Brief Backlog — Cycle 2026-05-05

Audience: Creative Director + Creative Producer Hypotheses: 39 Versions to brief: 39 Active concepts: 13 Kill recommendations: 3
5
SCALE
5
ITERATE
3
EXPLORE
3
KILL
39
total versions
How to read this document. Every recommendation below is anchored to evidence from the corrected Lumina D7 pipeline (cycle 2026-05-05). Each concept has a rationale and a list of versions to produce; each version specifies what to change from the existing DNA. Hypothesis IDs (e.g. H-2026-05-05-001) follow each recommendation so they can be referenced in briefs and resolved against next cycle's data.

Terminology. The 4-tag DNA categories — strategic_angle, hook, creative_focus, gameplay_narrative_arc — carry concept identity (61% of mutual information with concept_id). The other 5 categories are peripheral execution choices. The 3 Reliable Winners are specific tag values within 3 of those DNA categories (Cognitive Challenge, Fail → Win, Problem Statement) that lift Lumina D7 at p<0.01.

Variation design. Each new version of an existing concept varies one peripheral tag while holding DNA constant. We only test peripherals with defensible positive evidence across the portfolio: None (Silent) for audio, Core Loop / Real-gameplay for representation, Always-On Banner for CTA. After 5+ cycles of structured variation, we'll have controlled experimental data to compute concept-fixed-effect tag lifts.

Producer routing. Each concept gets a concept-level routing recommendation based on each producer's track record across the concept's 4 DNA tags. EXPLOIT = highest weighted-lift producer; EXPLORE = producer with thin reps on the DNA (capability-building). Per-version routing only appears as an inline flag when a specific version uses a Hard-difficulty tag (rare in this cycle).

SCALE briefs — 5 CONSISTENT WINNERS, hold DNA, vary one peripheral per version

Why every SCALE brief looks structurally identical. All 5 SCALE concepts get the same 3 version prescriptions: V1 = None (Silent) audio, V2 = Core Loop / Real-gameplay representation, V3 = Always-On Banner CTA. This is deliberate. Five different proven concepts each running the same 3 controlled peripheral swaps gives us, after this cycle plus a few more, 5 × 3 = 15 within-concept comparisons on the same peripheral swap — the data-generating mechanism this dataset has been missing. Today's tag lifts are observational (concept-confounded); the structured repetition is what turns them into causal evidence we can trust. The visual sameness across the 5 SCALE blocks is the feature, not a bug.
C518
SCALE CONSISTENT WINNER
5 versions shipped avg D7 = 79.4 std = 13.3
C518 is a CONSISTENT WINNER (avg D7=79.4, std=13.3, 5 versions). Goal: hold the proven DNA and systematically vary ONE peripheral tag per new version. The 3 versions below test the 3 peripheral swaps with positive cross-concept evidence — generating data the concept currently lacks.
Hold the existing DNA constant:
strategic_angle: Cognitive Challengehook: Problem Statementcreative_focus: Gameplay Mechanicgameplay_narrative_arc: Fail → Win
Concept-level producer routing — based on DNA-tag track records
EXPLOIT Julie Droz +20.9% weighted (n=41 on DNA tags, 2/4 covered) · Yevhenii Hrushetskyi +5.9% weighted (n=97 on DNA tags, 4/4 covered)
EXPLORE Vira Bilous n=12 on DNA tags (build capability) · Ayca Uyanik n=18 on DNA tags (build capability)
Version-by-version prescription (3):
V1 — Core Loop / Real-gameplay (gameplay_representation)
H-2026-05-05-001
Hold DNA constant; vary gameplay_representation to test whether the under-invested choice lifts performance.
Primary change: gameplay_representation: Exaggerated / Aspirational gameplay → Core Loop / Real-gameplay
MEDIUM CONFIDENCE Expected D7 lift: +16.5% n = 28
Core Loop / Real-gameplay shows a +16.5% Lumina D7 lift across 28 creatives portfolio-wide. Currently under-invested in production — worth testing on this concept's next versions.
V2 — None (Silent) (audio_narration)
H-2026-05-05-002
Hold DNA constant; vary audio_narration to test whether the under-invested choice lifts performance.
Primary change: audio_narration: Music/SFX → None (Silent)
HIGH CONFIDENCE Expected D7 lift: +13.7% n = 107
None (Silent) shows a +13.7% Lumina D7 lift across 107 creatives portfolio-wide. Currently under-invested in production — worth testing on this concept's next versions.
V3 — Always-On Banner (cta_strategy)
H-2026-05-05-003
Hold DNA constant; vary cta_strategy to test whether the under-invested choice lifts performance.
