Should rejected attempts be deleted?
No. Preserve them with failure reasons when policy and storage rules permit. Removing failures makes the test impossible to audit and can exaggerate apparent consistency.
Tool / Evaluation planner / task entry
Video Repeatability Test Generator
Video Repeatability Test Generator prepares fixed inputs for an inspectable trial packet.
Prepared workflow
From brief to reviewable handoff.
Prepare a video repeatability test with frozen inputs, invariant checks, attempt budgets, rejection codes, reviewer ownership, and a bounded workspace handoff.
- 01
Build the frozen test packet
Collect the exact prompt, authorized reference files and hashes, aspect ratio, duration target, visible model selection, and any exposed control values. Write a short intent statement explaining the shot's job. Store the packet under a test identifier so each attempt can be traced to the same inputs rather than reconstructed later from memory or a revised brief.
- 02
Score invariants without hiding judgment
Create separate fields for subject identity, object geometry, action order, camera intent, continuity, text, dialogue, sound timing, factual accuracy, rights, and delivery fit. Use observable notes beside pass labels. Assign a reviewer and escalation rule for borderline clips; an unexplained average can conceal a critical failure that would block actual publication.
- Attempt identifier and timestamp
- Criterion-level observation
- Failure code and severity
- Reviewer and disposition
- 03
Close the test with a bounded decision
Summarize the distribution of failures and repairs, not only the selected output. State the prompt, configuration, attempt count, date, and acceptance rule beside the conclusion. Keep a retest trigger for model, interface, policy, or delivery changes. The planner supports evidence collection; it does not calculate or certify a product-wide success guarantee.
Capability boundary
Dated evidence
Sources behind the current facts.
Product and model details can change. These links identify the evidence checked for the claims scoped below.
- X / user-supplied trend batch10 out of 10 repeatability representative X trend post
Representative post URL for the supplied trend signal only; it is not product, model, quality, repeatability, customer, or commercial-result evidence.
Reader notes
Before you hand off.
What does the workspace handoff prove?
It proves only that a prepared brief was handed into the currently visible workflow. Availability, model behavior, and the resulting observations still require direct verification and logging.
Adjacent task planners
Video Restyle Generator
Evaluate “video restyle” with a workflow learning checklist for inputs, controls, evidence limits, human review, rights, and a safe production handoff.
Open field noteVideo Reframe AI Generator
Evaluate Video Reframe AI for video Review focus: test worksheet, fact checker, action readability, and delivery package; keep reframe evidence separate from video assumptions.
Open field noteVideo Similar Generator
Evaluate “video similar” with a workflow learning checklist for inputs, controls, evidence limits, human review, rights, and a safe production handoff.
Open field note10 out of 10 repeatability is a claim to test, not a rate to promise.
Interpret 10 out of 10 repeatability as a testable video-evaluation claim, with fixed inputs, attempt logs, failure categories, sample disclosure, and no guaranteed rate.
Open field noteContinue the workflow
Take a prepared brief into the workspace.
Open the current SEELE Film & CG workspace to verify relevant controls and availability before applying this prepare a repeatability test scorecard workflow.