kevinatlee · Aug 20, 2026 · Aug 20, 2026 · Aug 20, 2026
diff --git a/apps/client/src/styles.css b/apps/client/src/styles.css
@@ -193,11 +193,20 @@ button:disabled {
   position: relative;
   z-index: 1;
   margin: 0 auto 1rem;
-  max-width: 44rem;
   text-align: center;
   font-weight: 700;
   font-size: clamp(1rem, 2vw, 1.4rem);
 }
+.display-results > h1 {
+  max-width: 44rem;
+}
+.display-results__quip {
+  width: fit-content;
+  max-width: 100%;
+  min-width: 0;
+  overflow-wrap: anywhere;
+  white-space: normal;
+}
 .display-results__fireworks {
   position: absolute;
   inset: 0;

diff --git a/apps/client/src/styles.test.ts b/apps/client/src/styles.test.ts
@@ -173,6 +173,21 @@ describe('tile typography', () => {
     expect(styles).not.toMatch(
       /\.display-results\s*\{[^}]*justify-content: center;/s,
     );
+    expect(styles).toMatch(
+      /\.display-results > h1\s*\{[^}]*max-width: 44rem;/s,
+    );
+    expect(styles).toMatch(
+      /\.display-results__quip\s*\{[^}]*width: fit-content;[^}]*max-width: 100%;[^}]*min-width: 0;[^}]*overflow-wrap: anywhere;[^}]*white-space: normal;/s,
+    );
+    expect(styles).not.toMatch(
+      /\.display-results__quip\s*\{[^}]*width: max-content;/s,
+    );
+    expect(styles).not.toMatch(
+      /\.display-results__quip\s*\{[^}]*max-width: 44rem;/s,
+    );
+    expect(styles).not.toMatch(
+      /\.display-results__quip\s*\{[^}]*white-space: nowrap;/s,
+    );
     expect(styles).toMatch(
       /\.display-results__cards\s*\{[^}]*width: fit-content;[^}]*max-width: 96vw;[^}]*grid-template-columns: repeat\(2, minmax\(16rem, max-content\)\);/s,
     );

diff --git a/apps/server/src/server.ts b/apps/server/src/server.ts
@@ -5,7 +5,7 @@ import {
   generateDefaultBoard,
   loadProductionDictionary,
   PRODUCTION_DICTIONARY_IDENTITY,
-  selectMedianBoard,
+  selectStrongBoard,
   type ProductionDictionaryLoadResult,
 } from '@words/game-data';
 import {
@@ -243,7 +243,7 @@ export function createWordsServer(
           attempts: 0,
         };
       }
-      return selectMedianBoard({
+      return selectStrongBoard({
         size,
         random: randomSource,
         dictionary: gameDataRuntime.dictionary,

diff --git a/apps/server/tests/server.test.ts b/apps/server/tests/server.test.ts
@@ -1978,7 +1978,7 @@ describe('Words Stage 4B server', () => {
     });
   });
 
-  it('uses the median selector with eight production-valid candidates', async () => {
+  it('uses the strong selector with 32 production-valid candidates', async () => {
     await server.stop();
     const randomSource = { next: vi.fn(() => 0.5) };
     const boardGenerator = vi.fn(({ size, random }) => {
@@ -2013,7 +2013,7 @@ describe('Words Stage 4B server', () => {
       ok: true,
       room: { round: { generationAttempts: 1 } },
     });
-    expect(boardGenerator).toHaveBeenCalledTimes(8);
+    expect(boardGenerator).toHaveBeenCalledTimes(32);
   });
 
   it('broadcasts one count-only progress snapshot while keeping accepted words private', async () => {

diff --git a/docs/DEVELOPMENT_LOG.md b/docs/DEVELOPMENT_LOG.md
@@ -1,5 +1,20 @@
 # Development log
 
+## 2026-08-20 — Strong high-density board candidate (draft)
+
+- Added a separate server-only strong selector that chooses the highest
+  playable-word count from 32 production-valid candidates, breaking ties by
+  cell coverage and then generation order. The median selector remains intact.
+- Audited 500 deterministic selections per size twice with seed
+  `strong-board-selection-v1`; outputs were byte-identical and selected medians
+  exceeded the retained original corpus P95 for 4 × 4, 5 × 5, and 6 × 6.
+- Let the display result quip prefer its intrinsic width with `fit-content`
+  while shrinking safely on narrow portrait screens; normal wrapping and the
+  heading's 44rem bound remain intact.
+- Kept generation profiles, dictionary, `QU`, network contracts, lifecycle,
+  result content, and production publication unchanged. This branch is only a
+  physical gameplay and TV test candidate.
+
 ## 2026-08-14 — Post-round player board review (draft)
 
