Feedback Vector Loop

Feedback Vector Loop

Use this skill to operationalize directional feedback:

• not just "good/bad"
• but "move output toward X and away from Y"

This pattern is reusable across brands (e.g., Jacksnaps, Digipop, any creative pipeline).

When to use

• You have candidate outputs and human critique.
• Binary labels are too weak for improvement.
• You want iterative quality gains while preserving deterministic control.

Core model

Each feedback record should include:

1. Verdict: up | revise | down
2. Direction:
   • more[]
   • less[]
   • avoid[]
   • replace{old:new}
3. Optional candidate context (title, slogan, rationale, etc.)

Deterministic pipeline

1. Capture feedback log (feedback_log.jsonl).
2. Compile vector profile (feedback_vector.json).
3. Rank new candidates with base score + vector score.
4. Apply replacement rules and penalties/bonuses.
5. Emit stage trace and artifacts for auditability.

Runbook

1) Compile vector from feedback

python scripts/compile_feedback_vector.py \
  --input <feedback_log.jsonl> \
  --output <feedback_vector.json>

2) Rank candidates with vector

python scripts/rank_candidates_with_vector.py \
  --candidates <candidates.json> \
  --vector <feedback_vector.json> \
  --output <ranked_candidates.json>

3) Validate data contract

Use references/feedback-data-contract.md for required fields.

Design invariants

• Keep feedback rows append-only.
• Use weighted token adjustments, not opaque hidden state.
• Persist artifacts every run (traceability).
• Separate creative generation from deterministic scoring.

Suggested artifact set per run

• feedback_log.jsonl
• feedback_vector.json
• ranked_candidates.json
• stage_trace.jsonl
• summary.json