Primary change: cta_strategy: End-Card Only → Always-On Banner
MEDIUM CONFIDENCE Expected D7 lift: +5.0% n = 46
Always-On Banner shows a +5.0% Lumina D7 lift across 46 creatives portfolio-wide. Currently under-invested in production — worth testing on this concept's next versions.
C431
SCALE CONSISTENT WINNER
3 versions shipped avg D7 = 73.0 std = 14.2
C431 is a CONSISTENT WINNER (avg D7=73.0, std=14.2, 3 versions). Goal: hold the proven DNA and systematically vary ONE peripheral tag per new version. The 3 versions below test the 3 peripheral swaps with positive cross-concept evidence — generating data the concept currently lacks.
Hold the existing DNA constant:
strategic_angle: Progress & Achievementhook: None (No Hook)creative_focus: Gameplay Mechanicgameplay_narrative_arc: Fail → Win
Concept-level producer routing — based on DNA-tag track records
EXPLOIT Julie Droz +10.1% weighted (n=89 on DNA tags, 3/4 covered) · Yevhenii Hrushetskyi +7.4% weighted (n=123 on DNA tags, 4/4 covered)
EXPLORE Vira Bilous n=19 on DNA tags (build capability) · Ana Narchemashvili n=10 on DNA tags (build capability)
Version-by-version prescription (3):
V1 — Core Loop / Real-gameplay (gameplay_representation)
H-2026-05-05-004
Hold DNA constant; vary gameplay_representation to test whether the under-invested choice lifts performance.
Primary change: gameplay_representation: Exaggerated / Aspirational gameplay → Core Loop / Real-gameplay
MEDIUM CONFIDENCE Expected D7 lift: +16.5% n = 28
Core Loop / Real-gameplay shows a +16.5% Lumina D7 lift across 28 creatives portfolio-wide. Currently under-invested in production — worth testing on this concept's next versions.
V2 — None (Silent) (audio_narration)
H-2026-05-05-005
Hold DNA constant; vary audio_narration to test whether the under-invested choice lifts performance.
Primary change: audio_narration: Music/SFX → None (Silent)
HIGH CONFIDENCE Expected D7 lift: +13.7% n = 107
None (Silent) shows a +13.7% Lumina D7 lift across 107 creatives portfolio-wide. Currently under-invested in production — worth testing on this concept's next versions.
V3 — Always-On Banner (cta_strategy)
H-2026-05-05-006
Hold DNA constant; vary cta_strategy to test whether the under-invested choice lifts performance.
Primary change: cta_strategy: End-Card Only → Always-On Banner
MEDIUM CONFIDENCE Expected D7 lift: +5.0% n = 46
Always-On Banner shows a +5.0% Lumina D7 lift across 46 creatives portfolio-wide. Currently under-invested in production — worth testing on this concept's next versions.
C419
SCALE CONSISTENT WINNER
4 versions shipped avg D7 = 71.8 std = 8.2
C419 is a CONSISTENT WINNER (avg D7=71.8, std=8.2, 4 versions). Goal: hold the proven DNA and systematically vary ONE peripheral tag per new version. The 3 versions below test the 3 peripheral swaps with positive cross-concept evidence — generating data the concept currently lacks.
Hold the existing DNA constant:
strategic_angle: Cognitive Challengehook: Problem Statementcreative_focus: Gameplay Mechanicgameplay_narrative_arc: Fail → Win
Concept-level producer routing — based on DNA-tag track records
EXPLOIT Julie Droz +20.9% weighted (n=41 on DNA tags, 2/4 covered) · Yevhenii Hrushetskyi +5.9% weighted (n=97 on DNA tags, 4/4 covered)
EXPLORE Vira Bilous n=12 on DNA tags (build capability) · Ayca Uyanik n=18 on DNA tags (build capability)
Version-by-version prescription (3):
V1 — Core Loop / Real-gameplay (gameplay_representation)
H-2026-05-05-007
Hold DNA constant; vary gameplay_representation to test whether the under-invested choice lifts performance.
Primary change: gameplay_representation: Exaggerated / Aspirational gameplay → Core Loop / Real-gameplay
MEDIUM CONFIDENCE Expected D7 lift: +16.5% n = 28
Core Loop / Real-gameplay shows a +16.5% Lumina D7 lift across 28 creatives portfolio-wide. Currently under-invested in production — worth testing on this concept's next versions.
V2 — None (Silent) (audio_narration)
H-2026-05-05-008
Hold DNA constant; vary audio_narration to test whether the under-invested choice lifts performance.
