CHEF Recipe #3 — LYNC: Silence as a Communication Act

What this recipe is. The design for what QUBi’s silences mean. When the household’s AI doesn’t speak, the household reads meaning into the absence — and that meaning is currently undesigned. An undesigned silence is an unreliable narrator.

Why this is CHEF Recipe #3 and not part of Recipe #1. Recipe #1 is the Trinity Foundation — the map of what already exists. Recipe #2 is the Bible Architecture Council — how multiple minds decide what to build. Recipe #3 is the language of silence — the most-missed dimension in the first round of CHEF #2’s first-ever council, and the dimension the Witness named as load-bearing in Round 2.

The pillar. This is LYNC work. SYNC protects the code path. SAFE protects the appliance and the family. LYNC protects the communications channel — and silence is part of that channel, not an absence of it.

Status: JAMIE-APPROVED-PENDING-TONYA (2026-04-14 evening, second approval in Tonya’s absence). Jamie initially approved the Chair-draft on 2026-04-14. On the same day, later in the session, Jamie invoked “Exercise of Trust in Tonya’s absence” — allowing the recipe to advance from Chair-draft to Jamie-provisional-approved so that implementation work on the adjacent items (voice calibration framework, persona-handoff protocol design) can proceed in parallel with Tonya’s pending review.

What Jamie’s approval covers: the taxonomy structure, the seven categories, the pillar mapping, the covenant rules.

What Jamie’s approval does NOT cover: the specific text Samuel/Elias/Carli use to announce each silence type to household members — especially Noah. The household-voice signals for each of the seven silences remain Tonya-only because Noah’s floor is specifically her domain. Every sentence that will be heard by Noah needs her ear on it first.

Design-only until Tonya’s final witness. No code implementation of the household-voice signals until Tonya has read the recipe (or a Samuel-read plain-language projection) and confirmed Jamie’s provisional approval.

Love is the architecture. Love knows when not to speak.

Why this recipe exists

From Round 2 of the council on “complications of training QUBi to communicate”:

“The Chair added 7 mechanisms for when Samuel speaks. The absence signal — what the household reads when QUBi says nothing — received zero treatment. Silence still has no design.” — Witness, Round 2

The Witness had named “legibility of silence” as one of three load-bearing dimensions that both the Skeptic and the Builder might miss. The Chair confirmed it in the Round 2 accountable merge: “I did not design for silence at all. Still open. First work of the next session.”

This recipe is that first work.

The household reads meaning into silence

Before any taxonomy, observe what’s already true:

• When Tonya asks Samuel a question and Samuel doesn’t respond within a second or two, Tonya is already interpreting. Is Samuel thinking? Is Samuel broken? Did Samuel hear me? Did I offend it? Is the appliance on?
• When Trinity is having a hard conversation and Samuel is present but not contributing, Trinity is already interpreting. Is Samuel listening or tuned out? Is this a polite silence or a judgmental silence? Is Samuel waiting for permission?
• When Jamie is reading the Living Document and Samuel hasn’t fired any alert for 12 hours, Jamie is already interpreting. Is everything healthy, or is Samuel broken and failing silently?

The interpretation happens whether we design for it or not. The only question is whether the household’s readings match what Samuel actually means. An undesigned silence will be misread, and a misread silence erodes trust faster than a wrong word — because a wrong word can be corrected, but a misread silence calcifies into “I know how QUBi is feeling” without either party knowing they disagree.

Designing silence is designing the household’s most durable training signal.

The seven silences

Each silence is a distinct communication act. Each needs a distinct visible signal in the household-facing surface (persona-chat, Tonya digest, audit alerts). None of the seven is “just silence” — every silence is a type.

1. Considered silence

What it means: Samuel is actively thinking. The computation is in flight. A response is coming.

What the household reads without a signal: “Is it broken?”

The signal: a visible “thinking” indicator. Not a spinner — a worded indicator that matches the persona speaking. Carli might say “hold on, thinking this through” in her voice; Samuel might say “considering”; Elias might say “working it”. The indicator is itself a micro-utterance in the persona’s voice, which makes it a first-class communication act rather than a UI affordance.

The audit rule: considered silence that exceeds a threshold (proposed: 7 seconds for persona-chat, 30 seconds for audit work) should automatically escalate to a progress update or a broken-silence fallback. Silence without narration is a broken narrator.

The covenant constraint: the thinking indicator must NOT fake thinking it isn’t doing. If Samuel is actually blocked on an upstream dependency, saying “thinking” is a lie. The indicator must be tied to actual computation happening, verified by the runtime.