 - Player phones now retain the exact authoritative finished board as a frozen,

diff --git a/docs/board-quality-research/STRONG_SELECTION_EXPERIMENT.md b/docs/board-quality-research/STRONG_SELECTION_EXPERIMENT.md
@@ -0,0 +1,51 @@
+# Strong board selection experiment
+
+This physical-test candidate keeps the production generator, dictionary, board
+acceptance profiles, `QU` policy, and generation-attempt bounds unchanged. For
+each round it solves 32 accepted production-valid candidates, then selects by:
+
+1. highest playable-word count;
+2. highest cell coverage;
+3. earliest candidate generation index.
+
+Long-word count and vowel spread are measured but are not ranking criteria.
+Production's existing median selector and targets remain available unchanged.
+
+## Deterministic audit
+
+`npm run research:strong-selection` ran 500 selections per board size with seed
+`strong-board-selection-v1`, covering 48,000 requested candidates. A second run
+produced a byte-identical
+`generated/strong-selection-summary.json`. Runtime values are intentionally not
+stored in that deterministic artifact because they vary by machine.
+
+| Size | Playable words P5 / P50 / P95 | Long words P5 / P50 / P95 | Cell coverage P5 / P50 / P95 | Mean successful candidates | Mean generation attempts |
+| ---: | ----------------------------: | ------------------------: | ---------------------------: | -------------------------: | -----------------------: |
+|  4×4 |         125.95 / 164 / 226.05 |          38 / 67 / 107.05 |               0.9375 / 1 / 1 |                         32 |                   39.316 |
+|  5×5 |               250 / 307 / 398 |          98 / 147 / 220.1 |                 0.92 / 1 / 1 |                         32 |                   38.832 |
+|  6×6 |         404.95 / 479 / 603.05 |     179.95 / 244 / 357.05 |   0.9444444444444444 / 1 / 1 |                     31.998 |                   43.086 |
+
+One representative local run measured these complete selector runtimes:
+
+| Size | Mean ms | P50 ms | P95 ms |
+| ---: | ------: | -----: | -----: |
+|  4×4 |   9.537 |  9.402 | 10.562 |
+|  5×5 |  20.046 | 19.994 | 22.020 |
+|  6×6 |  36.022 | 35.722 | 38.966 |
+
+## Original-corpus comparison
+
+The retained original corpus contains 5,000 independently generated boards per
+size. The strong selector's median is above that corpus's P95 for every size:
+
+| Size | Original P5 / P50 / P95 | Selected P5 / P50 / P95 | Selected P50 minus original P95 |
+| ---: | ----------------------: | ----------------------: | ------------------------------: |
+|  4×4 |           31 / 72 / 144 |   125.95 / 164 / 226.05 |                             +20 |
+|  5×5 |          77 / 153 / 267 |         250 / 307 / 398 |                             +40 |
+|  6×6 |      148 / 262 / 428.05 |   404.95 / 479 / 603.05 |                          +50.95 |
+
+Selected P5 remains slightly below the original P95, by 18.05, 17, and 23.10
+words respectively. The complete selected distributions nevertheless occupy
+the intended upper tail rather than clustering around the median targets. This
+audit establishes selection behavior, not whether denser boards are more fun;
+that remains the purpose of physical testing.
diff --git a/docs/board-quality-research/generated/strong-selection-summary.json b/docs/board-quality-research/generated/strong-selection-summary.json
@@ -0,0 +1,149 @@
+{
+  "schemaVersion": 1,
+  "seed": "strong-board-selection-v1",
+  "selectionsPerSize": 500,
+  "candidatePoolSize": 32,
+  "ranking": [
+    "highest playable-word count",
+    "highest cell coverage",
+    "earliest candidate generation index"
+  ],
+  "sizes": {
+    "4": {
+      "originalFiveThousandBoardCorpus": {
+        "playableWordCount": {
+          "min": 6,
+          "p5": 31,
+          "p50": 72,
+          "p95": 144,
+          "p99": 180,
+          "max": 276
+        },
+        "longPlayableWordCount": {
+          "min": 0,
+          "p5": 3,