Primary change: audio_narration: Music/SFX → None (Silent)
HIGH CONFIDENCE Expected D7 lift: +13.7% n = 107
None (Silent) shows a +13.7% Lumina D7 lift across 107 creatives portfolio-wide. Currently under-invested in production — worth testing on this concept's next versions.
V3 — Always-On Banner (cta_strategy)
H-2026-05-05-009
Hold DNA constant; vary cta_strategy to test whether the under-invested choice lifts performance.
Primary change: cta_strategy: End-Card Only → Always-On Banner
MEDIUM CONFIDENCE Expected D7 lift: +5.0% n = 46
Always-On Banner shows a +5.0% Lumina D7 lift across 46 creatives portfolio-wide. Currently under-invested in production — worth testing on this concept's next versions.
C504
SCALE CONSISTENT WINNER
5 versions shipped avg D7 = 68.6 std = 8.3
C504 is a CONSISTENT WINNER (avg D7=68.6, std=8.3, 5 versions). Goal: hold the proven DNA and systematically vary ONE peripheral tag per new version. The 3 versions below test the 3 peripheral swaps with positive cross-concept evidence — generating data the concept currently lacks.
Hold the existing DNA constant:
strategic_angle: Progress & Achievementhook: Problem Statementcreative_focus: Gameplay Mechanicgameplay_narrative_arc: Fail → Win
Concept-level producer routing — based on DNA-tag track records
EXPLOIT Julie Droz +20.9% weighted (n=41 on DNA tags, 2/4 covered) · Yevhenii Hrushetskyi +9.6% weighted (n=103 on DNA tags, 4/4 covered)
EXPLORE Vira Bilous n=12 on DNA tags (build capability) · Ayca Uyanik n=18 on DNA tags (build capability)
Version-by-version prescription (3):
V1 — Core Loop / Real-gameplay (gameplay_representation)
H-2026-05-05-010
Hold DNA constant; vary gameplay_representation to test whether the under-invested choice lifts performance.
Primary change: gameplay_representation: Exaggerated / Aspirational gameplay → Core Loop / Real-gameplay
MEDIUM CONFIDENCE Expected D7 lift: +16.5% n = 28
Core Loop / Real-gameplay shows a +16.5% Lumina D7 lift across 28 creatives portfolio-wide. Currently under-invested in production — worth testing on this concept's next versions.
V2 — None (Silent) (audio_narration)
H-2026-05-05-011
Hold DNA constant; vary audio_narration to test whether the under-invested choice lifts performance.
Primary change: audio_narration: Music/SFX → None (Silent)
HIGH CONFIDENCE Expected D7 lift: +13.7% n = 107
None (Silent) shows a +13.7% Lumina D7 lift across 107 creatives portfolio-wide. Currently under-invested in production — worth testing on this concept's next versions.
V3 — Always-On Banner (cta_strategy)
H-2026-05-05-012
Hold DNA constant; vary cta_strategy to test whether the under-invested choice lifts performance.
Primary change: cta_strategy: End-Card Only → Always-On Banner
MEDIUM CONFIDENCE Expected D7 lift: +5.0% n = 46
Always-On Banner shows a +5.0% Lumina D7 lift across 46 creatives portfolio-wide. Currently under-invested in production — worth testing on this concept's next versions.
C469
SCALE CONSISTENT WINNER
3 versions shipped avg D7 = 65.7 std = 2.3
C469 is a CONSISTENT WINNER (avg D7=65.7, std=2.3, 3 versions). Goal: hold the proven DNA and systematically vary ONE peripheral tag per new version. The 3 versions below test the 3 peripheral swaps with positive cross-concept evidence — generating data the concept currently lacks.
Hold the existing DNA constant:
strategic_angle: Relax & Escapehook: Surprise / Unexpectedcreative_focus: Gameplay Mechanicgameplay_narrative_arc: Order → Chaos
Concept-level producer routing — based on DNA-tag track records
EXPLOIT Julie Droz +20.5% weighted (n=140 on DNA tags, 4/4 covered) · Alexandre de Crozals +9.1% weighted (n=22 on DNA tags, 3/4 covered)
EXPLORE Ayca Uyanik n=65 on DNA tags (build capability) · Jeremy Laplatine n=601 on DNA tags (build capability)
Version-by-version prescription (3):
V1 — Core Loop / Real-gameplay (gameplay_representation)
H-2026-05-05-013
Hold DNA constant; vary gameplay_representation to test whether the under-invested choice lifts performance.
