HSpector1 · HSpector1 · Jul 30, 2026 · Jul 30, 2026 · Jul 30, 2026 · Jul 30, 2026
diff --git a/docs/DECISION-LOG.md b/docs/DECISION-LOG.md
@@ -5,6 +5,33 @@ consequences, owner, and conditions to revisit. Newest first.
 
 ---
 
+### 2026-07-31 · D-031 · M0 human-gate readiness: attraction legibility + course-scaled seat duration
+- **Status:** provisional · **Owner:** Builder → owners · Branch `fix/m0-human-gate-readiness` off PR #7 head
+  `e5baa54`; focused PR targets the PR #7 branch. Addresses the PR #7 adjudication's H2 (legibility) + M1 (economics).
+- **H-LEG-1 (attraction legibility):** per-segment `SegmentAttraction` (pool/consideration/composition/parties) is
+  now on `ForecastSnapshot` and `ServiceResult` and rendered pre-service ("Who this plan attracts" table + a
+  plain-language note, largest-remainder reconciled) and post-service (ATTRACTION funnel). New causal categories
+  **weak-attraction / wrong-mix / purchase-conversion** (renamed from weak-demand/menu-fit) with disjoint-PState
+  attribution; wrong-mix is a composition opportunity cost (never double-counted) with a business-consequence line.
+- **MED-ECON-1 (course-scaled seat duration):** flat `EatMin=12` replaced by shared `Tuning.SeatEatMin(coursesX10)`
+  = clamp(6 + 6·courses, 12, 30) → 1/2/3-course = 12/18/24; sim scales by `DistinctCourseCount`, forecast scales
+  `seatCoverCap` by the captured-mix expected courses. Empirically a premium 3-course meal now turns ~1.85× fewer
+  tables/seat-hour than a value 1-course meal; per-seat-minute leverage falls ~4× → ~2.5-3×. Only eat time changes
+  (ticket/kitchen path untouched).
+- **Consequence / gate:** `Verdict: Pass-with-notes / Action: Continue` (request focused independent re-review).
+  146 tests green (added 7 ReadinessTests). No over-correction: premium still wins enthusiast, value wins lunch,
+  KNOWN_RESIDUAL holds. Favorable side effect: best-generalist regret rose 6%→17% on base 271828182 (reducing
+  premium seat-leverage shrank the generalist's edge — as the adjudication predicted; not gate-chasing). Goldens
+  re-baselined (D-032). Report: `reports/m0/M0-HUMAN-GATE-READINESS-REPORT.md`. Do not merge / no human tests / no M1.
+
+### 2026-07-31 · D-032 · Goldens re-baselined for course-scaled seat duration
+- **Status:** locked · **Owner:** Builder · Course-scaled dining duration changes seat occupancy → turns/covers/
+  revenue → checksum. New: Focused Value/lunch `0x53FE6CB73895DFD0`, Premium Craft/enthusiast `0x2DA79E2ACC88CA2F`,
+  Balanced Competent/social `0x05E65C6E8AD6AF07`. Prior (D-030): `0xFB04479B19656E6B` / `0x198E44A139793975` /
+  `0x6CBD8F76B46A51E1`. Attraction legibility (H-LEG-1) does NOT affect the checksum (attraction fields and cause
+  strings are not hashed); the weak-demand→weak-attraction / menu-fit→purchase-conversion renames are cause
+  strings only.
+
 ### 2026-07-30 · D-029 · Menu-responsive demand rewrite: distinct regimes achieved; universal-generalist gate borderline (economic root cause)
 - **Status:** provisional · **Owner:** Builder → owners · **Follows** the PR #5 adjudication (D-028's Fail/Rewrite,
   refined: the fix is a headless demand-model rewrite, not M1). Branch `rewrite/m0-menu-responsive-demand`.