+          "p50": 18,
+          "p95": 58,
+          "p99": 81.01000000000022,
+          "max": 155
+        },
+        "cellCoverage": {
+          "min": 0.5625,
+          "p5": 0.875,
+          "p50": 1,
+          "p95": 1,
+          "p99": 1,
+          "max": 1
+        }
+      },
+      "selected": {
+        "playableWordCount": {
+          "p5": 125.95,
+          "p50": 164,
+          "p95": 226.04999999999995
+        },
+        "longPlayableWordCount": {
+          "p5": 38,
+          "p50": 67,
+          "p95": 107.04999999999995
+        },
+        "cellCoverage": { "p5": 0.9375, "p50": 1, "p95": 1 }
+      },
+      "upperTailComparison": {
+        "selectedP5MinusOriginalP95": -18.049999999999997,
+        "selectedP50MinusOriginalP95": 20,
+        "selectedP5AtOrAboveOriginalP95": false
+      },
+      "meanSuccessfulCandidateCount": 32,
+      "meanUnderlyingGenerationAttempts": 39.316
+    },
+    "5": {
+      "originalFiveThousandBoardCorpus": {
+        "playableWordCount": {
+          "min": 23,
+          "p5": 77,
+          "p50": 153,
+          "p95": 267,
+          "p99": 330.0100000000002,
+          "max": 436
+        },
+        "longPlayableWordCount": {
+          "min": 2,
+          "p5": 15,
+          "p50": 51,
+          "p95": 126,
+          "p99": 172,
+          "max": 256
+        },
+        "cellCoverage": {
+          "min": 0.6,
+          "p5": 0.88,
+          "p50": 1,
+          "p95": 1,
+          "p99": 1,
+          "max": 1
+        }
+      },
+      "selected": {
+        "playableWordCount": { "p5": 250, "p50": 307, "p95": 398 },
+        "longPlayableWordCount": {
+          "p5": 98,
+          "p50": 147,
+          "p95": 220.0999999999999
+        },
+        "cellCoverage": { "p5": 0.92, "p50": 1, "p95": 1 }
+      },
+      "upperTailComparison": {
+        "selectedP5MinusOriginalP95": -17,
+        "selectedP50MinusOriginalP95": 40,
+        "selectedP5AtOrAboveOriginalP95": false
+      },
+      "meanSuccessfulCandidateCount": 32,
+      "meanUnderlyingGenerationAttempts": 38.832
+    },
+    "6": {
+      "originalFiveThousandBoardCorpus": {
+        "playableWordCount": {
+          "min": 51,
+          "p5": 148,
+          "p50": 262,
+          "p95": 428.0500000000002,
+          "p99": 514.0100000000002,
+          "max": 672
+        },
+        "longPlayableWordCount": {
+          "min": 2,
+          "p5": 38,
+          "p50": 101,
+          "p95": 215,
+          "p99": 282,
+          "max": 423
+        },
+        "cellCoverage": {
+          "min": 0.75,
+          "p5": 0.9166666666666666,
+          "p50": 1,
+          "p95": 1,
+          "p99": 1,
+          "max": 1
+        }
+      },
+      "selected": {
+        "playableWordCount": { "p5": 404.95, "p50": 479, "p95": 603.05 },
+        "longPlayableWordCount": {
+          "p5": 179.95,
+          "p50": 244,
+          "p95": 357.04999999999995
+        },
+        "cellCoverage": { "p5": 0.9444444444444444, "p50": 1, "p95": 1 }
+      },
+      "upperTailComparison": {
+        "selectedP5MinusOriginalP95": -23.100000000000193,
+        "selectedP50MinusOriginalP95": 50.94999999999982,
+        "selectedP5AtOrAboveOriginalP95": false
+      },
+      "meanSuccessfulCandidateCount": 31.998,
+      "meanUnderlyingGenerationAttempts": 43.086
+    }
+  }
+}
diff --git a/package.json b/package.json
@@ -21,6 +21,7 @@
     "data:boards:audit": "tsx packages/game-data/scripts/audit-boards.mjs",
     "research:boards": "tsx scripts/research-board-quality.ts",
     "research:median-selection": "tsx scripts/audit-median-board-selection.ts",
+    "research:strong-selection": "tsx scripts/audit-strong-board-selection.ts",
     "privacy:check": "node scripts/check-public-privacy.mjs",
     "build:production": "npm run build --workspace @words/client && node scripts/build-production.mjs",
     "start:production": "node dist/production/server/index.mjs",