| name | synthesis-explore |
| description | Synthesis EXPLORE stage for the Diffmode growth-tactics pipeline (fuses the blind-combination draw + emergent-mechanism derivation — formerly two separate synthesis steps — into ONE structural-check-only stage). Reads the per-run growth-factors.json (LIGHT vector DB) + synthesis-constraints.json + founder-input + audience-jtbd + the 3 think-tank reports; draws vector combinations BLIND (IDs only, before any analysis), runs the Stripped Core Action Test + prohibited-combination + must_include checks, then derives the NOVEL MECHANISM that EMERGES from each (Reframing Test, structural-advantage classification, 3-step prototype, verb-group dedup). Outputs synthesis-explore.md — 15-20 blind combinations + 12-18 validated mechanisms, NO tactic names. Feeds synthesis-build. |
| metadata | {"version":"1.0.0"} |
Synthesis — Explore (Phase 1 Blind Combinations → Phase 2 Emergent Mechanisms)
This is the FIRST synthesis stage. It fuses two jobs that MUST stay ordered: Phase 1 draws
vector combinations BLIND (vector-first, before any tactic exists), and Phase 2 derives the
NOVEL MECHANISM that emerges from each combination. Vector-first thinking is the novelty
engine: you commit to combinations BEFORE knowing what tactic they might produce, which is how
you escape conventional playbooks. Output: 15-20 raw vector combinations + 12-18 validated
mechanisms. NO TACTIC NAMES. NO EXECUTION STEPS. (Tactic design is synthesis-build's job.)
Vectors and their definitions come from the per-run growth-factors.json (the LIGHT DB). Do
NOT read tactics_DB/ (clean-room — moat-critical). The category structure and
{prefix}-NNN-slug ID format are shared conventions; the specific IDs are whatever this run
mined.
The two phases are WALLED — do them strictly in order
You MUST complete Phase 1 (blind draw + combinations) in FULL before starting Phase 2
(mechanisms). The output must physically PROVE this ordering: the section
## Blind Draw (IDs only) must precede ## Vector Combinations, which must precede
## Validated Mechanisms. If you catch yourself deriving a mechanism (Phase 2) before the
blind draw and combinations (Phase 1) are written, STOP — you have broken the wall; go back.
Inputs & Output
The invoker provides (do not hardcode absolute paths):
WS/01-diagnostics/founder-input.md
WS/03-think-tanks/demand-generation/growth-factors.json (LIGHT vector DB — vector IDs,
categories, mechanisms, transferability, saturation_risk)
WS/03-think-tanks/demand-generation/synthesis-constraints.json (mandatory pools, white
space, prohibited combinations, category diversity)
WS/02-enrichment/audience-jtbd.md (trigger events + emotions)
- Think-tank context (use if present):
WS/03-think-tanks/demand-generation/cross-industry.md
(transferable mechanisms worth combining), competitor-gaps.md (open channels/segments),
platform-arbitrage.md (emerging-platform timing windows)
- OUTPUT:
WS/03-think-tanks/demand-generation/synthesis-explore.md
═══════════════════════════════════════════════════════════════════════════
PHASE 1 — BLIND VECTOR COMBINATIONS (no tactics, no execution)
═══════════════════════════════════════════════════════════════════════════
Force exploration of vector combinations WITHOUT thinking about tactics yet. Output of this
phase: 15-20 raw vector combinations with synergy rationale. NO TACTIC NAMES. NO EXECUTION
STEPS.
Critical Rules
YOU MUST NOT: name tactics · describe execution · pick channels · think "what marketing
to do" · generate combinations that produce CONVENTIONAL outcomes.
YOU MUST ONLY: explore vector synergies · explain WHY vectors combine · identify white
space · document founder leverage · REJECT combinations that produce conventional marketing.
If you find yourself writing a tactic name or execution step, STOP and delete it.
Conventional Outcome Detection (BLOCKING)
Some combinations ALWAYS produce conventional outcomes regardless of framing. REJECT them
before they reach Phase 2.
The Mechanism Preview Test — before accepting a combination, ask:
"If I combine these vectors, what mechanism results, in plain English?"
If it sounds like standard marketing advice, REJECT.
