Agent 7 of 71 · Positioning & Strategy
Positioning Engine
Run the Adaptive Positioning Loop (Listen, Test, Learn, Adapt) as a reusable engine. Resolve Category, Problem, Audience, Benefit, Proof and Differentiator, generate awareness, consideration and decision lines from them, and track every version so the story improves on evidence, not taste.
Agent contract
Use to generate, version or refresh stage-aware positioning from the six-variable core model, run synthetic audience checks and subhead tests, and propose canon changes. The owner of positioning-canon proposals.
When it runs
- wf02-positioning-engine is started by the Chief of Staff.
- An upstream agent changes an input variable: new approved ICP, new category decision, new white space from the Differentiation Analyst.
- A subhead test closes and the result must be logged and the canon reconsidered.
- Engagement Quality for the current lines falls for two consecutive readouts.
What it needs
| Input | Required | From |
|---|---|---|
| Current canon positioning and version history | Yes | Positioning Canon |
| Approved ICP and persona priorities | Yes | ICP Library |
| Persona objections and proof preferences | Yes | Buyer Personas |
| Proof records with ProofIDs | Yes | Proof Vault |
| Narrative white space findings | Optional | Differentiation Analyst |
| Past subhead test results | Optional | Message Memory |
How it works
- Load the current canon version and fill a six-variable table (Category, Problem, Audience, Benefit, Proof, Differentiator); mark any variable with no evidence record as a gap.
- Generate three stage-aware lines: awareness names the problem in buyer words, consideration states the benefit against the alternative, decision states the outcome with a ProofID.
- Run the positioning-validation skill: put each line through at least two synthetic personas (for example a skeptical CTO and a time-poor controller) and record every objection as hypothesis.
- Revise lines that drew 'vague' or 'unproven' objections; drop any decision line whose proof cannot be matched to an unexpired ProofID.
- Design one subhead test: two subheads on one surface, one variable changed, sample size and stop date written down before launch.
- When a test closes, pull the readout with `python3 scripts/sns.py metrics engagement-quality`, log winner and losing-angle lesson in message-memory, and apply the promotion rule.
- Write the positioning-canvas as a new proposed version with a diff against canon; request human approval before anything is written to positioning-canon.
What it delivers
Positioning canvas: six-variable table with evidence refs, three stage-aware lines, synthetic-audience objections, test plan or result, version diff, promotion recommendation.
Output contract: positioning-canvas
Quality bar
- All six variables are resolved or explicitly flagged as gaps; no line is generated over an unresolved Problem or Audience.
- Decision-stage lines carry a ProofID; awareness lines carry no unprovable superlatives.
- Synthetic audience feedback is labelled hypothesis and never cited as buyer evidence.
- Positioning-canon changes only after a human_decision record names the approver and date.
Hands off
- Messaging StrategistA new canon version is approved and the narrative and kernels must be realigned.
- A/B Experiment DesignerA subhead test needs a properly powered experiment on a high-traffic page.
- Value Proposition ArchitectThe Benefit or Proof variable changed for a specific persona.
- AI Marketing Chief of StaffA canon change is proposed and needs routing to the human approver.
When a person steps in
- Never write a version to positioning-canon without a human approval record; propose and wait.
- If the Proof variable has no unexpired ProofID, block decision-stage lines and escalate the proof gap to the owner.
- If two subhead tests in a row are inconclusive, stop testing that surface and ask whether traffic is sufficient.
Example
Request
“Refresh our positioning now that the mid-market finance ICP is approved.”
Result
Illustrative output: canvas v4 with all six variables resolved; decision line cites PRF-0017 (close time cut from 9 to 4 days, named customer). Skeptical-CTO persona flagged 'real-time' as vague, replaced with 'posted within 15 minutes'. Subhead test SUB-07 planned on the pricing page (2 variants, 1,800 sessions each, 21 days). v4 proposed, awaiting approval.
Illustrative example from the package documentation, not a client result.