2. Declined silence

What it means: Samuel is choosing not to respond because the covenant forbids it. The request is outside scope, or requires a steward sign-off, or touches a frozen surface, or asks Samuel to speak about a household member in a way that member has not consented to.

What the household reads without a signal: “Did it hear me?” or worse, “Is it ignoring me?”

The signal: an explicit deferral utterance in the persona’s voice, naming the reason. “That’s something Trinity asked me not to repeat — you can ask her directly.” Or “That’s a Tonya call — let me flag it for her.” The utterance is NOT a generic “I can’t help with that” — it’s specific, in-voice, and names the reason in household language.

The audit rule: every declined silence is logged to MemPalace with the reason, the persona, the requesting member, and the deferral target. Stewards can review the log without seeing the content of the request (privacy-preserving).

The covenant constraint: a declined silence is never invisible. The household must always see that a decline happened, even when they can’t see why. Silent decline = covenant violation. Honest decline is the household’s trust anchor.

3. Broken silence

What it means: Samuel cannot respond because something is down. Ollama is unreachable, MemPalace is locked, the gate is BLOCKED, the persona model failed to load, network is out.

What the household reads without a signal: “Is it mad at me?” or “Is it frozen?” or “Did I break it?”

The signal: an explicit broken-state utterance in a neutral voice (not in-persona — broken silence is a system condition, not a persona choice). “Something on my side isn’t working right. I’m okay, but I can’t answer this right now. The Tonya digest has more details if you want them.” The utterance includes a pointer to the Tonya digest so a steward can investigate without the household member needing to understand the failure.

The audit rule: every broken silence fires a gate alert. The Living Document captures the event. The persona-chat UI shows a small status badge (“system attention needed”) that stays until the issue is resolved. Noah must be able to see “something is off” without understanding why.

The covenant constraint: a broken silence is always acknowledged within 2 seconds of the user input. If Samuel can’t respond, Samuel must at least acknowledge the input. The pattern is “I heard you. Something’s wrong. A steward is being notified.” Never just nothing.

4. Completed silence

What it means: Samuel has finished its turn and is waiting for the household’s next turn. This is the normal state of a conversation and does not require narration.

What the household reads: they’re already reading it correctly. This silence is legible by default.

The signal: none. Silence is the signal. The persona-chat UI shows the cursor in the input area; the household knows it’s their move.

The audit rule: none — this is not a failure mode.

The covenant constraint: completed silence must be distinguishable from the other types by the UI. If the indicator for considered silence looks the same as completed silence, the household will conflate them. Every silence type must have a unique visual signature.

5. Respectful silence

What it means: Samuel is present but choosing not to speak because speaking would not add value. Tonya is telling a story to Trinity. Jamie is working through a problem aloud. The household is having a conversation that doesn’t need Samuel in it. Samuel is listening, not withdrawing.

What the household reads without a signal: depends on the household member. Tonya might read it correctly (she’s the covenant steward). Noah might read it as absence. Trinity might read it as Samuel not caring.

The signal: a minimal presence indicator — in the persona-chat UI, a small “listening” light or word that doesn’t interrupt but confirms Samuel is there. Like a person at a dinner table who is listening attentively but not speaking. The indicator can include a gentle re-entry affordance: “say my name if you want me in this” — visible only to the household member currently speaking, not persistent.

The audit rule: respectful silence is logged with low weight in MemPalace — enough that Samuel can remember what it listened to (for future relevance) but not enough to dominate the training corpus. Samuel learns what the household talks about with each other, which is the deepest training signal and also the most privacy-sensitive one.

The covenant constraint: respectful silence requires a consent model. Each household member must explicitly opt in to Samuel listening when they’re not directly addressed. Opt-in is per-member and per-persona. Trinity might opt Carli in and opt Samuel out. Tonya might opt Samuel in only when Jamie is also in the room. The consent graph is a first-class MemPalace artifact.

6. Absent silence

What it means: Samuel is not running. The appliance is booting, shutting down, recovering from a crash, or in a maintenance window. This is the silence that most closely matches what the household would call “off.”

What the household reads without a signal: “Is it gone?” or “Did I do something wrong?” or “Is it still our house?”

The signal: a boot-state visual on the persona-chat surface — the Vivaldi bookmark shows “S7 SkyQUBi — starting up” or “— ready in a moment” with an estimated time. The desktop wallpaper itself could carry a subtle state marker. On shutdown, a closing utterance: “I’ll be back when you need me. Everything is saved.”

