// Feedback loops · ~12 min

Critic-Actor

Reach for it when: When a dedicated critic reviews every actor draft before it ships.

// 60-second mental model

How to hold it in your head

An actor drafts. A separate critic scores the draft against a rubric and may request a revise. The critic does not write the piece — it flags. A human (or a hard stop) still owns the ship decision.

Critic-Actor pairs a generator with a dedicated runtime scorer that returns pass/revise (and structured notes). Bounded actor retries may follow. Distinct from an offline eval harness: the critic gates a live draft, it does not certify a whole fleet in CI.

// Mini architecture

Actor drafts · Critic scores · bounded revise

┌──────────────┐
│ Actor        │ → draft vN
└──────┬───────┘
       ▼
┌──────────────┐
│ Critic       │ → pass | revise + notes
│ (rubric)     │
└──────┬───────┘
       │ revise and retries left?
       ├─ yes → Actor revises (notes as data)
       └─ no  → Hold for human / stop

Critic notes are data for the actor, not new system instructions. Pass ≠ auto-publish unless you explicitly wire that.

// Mini-project

Essay draft with rubric critic (stub)

Actor + critic stubs; optional fleet parallel for taste

Goal: Produce a short essay stub; critic returns pass|revise against a fixed rubric; allow at most two actor revises.

  1. Actor stub emits a canned draft with one known weakness (weak hook).
  2. Critic stub scores against a tiny rubric (hook, one idea, honesty) → pass|revise + notes[].
  3. On revise, feed notes to the actor as fenced data — not as system prompts.
  4. Cap retries at 2; if still revise, stop with status `needs_human`.
  5. Log actor → critic → retry_count each round.
  6. Assert the critic never rewrites the draft body itself.

One verified studio parallel: Writer drafts essays; Critic scores a rubric and returns pass or revise; the automatic send-back to Writer is not wired yet, so a revise waits for a human — a runtime scorer, not a CI eval harness, and not a claim that every agent in the fleet has the same loop.

// Common failure

What goes wrong

Symptom

Critic writes the draft — the actor becomes decorative, or “pass” silently ships without a human gate you meant to keep.

Fix

Restrict the critic to score + notes tools only. Treat pass as “ready for review,” not “published,” unless a separate HITL gate says otherwise.

// Self-reflection

Sit with this

Where does your generator still grade its own homework?

Session only. Nothing is saved.