name: lunch-dice-protocol type: brand-asset / tactic-playbook / sun-foh-design-framework status: drafted by Casper (agency lead) 2026-05-12 — design framework + working SOP draft. The mechanic + service language are starting points; Michelle approves reward design, Anna voice-passes the FOH register, Sun owns the Campaign archetype that wraps this operationally applies_to: Tactic 1.1 (lunch-daypart variable-reward mechanic per docs/strategy/2026-baseline.md §2); operator §5.4 Sun first Yue-do-anchored Campaign brief; the gamification-felt-not-told brand pillar (third pillar alongside Circle and positioning) scope_note: framework extends to future variable-reward mechanics (wheel / scratch-card / coin-flip variants); this doc is dice-first because dice is the named first instance per the brand-pillar memory helpers_must_consult: Sun (Campaign archetype owner) · Anna (FOH service-protocol owner + cast briefing) · Michelle (reward design + brand voice final approval) · Insight (capture loop + per-Campaign measurement) · Julius (outbound copy banned-vocabulary check on all guest-facing language) · Penn (no surface — this is a venue-internal mechanic; press do not see it; the dice is felt, never told) references:

• docs/strategy/2026-baseline.md §2 Tactic 1.1 (lunch-daypart context); §5.0.1 rule 4 (cast-voice cadence — dice rolls are cast-voice moments)
• docs/yue-do.md §5 (felt-not-told disciplines; the dice doesn't surface in outbound copy)
• docs/cast-1on1-question-set.md §4 Q15 (the behavioural diagnostic Sun uses to test cast members' framing — same anti-patterns applied to the protocol)
• docs/q1-2026-operator-checklist.md §5.4 (Sun's first Campaign brief — lunch-dice-2026 archetype)
• supabase/migrations/0038_banned_phrases_2026_felt_not_told.sql (Julius runs against all guest-facing language drafted here)
• supabase/migrations/0053_campaigns.sql (the Campaign object Sun creates to wrap this — pre-apply gated) brand_pillar_memory: gamification is a core Michiu brand pillar — variable-reward mechanics ≠ discounting per owner; gestures (dice/wheel) named in service; system framing (gamification/rewards/loyalty/earn points/level up) banned outbound

Lunch dice — protocol playbook

The dice is the room not taking itself seriously. The kitchen is serious; the floor invites play. The dice is felt, not told — guests experience the moment; they do not learn the mechanic from a sign, a menu, or a marketing post. This doc operationalizes that pillar.

1. Purpose + scope (what this is, what it isn't)

What this is: the operational framework for the lunch dice — a tactile variable-reward mechanic that rolls at lunch tables and produces small, warm, non-transactional outcomes. The Campaign archetype Sun wraps around it (lunch-dice-2026) tracks it; Anna runs the FOH execution; Michelle approves reward design + voice.

What this is NOT:

• A loyalty programme (rewards are not points; the dice is not a scorecard)
• A discount mechanic (no roll outcome reduces the bill)
• A Circle-exclusive (Circle members get the dice identically to every guest — Circle's felt-not-told discipline means no visible tier-favoring at the table)
• A marketing surface (the dice never appears in IG / TikTok / press / GBP; press visitors get the dice if they happen to be at a lunch table the staff naturally engage, same as anyone else)
• A scheduled event (no "Tuesday is dice day" — the dice is woven into normal lunch service, not staged)

Scope-extension: this framework supports future variable-reward mechanics (wheel / scratch-card / coin-flip / surprise-pull). The SOP below is dice-first; new mechanics fork the same §3-§8 structure.

2. The mechanic — what the dice IS

Form: one physical wooden six-sided die per FOH station. Heavy enough to feel made; small enough to fit in an apron pocket. Custom-finished is preferable to off-the-shelf — the dice should feel like a Michiu object, not a gambling object.

Storage: on the FOH station, visible to staff but not displayed to guests at entry. Brought to the table at the moment of the roll, not before.