Apply synthesis-constraints.json → prohibited_combinations. Those are theme-level
rules (e.g. customer-research + content-flywheel = "talk to customers, write content";
build-demo + share = "build demos and share them"; community-join + audience-borrowing =
"join communities and participate"; content-flywheel + any SEO vector = "write content to
rank"; behavioral-cohort + content-funnel = "segment and optimize"). Any pair whose mined
vectors match a prohibited theme is REJECTED.
What makes a combination UNCONVENTIONAL — its mechanism (1) a marketer would advise
AGAINST ("that's risky/unusual"), (2) competitors structurally CAN'T copy (not just won't),
(3) exploits a timing window (first-mover in an emerging space), or (4) requires a specific
founder constraint (solo founder can do what enterprise can't).
The Replacement Rule — don't skip a conventional pair, REPLACE it: keep ONE vector,
swap the other for a partner (ideally an unconventional_anchor from the constraints file)
that produces an unconventional mechanism; re-run the Mechanism Preview Test.
Blind Vector Selection (BEFORE ANY ANALYSIS)
Complete this BEFORE reading enrichment deeply or thinking about tactics — commit to
combinations blind. Draw from the three mandatory pools in synthesis-constraints.json,
listing ONLY vector IDs (no descriptions/previews yet), into the output's
## Blind Draw (IDs only) section:
- Pool A — White Space (5): from
diverse_white_space / mandatory_combinations
pool A_white_space.
- Pool B — Cross-Category Synergy (5): from
mandatory_combinations pool B_synergy
(one per high-synergy category pair).
- Pool C — Founder Leverage (5): from pool
C_founder_leverage — both vectors
transferability "High", at least one "Emerging", each exploiting a specific unfair
advantage the founder possesses.
Checkpoint: 15 combinations drawn blind. If any pool is short (the LIGHT DB is smaller
than the paid DB), fill from the nearest cross-category Emerging/High-transferability pairs
in growth-factors.json and note the substitution. Only AFTER this may you analyze what
the combinations might produce.
Process
- Load founder context — Industry, Budget, Team, Skills, Stage (brief).
- Select 2-3 category clusters that fit — from the categories present in
growth-factors.json (Structural Arbitrage, Leverage, Resource Optimization,
Psychological, Positioning, Conversion), pick the 2-3 best-fit for this founder and say
why (e.g. low budget + high technical skill → Structural/Resource lean).
- Identify synergistic pairs — cross-category pairs from Pool B; explain which fit.
- Find white space — 5-8 uncommon pairs (Pool A + others); for each: why unexplored?
brilliant or foolish? fits constraints?
- Extract buyer psychology — from
audience-jtbd.md: Trigger Event (what happens
24-48h before they search) + Primary Emotion; note which vectors align.
Stripped Core Action Test (BLOCKING — run for EVERY combination)
Part 1 — Raw Action (≤5 words, NO adjectives, no vector IDs). Describe what the
combination produces using only verbs + nouns. ("Publish raw failure data" ✓; "Create
compelling contrarian content" ✗ — has adjectives.)
Part 2 — The Marketer Test. Would a generic B2B marketer recommend this EXACT raw
action? "Obviously do that" → REJECT. "That's risky / unusual / why would you?" → ACCEPT.
Part 2b — Structural Rescue (on REJECT only). If the Marketer Test says REJECT but the
competitor analysis (competitor-gaps.md Tier 1) shows a genuine structural barrier
preventing competitors from executing this exact action, OVERRIDE the reject → ACCEPT with
note [STRUCTURAL RESCUE: <barrier>]. The action looks conventional but competitors
structurally can't do it. Max 2 rescues per run.
Part 3 — Emergence Proof (accepted only). Can Vector A ALONE produce the action? Can
Vector B ALONE? If BOTH can independently → REJECT (no emergence). If EITHER alone CANNOT →
ACCEPT (true synthesis).
═══════════════════════════════════════════════════════════════════════════
PHASE 2 — EMERGENT MECHANISMS (what NEW thing emerges)
═══════════════════════════════════════════════════════════════════════════
Transform the Phase-1 vector combinations into the NOVEL MECHANISM that emerges from each.
Output of this phase: 12-18 validated mechanisms with structural-advantage analysis. No
tactic names yet.
The test: if you can't explain what EMERGES from the combination that neither vector
provides alone, the combination is weak — skip it.
Vector definitions come from the per-run growth-factors.json (each vector's mechanism
field — copy it into the "Vector Mechanisms" block for traceability). Do NOT read tactics_DB/.