Primary change: gameplay_representation: Exaggerated / Aspirational gameplay → Core Loop / Real-gameplay
MEDIUM CONFIDENCE Expected D7 lift: +16.5% n = 28
Core Loop / Real-gameplay shows a +16.5% Lumina D7 lift across 28 creatives portfolio-wide. Currently under-invested in production — worth testing on this concept's next versions.
V2 — None (Silent) (audio_narration)
H-2026-05-05-014
Hold DNA constant; vary audio_narration to test whether the under-invested choice lifts performance.
Primary change: audio_narration: Music/SFX → None (Silent)
HIGH CONFIDENCE Expected D7 lift: +13.7% n = 107
None (Silent) shows a +13.7% Lumina D7 lift across 107 creatives portfolio-wide. Currently under-invested in production — worth testing on this concept's next versions.
V3 — Always-On Banner (cta_strategy)
H-2026-05-05-015
Hold DNA constant; vary cta_strategy to test whether the under-invested choice lifts performance.
Primary change: cta_strategy: End-Card Only → Always-On Banner
MEDIUM CONFIDENCE Expected D7 lift: +5.0% n = 46
Always-On Banner shows a +5.0% Lumina D7 lift across 46 creatives portfolio-wide. Currently under-invested in production — worth testing on this concept's next versions.

ITERATE briefs — 5 HIGH-CEILING BETS, hold DNA, more aggressive peripheral variation

Same 3 peripheral swaps, plus a 4th DNA-swap version where evidence supports it. ITERATE briefs mirror the SCALE design (V1–V3 are the same 3 defensible peripheral swaps) so we can compare peripheral effects on CONSISTENT WINNERS vs HIGH-CEILING BETs directly. Where within-concept evidence already supports a DNA-tag swap, a V4 is added — flagged as a DNA change because the resulting creative effectively becomes a sibling concept rather than a tighter version of the same idea.
C484
ITERATE HIGH-CEILING BET
7 versions shipped avg D7 = 67.9 std = 25.6
C484 is a HIGH-CEILING BET (avg D7=67.9, std=25.6, 7 versions). High within-concept variance means the best version beats the worst by a lot — iteration captures real upside. The 3 peripheral variation versions below mirror the SCALE briefs (same defensible swaps); a 4th DNA-swap version is added when within-concept evidence supports it.
Hold the existing DNA constant:
strategic_angle: Cognitive Challengehook: Problem Statementcreative_focus: Gameplay Mechanicgameplay_narrative_arc: Fail → Win
Concept-level producer routing — based on DNA-tag track records
EXPLOIT Julie Droz +20.9% weighted (n=41 on DNA tags, 2/4 covered) · Yevhenii Hrushetskyi +5.9% weighted (n=97 on DNA tags, 4/4 covered)
EXPLORE Vira Bilous n=12 on DNA tags (build capability) · Ayca Uyanik n=18 on DNA tags (build capability)
Version-by-version prescription (3):
V1 — Core Loop / Real-gameplay (gameplay_representation)
H-2026-05-05-016
Hold DNA constant; vary gameplay_representation to test whether the under-invested choice lifts performance.
Primary change: gameplay_representation: Exaggerated / Aspirational gameplay → Core Loop / Real-gameplay
MEDIUM CONFIDENCE Expected D7 lift: +16.5% n = 28
Core Loop / Real-gameplay shows a +16.5% Lumina D7 lift across 28 creatives portfolio-wide. Currently under-invested in production — worth testing on this concept's next versions.
V2 — None (Silent) (audio_narration)
H-2026-05-05-017
Hold DNA constant; vary audio_narration to test whether the under-invested choice lifts performance.
Primary change: audio_narration: Music/SFX → None (Silent)
HIGH CONFIDENCE Expected D7 lift: +13.7% n = 107
None (Silent) shows a +13.7% Lumina D7 lift across 107 creatives portfolio-wide. Currently under-invested in production — worth testing on this concept's next versions.
V3 — Always-On Banner (cta_strategy)
H-2026-05-05-018
Hold DNA constant; vary cta_strategy to test whether the under-invested choice lifts performance.
Primary change: cta_strategy: End-Card Only → Always-On Banner
MEDIUM CONFIDENCE Expected D7 lift: +5.0% n = 46
Always-On Banner shows a +5.0% Lumina D7 lift across 46 creatives portfolio-wide. Currently under-invested in production — worth testing on this concept's next versions.
C488
ITERATE HIGH-CEILING BET
6 versions shipped avg D7 = 65.3 std = 23.4
C488 is a HIGH-CEILING BET (avg D7=65.3, std=23.4, 6 versions). High within-concept variance means the best version beats the worst by a lot — iteration captures real upside. The 3 peripheral variation versions below mirror the SCALE briefs (same defensible swaps); a 4th DNA-swap version is added when within-concept evidence supports it.