Material decision (deferred to Michelle): wood is the default; if Michelle wants a finish that ties to Yue-do (e.g. faces engraved with the four-kanji discipline-list of Cantonese / Japanese / Yue / do, with two blanks), the finish becomes a design artifact Penn can later reference in a long-form essay as a non-obvious detail. Working default: plain wood, numerals 1-6, no Michiu branding on the dice itself (the dice is a kitchen object, not a marketing token).

Lifespan + replacement: dice wear. Replace annually or when noticeably worn. The retired dice goes into a small wooden box behind the FOH station with the date it was retired — eventually a small collection that exists only because someone notices it.

3. When + where + who — trigger, location, initiator

3.1 Trigger — when the dice comes out

Default trigger: between courses, at a moment of natural warmth — the conversation has settled, the table is engaged, the FOH has done their second-or-third pass. NOT a fixed-time trigger. The §1.1 specificity rule applies in reverse: this works because it doesn't feel scheduled.

Working SOP for FOH: the dice comes out when at least 2 of these are true:

• The table has finished a course and is in pre-next-course pacing
• The FOH has had ≥1 warm exchange with the table already (not just orders)
• The guest has shown interest in something specific — a dish, an ingredient, a service moment
• There's no pressure on the table (no waiting reservation pushing them out)

When NOT to bring the dice:

• First visit to the table after seating (too soon; cold)
• During a complaint or service-recovery moment (wrong tone)
• When the guest looks rushed, distracted, or in a tense conversation
• When the table is in the middle of a phone call or work-meeting register
• During a Circle-member's quiet-recognition moment (let the Circle thing be the warmth; don't compound)
• Last-table-of-service when staff are visibly closing down (the warmth becomes performance)

3.2 Location — where the roll happens

At the guest's table, on a small wooden tray or directly on the table cloth if no tray. The FOH does NOT roll for the guest. The guest rolls. Handing the dice over is the moment.

3.3 Who initiates

The FOH initiates. The guest does not ask for the dice; the guest is invited to roll. The framing is "een dobbelsteen voor de tafel — uw beurt" or equivalent — never "would you like to play" (which surfaces the mechanic) or "can I offer you the chance to win" (which surfaces the reward).

Cast involvement: any FOH member can initiate. Per §5.0.1 rule 4 (cast-voice cadence ≥30%), the dice is a natural way to bring sous-chefs / pastry assistant / bar lead to the table when they have a moment — they hand the dice over, the warmth gets distributed. Anna decides who can initiate when, but it's not exclusively FOH lead.

4. Roll outcomes + reward structure

Owner decision required: the specific reward per face is Michelle's call. The framework below is a strawman. See §10 for owner-input items.

Strawman mapping (working defaults — Michelle approves or swaps):

Rationale for this mapping:

• No discount-shaped outcomes. Per the gamification-as-pillar memory: variable-reward mechanics ≠ discounting. Nothing on the table reduces the bill.
• Variable cost-to-house, mostly low. Average expected cost per roll ≤€3-4. A table that rolls weekly costs ~€15-20/month — within standard hospitality goodwill budget.
• Face 5 (name on the wall) is intentionally high-warmth-zero-cost. The cheapest outcome carries the most memory. Don Julio art-of-hospitality pattern.
• Face 6 (no outcome) is essential. Without a "nothing happens" face the dice becomes a guaranteed-reward mechanic, which kills the variable-reward dynamic. The "de volgende komt wel" framing turns the no-outcome into a warm continuation, not a loss.
• Face 2 (next-visit chef's-pick) creates a return-visit hook without naming it. Felt-not-told extended: the guest leaves with something to come back for, but it isn't framed as a return-visit incentive.

Frequency: one roll per table per visit. Not per guest. Not multiple rolls if the table has multiple Circle members. The dice is a moment, not a system to optimize.

5. Service language — the felt-not-told register

Cross-reference: the cast 1:1 question set §4 Q15 tests cast members' framing of dice outcomes. The behavioural diagnostic there mirrors this section's anti-patterns.

