Automatic Highlights Generator

Define the workflow learning job

Treat “automatic highlights” as a search job to investigate, not as proof that a SEELE feature exists. First turn the query into a repeatable sequence with explicit inputs, review points, and a reversible handoff. Write the intended audience, source owner, desired change, protected details, reviewer, and delivery condition before selecting any interface or model. That brief keeps the evaluation specific and makes an unsupported assumption visible early. While evidence is current, preserve the brief version, and ask the policy reviewer to record failure conditions before the bounded test.

Prepare inputs for clips, shorts, and highlights

For this topic, assemble the authorized long-form source, target channel, required context, duration range, and caption needs. Record where each source came from, who may use it, and what must remain unchanged. Use a small representative asset for the first pass, keep the original untouched, and define a fallback route so experimentation cannot silently become the production master. While evidence is current, preserve the failure note, and ask the claims reviewer to record control availability before the editorial approval.

• Confirm ownership, consent, and allowed reuse Before moving on, preserve the input snapshot, and ask the continuity editor to record visible continuity before the bounded test.
• Preserve an untouched source and version history Before moving on, preserve the control log, and ask the source custodian to record revision intent before the bounded test.
• Name the reviewer and acceptance condition Before moving on, preserve the decision history, and ask the model evaluator to record source fidelity before the bounded test.

Test observable controls for automatic highlights

A bounded evaluation should inspect moment selection, context preservation, reframing, pacing, captions, and the final call to action. Change one meaningful variable at a time and record the date, workspace, account context, input, setting, result, and failure. Topic selection can prioritize the question, but it does not establish availability, quality, speed, licensing, or a supported SEELE workflow. Before moving on, preserve the source ledger, and ask the source custodian to record control availability before the dated decision.

Review evidence, safety, and policy boundaries

A compelling excerpt must not distort the speaker's meaning or imply automatic publishing or platform approval. Use only authorized media, separate observed behavior from marketing language, and check current first-party documentation for any product-specific claim. For third-party products, competitors, plans, models, and platform rules, attach a verification date and primary source; an absent statement is an evidence gap rather than proof of a limitation. In the decision log, preserve the source ledger, and ask the brand reviewer to record format readiness before the reversible handoff.

Approve a reversible production handoff

Before advancing “automatic highlights”, compare each excerpt with its source, check that edits preserve meaning, and validate current channel specifications. Document remaining manual work, unresolved evidence, destination requirements, and the person accepting the result. The handoff should preserve sources and test notes, allow correction, and avoid promises about output quality, turnaround, business performance, publishing, or access that the evidence does not support. At this stage, preserve the source ledger, and ask the delivery owner to record control availability before the acceptance review.