Hold the existing DNA constant:
strategic_angle: Relax & Escapehook: None (No Hook)creative_focus: Gameplay Mechanicgameplay_narrative_arc: Order → Chaos
Concept-level producer routing — based on DNA-tag track records
EXPLOIT Julie Droz +15.0% weighted (n=179 on DNA tags, 4/4 covered) · Alexandre de Crozals +5.2% weighted (n=13 on DNA tags, 2/4 covered)
EXPLORE Jeremy Laplatine n=605 on DNA tags (build capability) · Ana Narchemashvili n=10 on DNA tags (build capability)
Version-by-version prescription (3):
V1 — Core Loop / Real-gameplay (gameplay_representation)
H-2026-05-05-019
Hold DNA constant; vary gameplay_representation to test whether the under-invested choice lifts performance.
Primary change: gameplay_representation: Exaggerated / Aspirational gameplay → Core Loop / Real-gameplay
MEDIUM CONFIDENCE Expected D7 lift: +16.5% n = 28
Core Loop / Real-gameplay shows a +16.5% Lumina D7 lift across 28 creatives portfolio-wide. Currently under-invested in production — worth testing on this concept's next versions.
V2 — None (Silent) (audio_narration)
H-2026-05-05-020
Hold DNA constant; vary audio_narration to test whether the under-invested choice lifts performance.
Primary change: audio_narration: Music/SFX → None (Silent)
HIGH CONFIDENCE Expected D7 lift: +13.7% n = 107
None (Silent) shows a +13.7% Lumina D7 lift across 107 creatives portfolio-wide. Currently under-invested in production — worth testing on this concept's next versions.
V3 — Always-On Banner (cta_strategy)
H-2026-05-05-021
Hold DNA constant; vary cta_strategy to test whether the under-invested choice lifts performance.
Primary change: cta_strategy: End-Card Only → Always-On Banner
MEDIUM CONFIDENCE Expected D7 lift: +5.0% n = 46
Always-On Banner shows a +5.0% Lumina D7 lift across 46 creatives portfolio-wide. Currently under-invested in production — worth testing on this concept's next versions.
C523
ITERATE HIGH-CEILING BET
6 versions shipped avg D7 = 62.7 std = 29.1
C523 is a HIGH-CEILING BET (avg D7=62.7, std=29.1, 6 versions). High within-concept variance means the best version beats the worst by a lot — iteration captures real upside. The 3 peripheral variation versions below mirror the SCALE briefs (same defensible swaps); a 4th DNA-swap version is added when within-concept evidence supports it.
Hold the existing DNA constant:
strategic_angle: Relax & Escapehook: None (No Hook)creative_focus: Gameplay Mechanicgameplay_narrative_arc: Levels mix
Concept-level producer routing — based on DNA-tag track records
EXPLOIT Julie Droz +13.5% weighted (n=139 on DNA tags, 4/4 covered) · Alexandre de Crozals +5.2% weighted (n=13 on DNA tags, 2/4 covered)
EXPLORE Vira Bilous n=36 on DNA tags (build capability) · Ana Narchemashvili n=10 on DNA tags (build capability)
Version-by-version prescription (3):
V1 — Core Loop / Real-gameplay (gameplay_representation)
H-2026-05-05-022
Hold DNA constant; vary gameplay_representation to test whether the under-invested choice lifts performance.
Primary change: gameplay_representation: Exaggerated / Aspirational gameplay → Core Loop / Real-gameplay
MEDIUM CONFIDENCE Expected D7 lift: +16.5% n = 28
Core Loop / Real-gameplay shows a +16.5% Lumina D7 lift across 28 creatives portfolio-wide. Currently under-invested in production — worth testing on this concept's next versions.
V2 — None (Silent) (audio_narration)
H-2026-05-05-023
Hold DNA constant; vary audio_narration to test whether the under-invested choice lifts performance.
Primary change: audio_narration: Music/SFX → None (Silent)
HIGH CONFIDENCE Expected D7 lift: +13.7% n = 107
None (Silent) shows a +13.7% Lumina D7 lift across 107 creatives portfolio-wide. Currently under-invested in production — worth testing on this concept's next versions.
V3 — Always-On Banner (cta_strategy)
H-2026-05-05-024
Hold DNA constant; vary cta_strategy to test whether the under-invested choice lifts performance.
Primary change: cta_strategy: Always-On Banner → Always-On Banner
MEDIUM CONFIDENCE Expected D7 lift: +5.0% n = 46
Always-On Banner shows a +5.0% Lumina D7 lift across 46 creatives portfolio-wide. Currently under-invested in production — worth testing on this concept's next versions.