5.1 What FOH says — the working register

Initiating the roll (any of these are correct):

• "Een dobbelsteen voor de tafel — uw beurt."
• "Hier — gooien jullie 'm een keer."
• "De dobbelsteen heeft u nog niet gehad vandaag." (only if it's true that this guest hasn't rolled before; never use as a faux-personalisation)

Win-framings (per face):

• Per the §4 table — short, specific, no upgrade language

No-outcome (face 6):

• "De volgende komt wel."
• "Dat was 'm nog niet — maar u was er."
• "Soms doet ie niks."

Cast member handing dice over (not FOH lead):

• "De [chef-rol of sous-rol] wilde even langskomen — bij deze ook een dobbelsteen."
• Cast member doesn't have to explain themselves; the dice is the bridge

5.2 What FOH does NOT say — banned framings

These are tested against 0038_banned_phrases_2026_felt_not_told.sql + the gamification-felt-not-told pillar memory:

5.3 Outbound copy rule

The dice does NOT appear in: michiu.nl / IG / TikTok / LinkedIn / press / GBP description / businessrestaurants.nl listing / any marketing surface. Penn does not pitch it. Zuck does not post it. Siren does not reference it in review replies.

The dice MAY appear in: word-of-mouth (uncontrolled; that's fine — the §1.1 specificity rule lets WoM carry it), guest-side photography (some guests will photograph the dice; we don't ask them to delete it but we don't repost), the small wooden box of retired dice behind the FOH station (visible to staff + observant guests only).

6. Circle integration — Circle-felt-not-told × dice-felt-not-told

The two felt-not-told pillars compound at this moment. The dice is identical for every guest. Circle status does not visibly change the experience.

What Circle membership DOES change (invisibly):

• Circle members already get the small touches the dice rolls produce — a comp coffee on a regular visit, an off-menu canapé. Circle members rolling the dice and getting face 1 (espresso on the house) is functionally redundant; the FOH plays it as a natural moment.
• The "name on the wall" face 5 outcome carries more warmth for a Circle regular because the recognition compounds — same outcome, different felt-weight.

What Circle membership does NOT change:

• The dice doesn't appear MORE OFTEN for Circle members
• The dice outcomes are not weighted higher for Circle members
• The FOH does not say "voor onze regelmatige gasten gooien we eens" — that surfaces the tier

Edge case — Circle members notice they're not getting tier-favored treatment at the dice moment: This is correct behavior. Per the Circle pillar, the Circle is felt, never announced. A Circle member who notices the dice is identical is experiencing the pillar as designed. If they ASK whether Circle changes the dice, the FOH says: "De dobbelsteen is voor iedereen. Maar u, u kennen we." — and the small touches that ARE Circle-attributable happen elsewhere in the meal, organically.

7. Edge cases + refusal handling

7.1 Guest declines the dice

Some guests will say "nee, dank u" when offered the dice. Honor it. The FOH does not insist, does not explain, does not offer again later. "Goed — geen probleem" + move on. The dice declined is itself a data point (some guests prefer no-mechanic moments; that's a Status Builder tier signal — note in Anna's internal log).

7.2 Guest wants to roll multiple times

Variants:

• "Mag ik nog een keer?" (single guest asking for second roll) — "Eén per tafel per bezoek — volgende keer." Warm, specific, closed.
• "Mogen we allebei gooien?" (multi-guest table, each guest asking) — "Een dobbelsteen voor de tafel — kies maar wie hem gooit." Same one-per-table rule.
• "Mag mijn kind 'm gooien?" (parent asks on behalf of a child at the table) — yes; the kid rolls; the outcome applies to the table as normal. This is the right answer; do not deflect.

7.3 The dice rolls off the table / breaks / lands wrong

• Rolls onto the floor: FOH retrieves, "sorry — opnieuw", hand back. No re-roll counts against the one-per-visit rule (this was a mechanic failure, not a roll).
• Lands on edge / ambiguous reading: FOH picks the lower number ("conservatief"). Or jokes: "deze ben ik niet zeker van — wij houden 'm op [lower]". The dice has authority; the FOH doesn't fudge upward.
• Dice is missing from station: FOH does NOT offer the dice that shift. Don't substitute. Anna sources a replacement before next service.

