Cultivate

Epic P1

Outcome measurement and evaluation

The product layer defines useful outcomes and tests changes against evidence that can falsify weak ideas.

Purpose

Agents can separate model scores, selected corrections and live product outcomes instead of treating all evidence as equal.

Current status

Next. Slop Filter has correction records and synthetic cases, but the first protected product baseline still needs approval.

Milestones

Next

Outcome and baseline are written down

The first product experiment needs agreed outcome measures, representative-enough examples and a held-out evaluation set.

Depends on: E3-M1

Proposed

Evaluation boundaries are explicit

Selected corrections, synthetic cases and model percentages are labelled as limited evidence rather than representative accuracy.

Depends on: P1-M1

Evidence

Verified

Existing narrative and capability roadmap Google Doc

Read through an authorised Google Drive connector and restructured into site pages rather than copied wholesale. The source document link is not published.

Verified

LinkedIn Slop Filter main at 8984c80

Local main and GitHub main matched after cleanup. README, AGENTS.md and recent pull requests were inspected.

Open source

Risks and open questions

Risks

  • Hand-selected corrections can overstate performance.
  • A classifier improvement may not improve what the owner reads.

Open questions

  • What is representative enough for a one-person LinkedIn product test?
  • How should held-out examples be stored without exposing private post text?

Linked Wardley components

  • User outcomes
  • Product experiments
  • Independent evaluation
  • Testing frameworks

Next steps

  • Create a private evaluation charter and baseline before running competing teams.