R1626
ITERATE HIGH-CEILING BET
6 versions shipped avg D7 = 62.4 std = 25.2
R1626 is a HIGH-CEILING BET (avg D7=62.4, std=25.2, 6 versions). High within-concept variance means the best version beats the worst by a lot — iteration captures real upside. The 3 peripheral variation versions below mirror the SCALE briefs (same defensible swaps); a 4th DNA-swap version is added when within-concept evidence supports it.
Hold the existing DNA constant:
strategic_angle: Relax & Escapehook: Surprise / Unexpectedcreative_focus: Gameplay Mechanicgameplay_narrative_arc: Order → Chaos
Concept-level producer routing — based on DNA-tag track records
EXPLOIT Julie Droz +20.5% weighted (n=140 on DNA tags, 4/4 covered) · Alexandre de Crozals +9.1% weighted (n=22 on DNA tags, 3/4 covered)
EXPLORE Ayca Uyanik n=65 on DNA tags (build capability) · Jeremy Laplatine n=601 on DNA tags (build capability)
Version-by-version prescription (3):
V1 — Core Loop / Real-gameplay (gameplay_representation)
H-2026-05-05-025
Hold DNA constant; vary gameplay_representation to test whether the under-invested choice lifts performance.
Primary change: gameplay_representation: Exaggerated / Aspirational gameplay → Core Loop / Real-gameplay
MEDIUM CONFIDENCE Expected D7 lift: +16.5% n = 28
Core Loop / Real-gameplay shows a +16.5% Lumina D7 lift across 28 creatives portfolio-wide. Currently under-invested in production — worth testing on this concept's next versions.
V2 — None (Silent) (audio_narration)
H-2026-05-05-026
Hold DNA constant; vary audio_narration to test whether the under-invested choice lifts performance.
Primary change: audio_narration: Music/SFX → None (Silent)
HIGH CONFIDENCE Expected D7 lift: +13.7% n = 107
None (Silent) shows a +13.7% Lumina D7 lift across 107 creatives portfolio-wide. Currently under-invested in production — worth testing on this concept's next versions.
V3 — Always-On Banner (cta_strategy)
H-2026-05-05-027
Hold DNA constant; vary cta_strategy to test whether the under-invested choice lifts performance.
Primary change: cta_strategy: None (No CTA) → Always-On Banner
MEDIUM CONFIDENCE Expected D7 lift: +5.0% n = 46
Always-On Banner shows a +5.0% Lumina D7 lift across 46 creatives portfolio-wide. Currently under-invested in production — worth testing on this concept's next versions.
C524
ITERATE HIGH-CEILING BET
5 versions shipped avg D7 = 62.3 std = 23.0
C524 is a HIGH-CEILING BET (avg D7=62.3, std=23.0, 5 versions). High within-concept variance means the best version beats the worst by a lot — iteration captures real upside. The 3 peripheral variation versions below mirror the SCALE briefs (same defensible swaps); a 4th DNA-swap version is added when within-concept evidence supports it.
Hold the existing DNA constant:
strategic_angle: Explore/Discoveryhook: Surprise / Unexpectedcreative_focus: Gameplay Mechanicgameplay_narrative_arc: Order → Chaos
Concept-level producer routing — based on DNA-tag track records
EXPLOIT Julie Droz +20.2% weighted (n=94 on DNA tags, 4/4 covered) · Alexandre de Crozals +11.0% weighted (n=14 on DNA tags, 2/4 covered)
EXPLORE Ayca Uyanik n=45 on DNA tags (build capability) · Jeremy Laplatine n=443 on DNA tags (build capability)
Version-by-version prescription (3):
V1 — Core Loop / Real-gameplay (gameplay_representation)
H-2026-05-05-028
Hold DNA constant; vary gameplay_representation to test whether the under-invested choice lifts performance.
Primary change: gameplay_representation: Exaggerated / Aspirational gameplay → Core Loop / Real-gameplay
MEDIUM CONFIDENCE Expected D7 lift: +16.5% n = 28
Core Loop / Real-gameplay shows a +16.5% Lumina D7 lift across 28 creatives portfolio-wide. Currently under-invested in production — worth testing on this concept's next versions.
V2 — None (Silent) (audio_narration)
H-2026-05-05-029
Hold DNA constant; vary audio_narration to test whether the under-invested choice lifts performance.
Primary change: audio_narration: Music/SFX → None (Silent)
HIGH CONFIDENCE Expected D7 lift: +13.7% n = 107
None (Silent) shows a +13.7% Lumina D7 lift across 107 creatives portfolio-wide. Currently under-invested in production — worth testing on this concept's next versions.