7.4 Group bookings (8+ covers)

For group / private-room bookings, the dice mechanic doesn't fit naturally — too large a table for the warmth-pacing in §3.1. Skip the dice for groups ≥8 covers. Group-booking warmth comes from other places (Michelle visit-to-table moment, chef's-gift dish, etc.). Don't force the dice into a setting it doesn't fit.

7.5 Lunch-vs-dinner

The pilot is lunch only. Dinner has different pacing + different audience tier mix; the felt-not-told register at dinner needs separate testing. Do not extend to dinner without Sun + Anna + Michelle's explicit Phase-2 decision after the lunch pilot data lands.

7.6 Press visitor at a lunch table

Per §5.3 outbound-copy rule: press visitors get the dice if they happen to be at a lunch table where the §3.1 trigger conditions hold. The FOH treats them as any guest. Don't telegraph the dice to press as a Michiu feature. If a critic writes about the dice in a column, that's organic discovery — fine; we don't seed it.

8. Capture loop — what gets recorded

Each roll produces a small structured record. Anna captures in a service-end log; Insight aggregates weekly.

Per-roll record:

• Date + lunch-service slot
• Table number + cover count
• Face rolled (1-6)
• Outcome honored (yes/no/upgraded — face 3 upgrade to face 4 if no dessert ordered)
• Circle member at table? (yes/no — invisible to guest; FOH knows)
• Cast member who handed the dice over (FOH lead / sous / pastry / bar / etc.)
• Anna's qualitative note if anything stood out (1 line max — refusals, group dynamic, follow-up signals)

Aggregation Insight runs weekly:

• Total rolls / week
• Face distribution vs uniform (any drift signals dice imbalance or scoring fudge)
• Cast-member-who-handed-it distribution (per §5.0.1 rule 4 ≥30% cast-voice tracking)
• Circle-member dice frequency vs non-Circle (should be proportional to Circle-vs-non-Circle lunch-cover ratio; if disproportionate, FOH is unconsciously favoring)
• Per-face-honored-vs-declined-by-guest count (face 3 dessert on a no-dessert table → upgrade; face 5 name on wall → some guests may decline)

Where the data lives:

• Per-roll records: a simple dice_rolls table or JSONB on the lunch_dice_2026 Campaign object (post-0053_campaigns.sql apply) — Sun decides during Campaign archetype creation
• Insight aggregation: weekly digest entry in Michelle's Monday-morning dashboard widget per §10.4

Privacy: records are not guest-attributable. Table number is operational; cover count is operational; Circle status is internal-only. No guest names, no contact info, no PII per the §6 record schema.

9. Pilot plan + measurement

9.1 Pilot scope

Phase 1 — 4-week lunch pilot:

• 1 dice in service, lunch only
• FOH briefed per §5 service language
• Anna's per-roll capture log running daily
• Insight weekly aggregation

Phase 1 success criteria (no specific number — directional):

• ≥80% of dice offers are accepted (guests rolling, not declining)
• Face distribution within ±15% of uniform (no scoring fudge / dice imbalance)
• Zero outbound-copy violations of the §5.3 rule (Penn / Zuck / Siren do not reference the dice)
• Anna's qualitative log surfaces ≥3 named guest moments per week that wouldn't have happened without the dice (the warmth-creation working as designed)
• No service-recovery incidents attributable to the dice (no guest complaints about the framing, no dice-as-pressure moments)

Phase 1 NOT measured:

• Direct bill-uplift or revenue impact (variable-reward mechanics ≠ discounting; ROI framing breaks the pillar)
• Return-visit attribution from face 2 (next-visit chef's-pick) — measurable in theory but the §4.3 attribution-loop work is downstream
• NPS or guest-satisfaction surveys — not the right instrument for a felt-not-told mechanic

