Creative Intelligence Operating System
The Critiqa Playbook
Use Critiqa as one connected learning loop: evidence → strategy → creative judgment → brief → production → review → experiment → performance → learning.
An illustrative creator workflow, not a customer result
From a weak opening to a testable next post
Imagine a fictional creator with a short product video that begins too slowly. First, review the existing creative and record the observable opening issue separately from AI interpretation. Next, write a brief that changes only the hook, keeping the product explanation and offer consistent. Create candidate versions in Creative Studio, check the resulting details and claims, then plan an evidence-aware comparison. Without observed results, neither variation can be called a proven winner.
01
The mental model
Critiqa is one creative learning loop: evidence → strategy → creative judgment → brief → production → review → experiment → performance → learning. A user can enter at any stage, but the system is strongest when context travels through the loop.
02
Start with brand or project context
For recurring work, select the relevant Client and Project so research, briefs, assets, experiments, performance evidence, and learning stay scoped correctly. For one off analysis or Studio work, do not force unnecessary workspace setup.
03
Build evidence before ideas
Use Customer Voice, verified research, Sidekick, Customer Avatars, Website Intelligence, Rival Monitor, Competitor Intelligence, existing analyses, connected performance data, and approved BrandLearning. Label what was observed, what was inferred, and what is only proposed.
04
Analyze an existing creative
Use Analyze when an image or video already exists. Read the overall diagnosis, 16 dimension performance fingerprint, evidence and limitations, Jury disagreement where entitled, Viewer Journey for time based video diagnosis where entitled, and the recommended next action. Do not use the overall score as universal truth.
05
Turn analysis into a creative route
Creative Forge converts diagnosis into executable routes. A useful route contains the hypothesis, target audience or persona context, message or angle, hook logic, format, evidence basis, test intent, and success measure.
06
Build the brief
Use Brief Builder when production needs a decision document. The brief should connect Context → Hypothesis → Audience → Test → Assets → Creative inspiration → Production direction. It should explain why the creative exists and what decision the content is meant to settle.
07
Produce in Creative Studio
Studio turns approved direction into image, video, or UGC assets. Confirm the source strategy or brief, reference assets, brand and product identity constraints, output type and ratio, what may change, what must remain stable, and the credit cost. Every intentional AI generation is billable once. Provider retries reconcile the same job; a genuine regeneration after edits is a new job.
08
Review before you publish
Use Review Assets and approval flows to separate generation from approval. Check grounding and claims, brand fit, creative quality, integrity, mobile readability, whether the output still answers the brief, and whether the intended test variable remained isolated. Do not publish material compliance issues, unsupported claims, a broken hook, an unclear message, or work that fails the experiment question.
09
Use Creative Multiplier for controlled descendants
Start from a source creative, extract Creative DNA, then declare what is LOCKED and what may CHANGE. Prefer one major variable at a time when causal learning matters. Refresh fights fatigue, Experiment varies a controlled dimension, Localize adapts market details while preserving the mechanism, and Propagate carries the winning idea into a different execution. Descendants preserve parent lineage and execution intelligence.
10
Build a first class Experiment
An Experiment records the hypothesis, variable under test, constants, control asset, variants, primary metric, guardrails, success and kill criteria, minimum evidence, review condition, result, decision, and learning linkage. The canonical lifecycle protects the test from being rewritten after evidence arrives.
11
Read performance as evidence, not certainty
Connected evidence can include spend, impressions, clicks, conversions, CTR, hook rate, and hold rate. Broader business context can include conversion rate, revenue, retention, and related metrics when valid sources exist. Judge metrics against the creative role rather than applying one universal performance rule.
12
Complete the learning loop
After an experiment, record the result, winning asset if one exists, decision, evidence strength, what changed, what stayed constant, where the learning applies, and what should be tested next. Completed experiments can create BrandLearning and the Creative Intelligence Graph preserves the relationships behind the result.
13
Winner Amplification V2 status
Winner Amplification V2 is live in production. It formalizes dynamic per variant change records, audience hypotheses, conservative attribution, deterministic decisions, winner extension, experiment tree lineage, and a human approval boundary before billable generation.
14
Agency operating pattern
Keep each Client context separate. Store verified research under the correct client. Use Project Workspace for active project context. Review assets before client approval or publication. Keep BrandLearning scoped to the brand and project that produced the evidence. Never turn one client result into another client rule without explicit evidence.
15
Action Inbox
Action Inbox is the daily operational layer for real states that need attention, such as pending review or other deterministic workflow conditions. It is not a generic task list populated by an LLM.
16
Billing discipline
Manual organization and deterministic workspace state changes remain free. Intentional AI analysis, generation, regeneration, strategy, brief, rewrite, branching, and other metered AI jobs are billable according to configuration. Never create free bulk generation loopholes, duplicate retry charges, or hidden regeneration after upstream edits.
17
Product availability and approval
Check the active application for which actions your plan supports, what each AI action costs and which sources are connected. A planned feature is not a delivered feature. Approval remains necessary before paid generation or publication.
Start where the evidence is.
You do not need to use every Critiqa surface. Start with the current decision, preserve the context, and carry the result into the next controlled step.
Start free