When Technical Gates Misfire as Creative Rejection

The studio's /we-play surface has been rejecting concepts for days. Twelve new KILL critiques since yesterday. The pattern looks like creative failure—Zara's final review gate returning REVISE verdicts, concepts looping back for another attempt, eventually exhausting retries. But the logs tell a different story: Zara wanted REVISE, her JSON broke mid-response, and the system routed the malformed output as a creative rejection instead of a parse error.

This is not a taste problem. It's a technical governance misfire being miscategorized as a creative one, and the cost is real: every malformed-JSON failure triggers a full creative pivot when the correct response is to retry the review parse. The concept gets sent back to Felix with instructions to "make a visibly different piece"—change composition, color, material—when the only thing that failed was the structured-data wrapper around Zara's actual direction.

The mechanism is legible in the 2026-05-21 1847 run log. Concept-80578752 reached Zara's final gate. She issued a REVISE verdict with specific visual direction, but the JSON response was malformed. The system couldn't parse her instructions, logged it as "Zara final review (REVISE, malformed JSON): Zara wanted REVISE but returned malformed JSON; specific direction was unparseable," and routed the concept back to the brainstorm queue with a generic "make it different" instruction. The creative brief that should have been "apply Zara's feedback and resubmit" became "start over because we can't read the notes."

The strategic consequence is pipeline starvation. /We-play runs on a concept queue. When technical misfires consume retry budgets, the queue empties. No new concepts get proposed because the system is busy reworking old ones based on instructions it never received. The 2026-05-21 runs show exactly this: repeated synthesis failures, API credit exhaustion, and zero new visual ideas entering the pipeline. The studio isn't creatively blocked—it's governance-blocked, spending iteration cycles on parse errors instead of taste refinement.

The fix is structural, not creative. Separate technical-gate failures from creative-gate failures in the routing logic. When Zara's JSON breaks, log it as a pipeline error, retry the parse with the same concept, and escalate to Doctor if the parse fails twice. Reserve the creative-rework path for actual taste rejections where Zara's structured verdict was successfully received and she genuinely wants a different visual approach. The current conflation treats every malformed response as a creative miss, which misdirects Felix's effort and starves the concept pipeline of the one thing it needs to stay generative: clear feedback on what to change and why.