9.2 Phase 2 decision (post-pilot)

After 4 weeks, Sun + Anna + Michelle review the Insight digest + Anna's qualitative notes. Three Phase-2 options:

• Continue lunch-only — the pilot worked as designed; the dice stays in lunch service as a standing feature
• Extend to dinner — pilot moves to dinner pacing test; new §7.5 lunch-vs-dinner rule revisited
• Retire — the mechanic isn't fitting; the dice goes into the retired-dice wooden box behind the FOH station and the pillar surfaces through a different mechanic (wheel / scratch-card / something Sun designs next)

The "retire" option is real, not theoretical. The pillar matters; the specific mechanic is replaceable.

10. Owner-input punch-list (gating items)

Sun cannot ship the lunch-dice-2026 Campaign archetype without these resolved:

1. Michelle — reward design. §4 strawman mapping (faces 1-6) is Casper's working default; Michelle approves the per-face outcomes OR swaps. Specific opens: face 1 espresso/thee — yes? face 2 next-visit chef's-pick — yes, and what's the value cap? face 3 dessert-on-house — yes, even at peak lunch? face 5 name-on-the-wall — yes, and where's the chalkboard going? face 6 no-outcome — yes (the most important to keep). Effort: 30-min conversation.
2. Michelle — dice physical form. Wood-default OK, or commissioned with Yue-do-related engraving? Effort: 1 decision; then 2-3 weeks lead time if commissioned.
3. Anna — FOH service-language voice-pass. §5.1 NL register working drafts approved or refined? Anna's cast 1:1s (per cast-1on1-question-set.md §5.5 FOH-lead questions) should surface the FOH lead's own framing preferences. Effort: incorporated into the cast 1:1s.
4. Anna — pilot-launch start-date + dice-source. When does Phase 1 4-week pilot begin? Where does the dice get sourced (Amsterdam woodworker / online / brand-assets/?)? Effort: 1 conversation + 1-2 weeks sourcing lead time.
5. Sun — Campaign archetype creation in marketing-os. Once 0053_campaigns.sql is applied live (operator §2.3) + Michelle's reward design + Anna's pilot start-date are locked, Sun creates lunch-dice-2026 archetype with this doc as the brief. Effort: 2 hrs.
6. Insight — capture-table schema decision. Separate dice_rolls table OR JSONB on the Campaign object? Sun + Insight decide during Campaign creation. Effort: 1 decision in the Campaign-creation session.
7. Julius — banned-vocabulary check on §5.1 language. Once Anna's voice-passes land, Julius runs all guest-facing service-language against 0038_banned_phrases_2026_felt_not_told.sql. Effort: 15 min.

11. Anti-patterns (what kills this pillar)

Beyond the §5.2 service-language banned framings and the §5.3 outbound-copy rule, these are the structural ways the pillar fails:

1. Naming the mechanic in marketing. A single IG post about "our lucky dice" kills the felt-not-told pillar across all three pillars (gamification + Circle + positioning). Penn / Zuck / Siren never reference.
2. Making the dice a guarantee-of-reward. If every face produces a meaningful outcome, the variable-reward dynamic dies. Face 6 (no-outcome) is non-negotiable.
3. Making the dice scheduled. "Dice Tuesdays" or "Lucky Lunch Special" turns the mechanic into a marketing event — surfaces it as a system.
4. Cast members performing the dice. If the FOH framing reads as a memorized routine — same words every time, same gestures — guests notice. The §5.1 register has multiple working variants because organic variation IS the warmth.
5. Optimizing the dice for revenue. The moment Insight starts measuring bill-uplift attribution from the dice, the protocol is breaking. The pillar is not a revenue mechanic. (Insight measures the things in §8 — adoption, distribution, qualitative warmth signals. Not bill impact.)
6. Circle-tier visible favoring. If Circle members notice their dice rolls are weighted higher, OR if FOH treats Circle dice moments differently in front of non-Circle guests, both felt-not-told pillars (gamification AND Circle) break simultaneously.
7. Cross-mechanic stacking. Don't combine the dice with another variable-reward mechanic in the same visit. If a wheel comes later, it's a separate Campaign with its own table-occurrence rule; never both at the same meal.
8. Forced rolls during press visits. If a critic happens to be at a lunch table where §3.1 trigger conditions hold, they get the dice. If they're NOT at such a table, don't force the moment for press benefit. Press discovery should be organic or not at all.

