The Strategist

The Strategist is the control room. ReviewHub is where taste goes in; the Strategist is where you steer the machine and watch what your taste did to it. It has seven working areas, and each one answers a specific question.

Overview — "is it learning?"

The front page. Grade distribution across everything ever graded, the outlier and catalyst rates, output diversity (how many of the recent looks were unique compositions), and the learning crossover — how many learnings apply across the whole roster versus being scoped to one designer or colorway. The cards deep-link into the Lab and Learning tabs.

Control — "what does it make, and when?"

The entire daily run on one page, written as sentences you edit in place: when it fires and in which timezone, how many looks per day and the budget that stops it, how aggressively it innovates, which models compose and render, the frame and per-look render count, and who gets notified. Below that sits the roster — every designer, colorway, garment, and template the engine can draw from, each with a "1 in N outfits" frequency dial, usage stats, and retire/reactivate controls. Every change saves as you make it; the full knob list is in the Control reference.

Lab — "how are winners evolving?"

The evolution arm made visible. The seed queue shows every graded winner waiting for the Lab. Evolution chains trace each seed to the experiments it spawned and the grades those descendants earned. The experiments feed shows each experiment explaining its own reasoning. The Lab's daily quota lives here too, with a "Run lab now" button.

Learning — "what does it know?"

The brain, browsable. The eight innovation axes it rotates through, the proven directions it grows from, the live anti-repetition fence (the exact phrases the next look is forbidden to reuse, with reuse counts), the novelty score of the latest looks, and the full register of learnings — thousands of notes, filterable by designer / colorway / garment and by guidance vs bans.

Ideas — "make something now"

Fire a manual batch with a chosen size, frame, and renders per look.

Runs — "what happened?"

Every batch the engine has run. Click a row and it opens into that run's looks — thumbnails with their grades — headed by the run's innovation report card: novelty versus prior work, whether the scene stayed locked, and whether any banned language slipped through.

Status — "is anything wrong?"

Engine health checks, plus the pipeline failure panel: every stage's ok/fail counts and the most recent failure reasons, refreshing every 30 seconds. If a stage starts failing repeatedly, the sidebar health badge goes red on its own — see Observability.