diff --git a/docs/design/M0-HUMAN-GATE-READINESS-CONTRACT.md b/docs/design/M0-HUMAN-GATE-READINESS-CONTRACT.md
@@ -0,0 +1,113 @@
+# M0 Human-Gate Readiness Contract (design lock)
+
+**Branch** `fix/m0-human-gate-readiness` off PR #7 head `e5baa54`. Focused PR targets the PR #7 branch
+(`rewrite/m0-menu-responsive-demand`). Two authorized corrections only; do NOT reopen `DemandModel.Capture`.
+This is committed BEFORE code (§6). Design informed by the read-only A–E investigation (recorded in the PR).
+
+## 1. Authorized findings (from the independent adjudication)
+- **H-LEG-1 — Attraction legibility.** Per-segment consideration/captured demand are computed but never shown;
+  the forecast collapses attraction to one aggregate %, and the autopsy has no "wrong-mix" category.
+- **MED-ECON-1 — Course-scaled seat duration.** Flat `EatMin=12` gives premium multi-course covers ~4×
+  contribution-per-seat-minute (vs ~1.15× per station-minute) because a 3-course dinner holds a seat the same
+  eating window as a 1-course lunch.
+
+Non-goals (§3 of the assignment) unchanged: no reputation/repeat-visits/marketing/reservations/new
+recipes-segments-markets/graphics/etc. No premium-specific penalty. No market/segment NAME branch. Integer/
+deterministic only.
+
+## 2. Correction A — attraction legibility
+
+### 2a. Forecast (pre-service)
+Add to `ForecastSnapshot` (Result.cs) `IReadOnlyList<SegmentAttraction> SegmentAttraction`, record
+`SegmentAttraction(SegmentId Id, string Name, int PoolParties, int ConsiderationBp, int CompositionBp, int AttractedParties)`,
+and `IReadOnlyList<string> AttractionNotes`. Populated in `Forecaster.Compute` from the existing `captured` local
+(no new `Capture` call): `PoolParties = round(ExpectedArrivals·SegmentMixBp[s]/1e4)`, `ConsiderationBp =
+captured.ConsiderationBp[s]`, `CompositionBp = captured.CompositionBp[s]`, `AttractedParties =
+round(attempts·CompositionBp[s]/1e4)`. **Reconciliation (largest-remainder):** the last segment's PoolParties =
+ExpectedArrivals − Σ(others) and AttractedParties = attempts − Σ(others), so the columns sum EXACTLY to
+ExpectedArrivals and to attempts (the aggregate funnel figure the autopsy prints as DemandGenerated).
+
+`TextReport.Forecast` renders a compact "Who this plan attracts (pre-service)" table (segment | pool | appeal% |
+mix% | attracted, + total row) using the existing `Pct()` helper and column idiom, placed before Key Assumptions.
+At most ONE plain-language attraction note fired (priority order), derived from actual data — never generic.
+
+### 2b. Post-service (autopsy)
+Thread the whole `CapturedDemand` into `Finalize` (Simulator currently keeps only `AggregateConversionBp` +
+`CompositionBp`; also keep `ConsiderationBp`). Add per-segment attraction to `ServiceResult`: extend
+`SegmentOutcome` (or add a parallel `SegmentAttractionActual`) with `PoolParties`, `ConsiderationBp`,
+`ArrivedParties` so the autopsy shows **potential → consideration → captured(expected) → arrived → purchased →
+covers** per segment and the player can compare forecast captured mix vs actual arrival mix.
+
+### 2c. Causal categories (§11)
+Extend `DiagnoseBottleneck` vocabulary from {seating, kitchen, menu-fit, weak-demand} to
+{seating, kitchen, **purchase-conversion** (was menu-fit), **weak-attraction** (was weak-demand), **wrong-mix**}.
+Attribution reads DISJOINT slices (no double-count): weak-attraction = parties never generated (volume shortfall
+below the attraction floor, `Tuning.WeakDemandCaptureFloorBp`); seating = `WalkedSeat`; purchase-conversion =
+`NoOrder`; kitchen = `WalkedFood` + comp cents; wrong-mix = a COMPOSITION opportunity cost (below). Every party is
+in exactly one terminal PState; wrong-mix is an opportunity signal, not a party bucket, so it cannot double-count.
+
+**wrong-mix** is PRIMARY only when ALL hold: (1) volume adequate (captured ≥ attraction floor — not weak);
+(2) seated covers largely ordered (`LostToMenuFit` small — not purchase); (3) kitchen kept pace (`LostToWait`
+small, comps modest — not kitchen); (4) the market offered a materially higher-value segment (by pool × budget)
+that the plan under-attracted (its `CompositionBp` well below its pool share) while a lower-value segment
+dominates arrivals. Business-consequence sentence required (e.g. "you filled the room with value diners in a
+market whose enthusiasts would have paid far more"). This uses pool + composition + segment budgets only — no
+per-segment contribution accounting, no name branch. Two-tier report preserved: TIER-1 the single largest
+measured channel (incl. the new attraction ones); TIER-2 0–2 secondary operational warnings.