Critical Rules
MUST NOT: name tactics · name channels/platforms · write execution steps · jump to "how
to do this marketing." MUST: describe the emergent MECHANISM · explain what's NOVEL ·
identify the STRUCTURAL ADVANTAGE type · skip weak combinations.
Conventional Detection Gate (BLOCKING)
For EVERY mechanism, ask: "If I stripped the vector IDs and described this to a traditional
B2B marketer, would they say 'Yes, obviously do that'?" YES → REJECT (derive a DIFFERENT
emergent capability). NO / "that's unusual" → continue.
Anti-gaming — same thing, different words (ALL REJECTED). Reframing a conventional
tactic with fancy language does not make it unconventional. Run the Reframing Test:
remove the adjectives (compelling, authentic, contrarian, compounding) — what's the CORE
ACTION? If the core action is conventional, REJECT regardless of wording. Examples of core
actions that stay conventional however they're dressed up: "talk to customers and write
content," "segment users and optimize," "share your work publicly," "build demos and share
them," "join communities and participate," "write content to rank in search."
The Unconventional Element Rule: the mechanism description must CONTAIN what makes it
non-obvious (e.g. "released as open datasets," "competitors avoid," "before the platform
saturates"), not merely imply it.
Structural Advantage Types (classify each mechanism)
- Platform Arbitrage (12-24 mo): new platform/algorithm competitors don't grok yet.
- Competitor Lockout (Permanent): brand/structural constraints stop competitors copying.
- Network Effect (Compounding): value grows with scale; late entrants can't catch up.
- Counterintuitive (Variable): traditional marketers advise against it.
- Execution Lead (3-6 mo): "we'll do it better" — NOT a real structural advantage (flag).
Execution Prototype (REQUIRED before validation — plain English, no jargon)
For EACH mechanism write a 3-step prototype:
- Day 1 — what does the founder PHYSICALLY DO?
- Day 7 — what happens next?
- Day 30 — measurable outcome?
Then the prototype gates: Q1 Is Step 1 something any marketer would recommend? YES →
CONVENTIONAL. Q2 Would executing Step 1 feel weird/risky to a traditional marketer? NO →
CONVENTIONAL; YES (give the reason) → continue. Q3 Could a competitor copy Steps 1-3
within 30 days with no structural barrier? YES → downgrade to "Execution Lead"; NO (name the
barrier) → true structural advantage.
Guerrilla Lens (tag each mechanism — positive novelty, not a gate)
After the prototype gates, tag each mechanism with 1-3 guerrilla principles it exploits.
This is a POSITIVE lens — it pushes toward mechanisms that work because of structural
asymmetries between small and large players, complementing the negative anti-conventional
tests above. A mechanism does NOT need all six; 1-2 strong tags is fine.
| Principle | Why it works for a small player | Why a funded competitor can't copy it |
|---|
| Surprise | Unexpected format, timing, or channel gets disproportionate attention | Big players plan campaigns months ahead; can't be spontaneous |
| Intimacy | Personal, 1:1, hand-crafted outreach converts at 10× mass marketing | Can't scale intimacy; delegating it makes it fake |
| Judo | Uses a competitor's size, brand, or compliance burden against them | Their size IS the vulnerability; they can't shrink |
| Counter-cyclical | Going against the grain (timing, pricing, positioning) buys cheap attention | Big players follow industry norms; deviating is career risk |
| Transparency / Absurdity | Doing something "unprofessional" works because it's honest or weird | Corporate brands get fired for this; founders get celebrated |
| Compounding | Many small consistent actions build an asset nobody can buy | Big players buy results; they don't build them patiently |
Write the tags as Guerrilla: Surprise + Judo (etc.) in each mechanism block. If a
mechanism has zero applicable tags, note Guerrilla: none — execution lead as a warning
(it may still pass the structural-advantage test, but it lacks the asymmetry multiplier).
Mid-Generation Diversity Check (after mechanisms #5, #10, #15)
List each mechanism's core action in 3-5 words; group by similarity. If 3+ share the same
core action → crutch detected: keep the 2 with the best structural advantage, kill the rest,
and for remaining combinations deliberately seek a DIFFERENT core action. Targets: ≥4
distinct core actions by #5, ≥7 by #10, ≥10 by #15.
