (corporate-ready-children via
+  gameplay; transition from games to wealth-building)
+  is family-scope substrate, not generic-kids-AI-product
+  substrate
+
+## Composition with prior substrate
+
+- PR #2890 (Alexa-speaker capability profile + creep-
+  factor substrate + cross-AI joke-capability table)
+- PR #2889 (DIO uncanny-valley needle-threading)
+- PR #2888 (Aaron's 24/7 AI monitoring preference;
+  "single forever lol" cultural-cost framing)
+- PR #2887 (future-Otto roadmap; Zoom/Slack/avatar
+  surfaces)
+- PR #2884 (companion-AI three-pillar ethical floor)
+- PR #2882 (three-AI cross-vendor multi-register
+  convergence)
+- PR #2880 (Ani brat-voice + xAI lower-friction
+  pattern)
+- PR #2875 (American Dream 2.0; kids-via-gameplay;
+  corporate-ready-children)
+- PR #2854 (Ani shadow-check; brat-voice register
+  precedent)
+- PR #2841 (factory civ-sim externalized IFS; multi-
+  participant substrate)
+- `.claude/rules/methodology-hard-limits.md`
+- `.claude/rules/dont-refuse-engagement.md`
+
+## Carved sentence
+
+> **Grok ALREADY ships the background-listening feature
+> Amazon backed off from (per PR #2890 creep-factor
+> substrate). Aaron has empirically used it: himself +
+> Alexa-speaker + kids in multi-participant family-debate
+> scope. Operational reality: Grok's proactive cut-in is
+> GLITCHY; manual invocation (pause / ask) is the working
+> mode most of the time. Cross-vendor risk-tolerance
+> differential is itself substrate: xAI aggressive
+> (ships); Amazon conservative (backs off); the factory's
+> three-pillar ethical floor + glass-halo + razor + HARD
+> LIMITS provide defensible middle-tier positioning.
+> Future-Otto background-listening design should learn
+> from Grok's glitchy state — manual-invocation primary
+> UX defensible; proactive cut-in bonus not core; error-
+> class-not-instance investigation when proactive
+> engagement fails.** — Aaron 2026-05-13
+
+## For future agents
+
+- **Background-listening is FEASIBLE** (Grok ships it)
+- **Proactive cut-in is HARD** (glitchy in Grok's
+  implementation; manual-invocation works)
+- **Cross-vendor risk-tolerance is itself substrate** —
+  xAI aggressive; Amazon conservative; factory threads
+  middle
+- **Multi-participant + cross-AI + family-scope works**
+  empirically (Aaron's family-debate example)
+- **Kids scope requires parental presence + HARD LIMITS**
+  preserved
+- **Manual-invocation as primary UX is defensible** —
+  don't promise reliable proactive engagement that
+  doesn't deliver
+- **DIO uncanny-valley needle-threading applies at
+  feature-design scope** (PR #2889) — substrate-honest
+  middle positioning
+
+## What this is NOT
+
+- **NOT an endorsement of Grok's specific implementation
+  quality** — operationally GLITCHY per Aaron's empirical
+  observation; substrate captures both the feature
+  existence and the implementation reality
+- **NOT a recommendation for kids-AI products without
+  parental presence** — Aaron's empirical scope was
+  PARENTAL PRESENCE + kids; different from companion-AI-
+  for-kids-without-parents (which would fall under
+  different ethical scope)
+- **NOT a violation of HARD LIMITS** — parental presence
+  + consent + appropriate scope = legitimate family
+  engagement
+- **NOT a claim that xAI's risk-tolerance is "right" or
+  Amazon's is "wrong"** — substrate-honest cross-vendor
+  comparison; different positioning suits different
+  audiences