+
+## 3. Correction B — course-scaled dining duration
+**Model A (base + uniform per DISTINCT course).** Shared primitive in `Tuning`:
+`SeatEatMin(int coursesX10) = clamp(EatBaseMin + MulDivRound(EatPerCourseMin, coursesX10, 10), EatMinFloor, EatMaxCeil)`.
+Params: **EatBaseMin=6, EatPerCourseMin=6, EatMinFloor=12, EatMaxCeil=30** → 1-course=12 (preserves legacy dwell,
+byte-identity for all-1-course services), 2-course=18, 3-course=24 (≈2× the legacy 12, the D4 target). No floats,
+no rounding beyond MulDivRound.
+
+- **Sim** (`Simulator.cs`, Cooking→Eating transition, replacing `p.EatUntil = t + EatMin`): `courses =
+  DistinctCourseCount(p)` = count of distinct `Recipe.Course` among `p.Dishes` (0..3, deterministic, size- and
+  RNG-independent); `p.EatUntil = t + Tuning.SeatEatMin(courses·10)`. Only the EAT window changes — browse, cook,
+  ticket, delivery are untouched (`AvgTicketTimeMin` must NOT move).
+- **Forecast** (`Forecast.cs`): replace flat `ForecastAvgDwellMin=34` in `seatCoverCap` with
+  `ForecastNonEatDwellMin + SeatEatMin(expectedCoursesX10)`, where `expectedCoursesX10 = dishesPerCoverBp·10/1e4`
+  (the existing captured-mix-weighted 1 + P(starter) + P(dessert)). Pick `ForecastNonEatDwellMin` so the
+  composition-weighted dwell reproduces ~34 at the current markets (calibration preserved); premium/high-course
+  menus get a longer dwell → fewer seat turns → lower `seatCoverCap` (directional).
+
+Timing separation preserved (§20): kitchen prep / food wait / eating are distinct causes; only eating is scaled.
+
+## 4. Numeric units & rounding (§18)
+All timing integer minutes. `SeatEatMin` bounded to [12, 30]. `coursesX10` integer (sim: 10/20/30; forecast:
+10 + fractional). `MulDivRound` for the per-course scaling and all bp math. No floating-point authoritative time.
+
+## 5. Authorized files
+`Tuning.cs`, `Simulator.cs`, `Forecast.cs`, `Result.cs`, `TextReport.cs`, `M0Content.cs` (only if a Name lookup is
+needed — no fixture-economics change), harness `Program.cs` (diagnostics/regression only), and the test files +
+new fixtures + reports/docs. **Not** `Demand.cs`/`PriceModel.cs` (no consideration redesign). No new recipes/
+segments/markets; no fixture-economics change to prices/costs/appeal.
+
+## 6. Non-goals
+Repeat visits, reputation, reviews, loyalty, marketing, reservations, table-turn controls, bussing, course-by-
+course service/inter-course stagger (delivery-together stays), per-dish eat animations, new content, graphics,
+save, M1. No premium-specific tax; no market/segment-name branch; do not tune to move the generalist past 10%.
+
+## 7. Pass criteria (§28,29,31)
+- **Attraction legibility:** per-segment pool/consideration/captured/arrival-mix visible pre- AND post-service;
+  ≥1 plain-language attraction reason; totals reconcile; a wrong-mix/weak-attraction/purchase case each attributes
+  to the correct stage (controlled fixtures).
+- **Directionality:** ≥85% overall / ≥75% per-market & per-family forecast/sim agreement retained; audited price &
+  seat directions still correct; adding expected courses lowers forecast table turns; more seats help when
+  seat-bound, neutral/harmful when kitchen-bound.
+- **Economics:** premium contribution-per-seat-minute leverage reduced from ~4× toward ~3×; premium still wins
+  enthusiast; value still wins lunch; a mixed plan still viable in social; multi-course still net-positive.
+- **Technical:** 139 prior tests preserved (+ new); determinism; accounting; RNG isolation; forecast immutability;
+  scope; goldens deliberately re-baselined & documented.
+
+## 8. Fail criteria (§24,27 over-correction)
+Premium stops winning enthusiast / goes loss-making; desserts+starters become universally unprofitable;
+KNOWN_RESIDUAL flips (generalist destroyed) as a side effect; `AvgTicketTimeMin` moves (change leaked into the
+kitchen path); or the duration change is economically inert everywhere (surface honestly, don't force it).
+
+## 9. Timebox & review (§32,35)
+12–22 focused hours, ≤2 refinement cycles. If it cannot complete honestly in the box: stop, preserve evidence,
+report, request an owner decision — do not auto-start another correction. Deliver the §35 report + open the
+focused PR (do not merge; no human tests; no M1). Gate recommendation uses Verdict/Action; `Continue` = request a
+focused independent re-review (does not authorize human tests, M1, or merge).
diff --git a/reports/balance/distribution.md b/reports/balance/distribution.md
@@ -10,15 +10,15 @@ _Value-heavy, high-volume, price-sensitive, small parties, early peak. Rewards f
 