V3 — Always-On Banner (cta_strategy)
H-2026-05-05-030
Hold DNA constant; vary cta_strategy to test whether the under-invested choice lifts performance.
Primary change: cta_strategy: End-Card Only → Always-On Banner
MEDIUM CONFIDENCE Expected D7 lift: +5.0% n = 46
Always-On Banner shows a +5.0% Lumina D7 lift across 46 creatives portfolio-wide. Currently under-invested in production — worth testing on this concept's next versions.

EXPLORE briefs — 3 new concepts on under-explored DNA recipes

NEW: Recipe-01-Relax&Esca-Order→Ch
EXPLORE Under-explored DNA recipe
3 creatives in this DNA today observed avg D7 = 73.0
This 4-tag DNA combination has 3 existing creatives averaging D7=73.0. The recipe is flagged is_underexplored=True (appears far less often than expected given the tag distribution). Probe whether the signal scales as a fresh concept.
Prescribed DNA for the new concept:
strategic_angle: Relax & Escapehook: None (No Hook)creative_focus: Gameplay Mechanicgameplay_narrative_arc: Order → Chaos
target_concept_name: [Creative Producer assigns]  ·  Producer routing: To be decided when concept_id is assigned — per-version best-executor info shown inline below.
Version-by-version prescription (3):
V1 — Recipe replication
H-2026-05-05-031
Replicate the observed-good DNA combination as a fresh concept. Measures whether the recipe's signal in existing creatives transfers when a brand-new concept_id is shipped under it.
Primary change: (no change — replicate as-is)
EXPLORATORY Expected D7 lift: ~73 D7 (matching observed cohort avg) n = 3
Existing creatives in this DNA (n=3) average D7=73.0. Replicating the recipe as a fresh concept tests whether the signal is real or sample-size-driven.
⚠ Hard-difficulty tag
Best on this tag value: Julie Droz +15.6% (n=51) · Nikola Kachanski +2.7% (n=231)
V2 — Core Loop / Real-gameplay (gameplay_representation)
H-2026-05-05-032
Hold DNA constant; vary gameplay_representation to test whether the under-invested choice lifts performance.
Primary change: gameplay_representation: varies → Core Loop / Real-gameplay
MEDIUM CONFIDENCE Expected D7 lift: +16.5% n = 28
Core Loop / Real-gameplay shows a +16.5% Lumina D7 lift across 28 creatives portfolio-wide. Currently under-invested in production — worth testing on this concept's next versions.
Best on this tag value: Alexandre de Crozals +-13.8% (n=11)
V3 — None (Silent) (audio_narration)
H-2026-05-05-033
Hold DNA constant; vary audio_narration to test whether the under-invested choice lifts performance.
Primary change: audio_narration: varies → None (Silent)
HIGH CONFIDENCE Expected D7 lift: +13.7% n = 107
None (Silent) shows a +13.7% Lumina D7 lift across 107 creatives portfolio-wide. Currently under-invested in production — worth testing on this concept's next versions.
NEW: Recipe-02-Destruction-None(NoA
EXPLORE Under-explored DNA recipe
4 creatives in this DNA today observed avg D7 = 66.4
This 4-tag DNA combination has 4 existing creatives averaging D7=66.4. The recipe is flagged is_underexplored=True (appears far less often than expected given the tag distribution). Probe whether the signal scales as a fresh concept.
Prescribed DNA for the new concept:
strategic_angle: Destruction & Chaoshook: Social Proofcreative_focus: Gameplay Mechanicgameplay_narrative_arc: None (No Arc)
target_concept_name: [Creative Producer assigns]  ·  Producer routing: To be decided when concept_id is assigned — per-version best-executor info shown inline below.
Version-by-version prescription (3):
V1 — Recipe replication
H-2026-05-05-034
Replicate the observed-good DNA combination as a fresh concept. Measures whether the recipe's signal in existing creatives transfers when a brand-new concept_id is shipped under it.
Primary change: (no change — replicate as-is)
EXPLORATORY Expected D7 lift: ~66 D7 (matching observed cohort avg) n = 4
Existing creatives in this DNA (n=4) average D7=66.4. Replicating the recipe as a fresh concept tests whether the signal is real or sample-size-driven.
V2 — Core Loop / Real-gameplay (gameplay_representation)
H-2026-05-05-035
Hold DNA constant; vary gameplay_representation to test whether the under-invested choice lifts performance.