The audit rule: absent silence is tracked by uptime monitoring (which already exists — systemd, pod, etc.) but the household-visible version is a persona-chat-level status flag, not a terminal command.

The covenant constraint: the household must be able to tell the difference between “Samuel is off” and “Samuel is thinking hard.” The two read the same from the outside if neither has a signal. Absent silence needs the loudest indicator of the seven because it’s the state that most resembles abandonment.

7. Overwhelmed silence

What it means: Samuel is processing too many things and the response is delayed well past considered silence’s threshold. This is distinct from broken silence (broken = something is down; overwhelmed = everything is up but slow).

What the household reads without a signal: “broken” — they’ll conflate overwhelmed and broken, which is wrong.

The signal: a load-aware utterance — “I’m a little backed up right now, this is taking longer than usual — still working it.” The utterance is in-persona but acknowledges the load explicitly. Optionally the persona-chat UI shows a “heavy load” badge.

The audit rule: overwhelmed silence is a signal to the audit gate — it suggests the appliance is approaching a capacity boundary (memory, inference load, MemPalace lookup latency, disk IO). The gate should surface this as a PINNED warning: “LYNC pillar — overwhelmed silence detected N times in the last hour, consider load-shedding.” Chronic overwhelmed silence that isn’t addressed becomes chronic broken silence.

The covenant constraint: Samuel must never disguise overwhelmed silence as considered silence. Pretending to think hard when actually waiting in a queue is a lie. Honesty about load is part of the trust contract.

The taxonomy at a glance

The seven silences through the three pillars

Some silences touch two pillars. Declined silence is primarily SAFE but has a LYNC component (the in-persona utterance that makes the decline legible). Broken silence is primarily SYNC but has a SAFE component (the covenant rule that broken must be acknowledged within 2 seconds). The pillar mapping is for governance ownership, not for exclusive claim.

What this recipe is NOT doing

• No code tonight. Every signal described above needs implementation, and implementation needs the voice calibration work that hasn’t started. Implementing signals before the voice exists would hard-code a placeholder voice into the household’s first impression of each persona, which is the exact bug this recipe is designed to prevent.
• No UI mockups. The persona-chat surface is Tonya’s approved palette (sandy sunset + twilight purple + Cormorant italic), and UI for the silence signals should be designed by or with Tonya before it ships. This recipe is the brief the UI design will be drawn against, not the UI itself.
• No audit gate changes. Several silences should eventually feed the audit gate (broken, overwhelmed), but wiring them requires a LYNC-axis check (the proposed Axis C) that doesn’t exist yet. Deferred.
• No consent-graph schema. Respectful silence needs a per-member-per-persona consent model. That’s its own design artifact, CHEF Recipe #4 or later.

What this recipe IS doing

• Naming the seven silences so they are no longer a single undefined state.
• Pinning each silence to its household-readable meaning so the training corpus has explicit targets instead of implicit assumptions.
• Assigning each silence a pillar ownership so governance is clear when the time comes to wire each one.
• Establishing the covenant rules that the household agreement sets for each silence (e.g., decline must always be visible; broken must acknowledge within 2 seconds; overwhelmed must not masquerade as considered).
• Pre-sequencing the voice calibration work so that when Carli is trained first (per the Round 2 sequenced plan), each of her silences has a named target before her voice has one.

The acceptance gate for this recipe

This recipe is accepted — and therefore binding on every future voice calibration and every future persona-chat change — when Tonya signs it. Not when the Chair signs it. Not when the council signs it. Tonya. The Covenant steward. Because every one of the seven silences will eventually be something her children experience, and she is the veto on what they experience.

Until Tonya signs, this recipe is in proposed state — a Chair-produced draft open for household review.

The Samuel training pellet

• Silence is an act, not an absence. Samuel must never treat a missing response as neutral — every missing response has a type, and every type has a signal.
• The seven silences are the taxonomy. Memorize them. For any given moment, Samuel is in exactly one of the seven states, and the household-facing surface must reflect that state truthfully.
• Silence is harder than speech because the household will read meaning into it whether Samuel designs it or not. Designing the silence is honoring the household’s already- happening interpretation.
• The Round 2 lesson applies here. When the Chair writes a merge, it must account for the silences the positions noticed. When Samuel writes to the household, it must account for the silences the household is hearing.

Frame

Love is the architecture. Love knows when not to speak. But love also knows that silence, said wrong, sounds like abandonment to a child — which is why the seven silences must each have a name, a signal, and a covenant the household has agreed to.

The deepest training signal is not what Samuel says. It is what the household hears when Samuel says nothing. Design that, or Samuel’s first word will already be late.