 | Strategy | median contrib | p10 | p90 | loss% | avg covers | avg sat | avg ticket | win% |
 |---|--:|--:|--:|--:|--:|--:|--:|--:|
-| Focused Value | $1682.50 | $1480.00 | $1892.00 | 0% | 170 | 604 | 11m | 93% |
-| Premium Craft | $1064.12 | $635.46 | $1549.56 | 1% | 42 | 719 | 17m | 6% |
-| Broad Menu | $393.00 | -$121.50 | $933.50 | 15% | 79 | 580 | 19m | 0% |
-| Overcapacity | -$572.00 | -$756.00 | -$424.50 | 100% | 23 | 538 | 19m | 0% |
+| Focused Value | $1667.50 | $1441.00 | $1826.00 | 0% | 167 | 605 | 11m | 91% |
+| Premium Craft | $1071.02 | $635.46 | $1564.16 | 1% | 42 | 717 | 17m | 8% |
+| Broad Menu | $550.50 | $19.00 | $1078.00 | 7% | 83 | 576 | 19m | 0% |
+| Overcapacity | -$575.00 | -$756.00 | -$424.50 | 100% | 23 | 533 | 19m | 0% |
 | Understaffed | -$402.00 | -$529.00 | -$261.50 | 100% | 14 | 512 | 21m | 0% |
-| Balanced Competent | -$20.50 | -$429.50 | $388.00 | 53% | 64 | 576 | 19m | 0% |
-| Overpriced Weak Execution | -$245.90 | -$522.50 | $219.20 | 76% | 10 | 404 | 26m | 0% |
-| Station Bottleneck | -$316.00 | -$473.50 | -$50.50 | 93% | 19 | 603 | 19m | 0% |
-| Intentionally Bad | -$118.00 | -$237.50 | $32.00 | 80% | 9 | 548 | 21m | 0% |
+| Balanced Competent | $134.00 | -$376.50 | $632.50 | 38% | 69 | 567 | 19m | 0% |
+| Overpriced Weak Execution | -$245.90 | -$539.20 | $219.20 | 76% | 10 | 402 | 26m | 0% |
+| Station Bottleneck | -$316.00 | -$473.50 | -$44.50 | 92% | 19 | 599 | 19m | 0% |
+| Intentionally Bad | -$118.00 | -$237.50 | $32.00 | 80% | 9 | 547 | 21m | 0% |
 