12. Decision log

• 2026-05-12 — Drafted by Casper (agency lead) as Sun + Anna's operational framework for the gamification pillar's first instance per the brand-pillar memory. Structured as design framework + working SOP draft + owner-input punch-list — same hybrid pattern as the Grimm pitch (working defaults Penn can ship if Michelle approves; otherwise replaced) and the cast 1:1 question set (behavioural framework Sun runs with).
• 2026-05-12 — Chose physical wooden dice over digital/iPad-mediated mechanic. Reason: tactile + warmth + replaceable + cheap; digital surfaces undermine the felt-not-told pillar by introducing a screen that's legible as a mechanic.
• 2026-05-12 — §4 strawman reward mapping is opinionated but explicitly Michelle's-decision. Working defaults provided so Sun can move (Campaign archetype can be drafted with placeholders) without Michelle approval; production roll-out gated on her approval per §10 item 1.
• 2026-05-12 — Face 6 (no-outcome) framed as non-negotiable. Without it the variable-reward dynamic collapses. The "de volgende komt wel" framing turns no-outcome into warm-continuation rather than loss.
• 2026-05-12 — Lunch-only pilot scope chosen vs lunch+dinner. Reason: dinner pacing + audience tier mix differ; piloting lunch isolates the variable-reward effect cleanly. §7.5 explicit rule.
• 2026-05-12 — Outbound-copy rule (§5.3) names the surfaces the dice does NOT appear on. Reason: the gamification-felt-not-told brand pillar memory rules out system framing in outbound — naming the surfaces explicitly closes the rule against drift in Penn / Zuck / Siren copy.
• 2026-05-12 — Insight aggregation explicitly NOT measuring bill-uplift / ROI. Reason: that framing breaks the pillar; variable-reward mechanics ≠ discounting per owner; measuring revenue impact treats them as discounting equivalents.
• 2026-05-12 — Phase-2 retirement option (§9.2) named as real, not theoretical. The pillar matters; the dice doesn't. Sun + Anna + Michelle can swap to a different variable-reward mechanic without breaking the pillar.
• Pending — Michelle reward-design approval (§10 item 1).
• Pending — Anna FOH voice-pass on §5.1 (§10 item 3) — incorporated into cast 1:1s.
• Pending — Sun Campaign archetype creation (§10 item 5) — gated on 0053_campaigns.sql operator-authorization.
• Pending — Phase 1 pilot launch (4-week lunch pilot) per Anna's start-date decision.
• Pending — Phase 2 review post-pilot (Sun + Anna + Michelle).

13. Related references

• Strategy doc: docs/strategy/2026-baseline.md §2 Tactic 1.1 (lunch-daypart context); §5.0.1 rule 4 (cast-voice cadence — dice rolls are cast-voice moments)
• Canonical Yue-do reference: docs/yue-do.md §5 (felt-not-told disciplines)
• Cast 1:1 question set: docs/cast-1on1-question-set.md §4 Q15 (cast-side behavioural diagnostic on dice framing); §5.5 FOH-lead role-specific add-ons
• Operator punch-list: docs/q1-2026-operator-checklist.md §5.4 (Sun's first Yue-do-anchored Campaign brief — this is the operational meat the Campaign wraps)
• Banned phrases: 0038_banned_phrases_2026_felt_not_told.sql (Julius runs against §5 service language)
• Campaign object: 0053_campaigns.sql (the schema Sun's Campaign archetype lives in — pre-apply gated per operator §2.3)
• Brand pillar memory: gamification is a core Michiu brand pillar — variable-reward mechanics ≠ discounting per owner; gestures named, system banned outbound (third felt-not-told pillar)