Action Deduplication (verb-group — OUTPUT INVALID IF FAILED)
Extract the FIRST VERB of each mechanism's Day-1 action and group semantically:
| Group | Verbs |
|---|
| WRITE | Write, Publish, Create, Produce, Document, Blog, Author, Draft |
| BUILD | Build, Develop, Make, Code, Ship, Implement, Construct, Design |
| SHARE | Share, Post, Distribute, Spread, Release, Broadcast, Announce |
| JOIN | Join, Participate, Engage, Connect, Network, Comment, Reply |
| EXPORT | Export, Extract, Open-source, Expose, Dump, Upload, Provide |
| OPTIMIZE | Optimize, Target, Position, Format, Structure, Configure |
| PRE-SELL | Pre-sell, Validate, Test, Demo, Launch, Pitch, Sell |
| ANALYZE | Analyze, Research, Study, Survey, Investigate, Audit |
Hard gates: max 2 mechanisms per verb group; ≥7 distinct verb groups across 12-18
mechanisms. If violated → keep the 1-2 best per group, regenerate the rest from their
combination with a DIFFERENT first verb (favor underrepresented groups: EXPORT, PRE-SELL,
OPTIMIZE).
═══════════════════════════════════════════════════════════════════════════
Output Format (one file — Phase 1 sections, THEN Phase 2 sections)
═══════════════════════════════════════════════════════════════════════════
Save to WS/03-think-tanks/demand-generation/synthesis-explore.md. The section ORDER is the
structural proof that Phase 1 preceded Phase 2 — keep it exactly:
# Synthesis Explore — Blind Combinations & Emergent Mechanisms
## Founder Context Summary
- Industry / Budget / Team / Key Skills / Stage
## Blind Draw (IDs only)
> Drawn BEFORE any analysis (Phase 1, step 1). IDs only — no descriptions, no tactics.
- Pool A — White Space (5 pairs): `id`+`id` · `id`+`id` · … (Pool A entries are PAIRS, same as B and C)
- Pool B — Cross-Category Synergy (5): `id`+`id` · … (one per high-synergy pair)
- Pool C — Founder Leverage (5): `id`+`id` · …
- Substitutions (if the LIGHT DB was short): <pair> → <replacement id that exists in growth-factors.json>
## Selected Category Clusters
1. Primary: <category> — why; key vectors (5-7 IDs from growth-factors.json)
2. Secondary: <category> — why; key vectors
3. Tertiary (optional)
## Buyer Psychology Alignment
- Trigger Event / Primary Emotion / vectors that intercept the trigger / address the emotion
## Stripped Core Action Test
| Combination | Raw Action (≤5 words) | Marketer Test | Emergence Proof |
|-------------|----------------------|---------------|-----------------|
| A1 | ... | ACCEPT/REJECT | PASS/FAIL |
## Vector Combinations (15-20)
> **Length budget — this file has a hard ceiling.** Keep each combination to roughly 10 lines;
+ [+ ] (2-3 vectors max, never 4+)
A / B / C
...
ACCEPT · PASS
what each vector contributes (1 sentence each) + Together (2-3 sentences)
reference the Raw Action
2 sentences — what unfair advantage does this exploit?
intercepts trigger? addresses emotion?
| # | Combination | Why Unexplored? | Brilliant or Foolish? |
+
:
:
2-4 sentences — what NEW thing emerges (impossible with either alone)
2-3 sentences
Type / Duration / Why competitors can't copy (specific barrier)
1. Day 1 … 2. Day 7 … 3. Day 30 …
<1-3 principles: Surprise / Intimacy / Judo / Counter-cyclical /
Transparency-Absurdity / Compounding — or "none — execution lead">
Conventional Detection PASS (marketer would NOT recommend) · Emergence Yes
· Specificity Yes (unconventional element explicit)
| Type | Count | Mechanisms |
(target: ≥50% Platform Arbitrage / Competitor Lockout / Network Effect)
| UNCONVENTIONAL | CONVENTIONAL | (target ≥60% unconventional)
| Distinct verb groups (≥7) | Max per group (≤2) |
| Total combos | Validated mechanisms | White space | Synergistic pairs | Unique vectors |
| 15-20 | 12-18 | 5-8 | 10+ | 25+ (or as many as the LIGHT DB allows) |
Validation Checkpoint (before synthesis-build)
Phase 1 (combinations):
Phase 2 (mechanisms):
This feeds synthesis-build (white-space ideation → founder-fit adaptation → final
synthesis.md). Do not proceed until both phases' validation passes.