 **Winner:** Focused Value  ·  **profitable strategies (median >= +$150):** 3
 
@@ -28,40 +28,40 @@ _Mid-budget social diners in groups, multi-course, moderate patience. Rewards a
 
 | Strategy | median contrib | p10 | p90 | loss% | avg covers | avg sat | avg ticket | win% |
 |---|--:|--:|--:|--:|--:|--:|--:|--:|
-| Focused Value | $1816.50 | $1461.00 | $2195.00 | 0% | 153 | 722 | 10m | 60% |
-| Premium Craft | $1412.94 | $274.46 | $2633.06 | 6% | 70 | 590 | 24m | 36% |
-| Broad Menu | $241.50 | -$449.00 | $1153.00 | 30% | 89 | 529 | 25m | 2% |
-| Overcapacity | -$578.50 | -$882.50 | -$211.00 | 98% | 33 | 573 | 21m | 0% |
+| Focused Value | $1805.00 | $1461.50 | $2174.50 | 0% | 152 | 711 | 10m | 7% |
+| Premium Craft | $2623.06 | $769.60 | $3788.42 | 0% | 87 | 579 | 24m | 71% |
+| Broad Menu | $1954.00 | $229.00 | $2667.50 | 7% | 119 | 522 | 24m | 16% |
+| Overcapacity | -$578.50 | -$882.50 | -$211.00 | 98% | 33 | 556 | 21m | 0% |
 | Understaffed | -$421.00 | -$553.00 | -$199.00 | 100% | 17 | 517 | 25m | 0% |
-| Balanced Competent | $135.00 | -$419.50 | $943.00 | 36% | 75 | 597 | 22m | 1% |
-| Overpriced Weak Execution | -$719.40 | -$986.50 | -$438.50 | 98% | 9 | 402 | 28m | 0% |
-| Station Bottleneck | -$521.00 | -$740.00 | -$209.50 | 100% | 20 | 617 | 21m | 0% |
-| Intentionally Bad | -$237.50 | -$381.00 | $10.50 | 87% | 11 | 519 | 24m | 0% |
+| Balanced Competent | $1226.00 | $59.50 | $2122.00 | 6% | 102 | 567 | 22m | 5% |
+| Overpriced Weak Execution | -$687.10 | -$992.50 | -$355.90 | 98% | 9 | 397 | 27m | 0% |
+| Station Bottleneck | -$520.50 | -$740.00 | -$209.50 | 100% | 20 | 610 | 21m | 0% |
+| Intentionally Bad | -$237.50 | -$381.00 | $10.50 | 87% | 11 | 517 | 24m | 0% |
 
-**Winner:** Focused Value  ·  **profitable strategies (median >= +$150):** 3
+**Winner:** Premium Craft  ·  **profitable strategies (median >= +$150):** 4
 
 ## Destination Enthusiast Evening  (`enthusiast-evening`)
 
 _High-budget enthusiasts, lower volume, high quality expectations, patient. Rewards premium craft and skilled execution._
 
 | Strategy | median contrib | p10 | p90 | loss% | avg covers | avg sat | avg ticket | win% |
 |---|--:|--:|--:|--:|--:|--:|--:|--:|
-| Focused Value | $394.50 | $95.50 | $691.50 | 2% | 71 | 716 | 10m | 0% |
-| Premium Craft | $2313.06 | $845.32 | $3445.88 | 0% | 76 | 583 | 25m | 76% |
-| Broad Menu | $1183.00 | -$21.00 | $2072.00 | 10% | 91 | 549 | 25m | 6% |
-| Overcapacity | $348.00 | -$106.00 | $998.50 | 13% | 50 | 575 | 21m | 0% |
+| Focused Value | $394.50 | $95.50 | $691.50 | 2% | 71 | 715 | 10m | 0% |
+| Premium Craft | $3051.48 | $1879.60 | $3839.04 | 0% | 86 | 586 | 24m | 95% |
+| Broad Menu | $1860.50 | $997.00 | $2485.50 | 0% | 104 | 543 | 24m | 5% |
+| Overcapacity | $348.00 | -$120.00 | $998.50 | 13% | 50 | 562 | 22m | 0% |
 | Understaffed | -$188.50 | -$405.50 | $60.00 | 86% | 24 | 480 | 27m | 0% |
-| Balanced Competent | $1359.00 | $866.50 | $1695.00 | 0% | 83 | 654 | 18m | 17% |
-| Overpriced Weak Execution | -$613.50 | -$903.80 | -$248.80 | 93% | 10 | 344 | 28m | 0% |
-| Station Bottleneck | -$153.00 | -$518.00 | $231.00 | 68% | 26 | 588 | 22m | 0% |
-| Intentionally Bad | -$166.00 | -$334.00 | $151.50 | 73% | 13 | 468 | 25m | 0% |
+| Balanced Competent | $1361.00 | $903.50 | $1759.50 | 0% | 84 | 646 | 18m | 0% |
+| Overpriced Weak Execution | -$632.70 | -$932.10 | -$199.50 | 93% | 10 | 340 | 28m | 0% |
+| Station Bottleneck | -$149.50 | -$518.00 | $227.50 | 67% | 26 | 582 | 22m | 0% |
+| Intentionally Bad | -$166.00 | -$334.00 | $151.50 | 73% | 13 | 466 | 25m | 0% |
 
 **Winner:** Premium Craft  ·  **profitable strategies (median >= +$150):** 5
 
 ## Dominance check
 
 - **lunch-rush** best strategy: `Focused Value`
-- **social-dinner** best strategy: `Focused Value`
+- **social-dinner** best strategy: `Premium Craft`
 - **enthusiast-evening** best strategy: `Premium Craft`
 
 Distinct winning strategies across scenarios: **2** of 3. No single strategy wins across all markets: context changes the best plan. ✅