Primary change: gameplay_representation: varies → Core Loop / Real-gameplay
MEDIUM CONFIDENCE Expected D7 lift: +16.5% n = 28
Core Loop / Real-gameplay shows a +16.5% Lumina D7 lift across 28 creatives portfolio-wide. Currently under-invested in production — worth testing on this concept's next versions.
Best on this tag value: Alexandre de Crozals +-13.8% (n=11)
V3 — None (Silent) (audio_narration)
H-2026-05-05-036
Hold DNA constant; vary audio_narration to test whether the under-invested choice lifts performance.
Primary change: audio_narration: varies → None (Silent)
HIGH CONFIDENCE Expected D7 lift: +13.7% n = 107
None (Silent) shows a +13.7% Lumina D7 lift across 107 creatives portfolio-wide. Currently under-invested in production — worth testing on this concept's next versions.
NEW: Recipe-03-Relax&Esca-Order→Ch
EXPLORE Under-explored DNA recipe
4 creatives in this DNA today observed avg D7 = 66.3
This 4-tag DNA combination has 4 existing creatives averaging D7=66.3. The recipe is flagged is_underexplored=True (appears far less often than expected given the tag distribution). Probe whether the signal scales as a fresh concept.
Prescribed DNA for the new concept:
strategic_angle: Relax & Escapehook: None (No Hook)creative_focus: Gameplay Mechanicgameplay_narrative_arc: Order → Chaos
target_concept_name: [Creative Producer assigns]  ·  Producer routing: To be decided when concept_id is assigned — per-version best-executor info shown inline below.
Version-by-version prescription (3):
V1 — Recipe replication
H-2026-05-05-037
Replicate the observed-good DNA combination as a fresh concept. Measures whether the recipe's signal in existing creatives transfers when a brand-new concept_id is shipped under it.
Primary change: (no change — replicate as-is)
EXPLORATORY Expected D7 lift: ~66 D7 (matching observed cohort avg) n = 4
Existing creatives in this DNA (n=4) average D7=66.3. Replicating the recipe as a fresh concept tests whether the signal is real or sample-size-driven.
⚠ Hard-difficulty tag
Best on this tag value: Julie Droz +15.6% (n=51) · Nikola Kachanski +2.7% (n=231)
V2 — Core Loop / Real-gameplay (gameplay_representation)
H-2026-05-05-038
Hold DNA constant; vary gameplay_representation to test whether the under-invested choice lifts performance.
Primary change: gameplay_representation: varies → Core Loop / Real-gameplay
MEDIUM CONFIDENCE Expected D7 lift: +16.5% n = 28
Core Loop / Real-gameplay shows a +16.5% Lumina D7 lift across 28 creatives portfolio-wide. Currently under-invested in production — worth testing on this concept's next versions.
Best on this tag value: Alexandre de Crozals +-13.8% (n=11)
V3 — None (Silent) (audio_narration)
H-2026-05-05-039
Hold DNA constant; vary audio_narration to test whether the under-invested choice lifts performance.
Primary change: audio_narration: varies → None (Silent)
HIGH CONFIDENCE Expected D7 lift: +13.7% n = 107
None (Silent) shows a +13.7% Lumina D7 lift across 107 creatives portfolio-wide. Currently under-invested in production — worth testing on this concept's next versions.

KILL recommendations — 3 STABLE UNDERPERFORMERS, deprioritize

C418
KILL
avg D7 = 6.8 · std = 5.3 · 3 versions
C418 is a STABLE UNDERPERFORMER. Avg D7=6.8 after 3 versions with std=5.3 — reliably underperforming, low iteration upside, no outlier hope. Recommend: stop iteration. Reallocate the production budget to SCALE / ITERATE / EXPLORE briefs above.
~4 versions of motion-design capacity freed per cycle
C483
KILL
avg D7 = 14.8 · std = 12.0 · 5 versions
C483 is a STABLE UNDERPERFORMER. Avg D7=14.8 after 5 versions with std=12.0 — reliably underperforming, low iteration upside, no outlier hope. Recommend: stop iteration. Reallocate the production budget to SCALE / ITERATE / EXPLORE briefs above.
~8 versions of motion-design capacity freed per cycle
E33
KILL
avg D7 = 16.7 · std = 14.4 · 3 versions
E33 is a STABLE UNDERPERFORMER. Avg D7=16.7 after 3 versions with std=14.4 — reliably underperforming, low iteration upside, no outlier hope. Recommend: stop iteration. Reallocate the production budget to SCALE / ITERATE / EXPLORE briefs above.
~4 versions of motion-design capacity freed per cycle
Brief Backlog v1 · Cycle 2026-05-05 · Source: outputs_3/brief_backlog.json · Generated by build_brief_html.py