Collections Agent | SPL RightNow

1. The basics

Collections Agent turns a plain-language request into a collection of time-coded clips from the archive. You describe what you want. The agent finds the matching moments in the annotation data, checks them, and builds the collection. You then approve or reject it.

A good request tells the agent three things: which matches, which moments, and how many clips. Anything else you add (order, extras, exclusions) makes the result more useful.

Tip: You do not need special words or formats. Write the request as you would brief a colleague. Use the names and match-week codes that you see in the archive.

2. How the agent builds a collection

The agent draws on three sources of annotation data. You do not need to name them in your request, but understanding them helps you ask for the right things.

Source	What it tells the agent	What it is good for
Opta Highlights	What happened: the event (goal, save, card and so on), the player, the team, the minute, and a pre-built clip window.	Finding goals, saves, cards and shots by player, team or minute.
CobraHighlight	Editorial in and out points for those moments, often with several versions of the same moment.	Setting clean start and end points for each clip.
CobraAnnotation	What the footage shows: build-up, celebrations, reactions, replays, set pieces, and scorelines.	Adding context before and after a moment, finding set pieces, and answering questions about what happened on the pitch.

Every clip is a time-coded sub-clip taken from the match footage. The agent does not add a fixed number of seconds before or after a moment. It uses the editorial points where they exist, and the clip window from the event data where they do not.

3. A simple recipe for a prompt

Most good prompts answer four questions. You do not need to write them as a list, but covering each one helps the agent get it right the first time.

Which match or matches? Name the teams, the match, or the match-week code (for example, MD12).
Which moments? Name the events and players, such as goals by a player, set-piece goals, or saves.
How many clips? Give a number or a range, such as 8 to 10 clips.
Anything extra? Say whether to include the build-up or celebration, the order, what to leave out, and any collection name you want.

Template

Copy and fill inCreate a collection of [moment type] from [match(es) or match-week]. Players or teams: [names, or "all"] Events: [for example goals and penalties, or "not sure, please ask"] Volume: [for example 10 to 12 clips] Extras: [for example include the build-up and celebration, in match order] Leave out: [for example penalties, or nothing] Name: [optional collection name]

4. What the agent will ask you

The agent asks questions only when the answer would change the collection, and it tries to find answers in the data first. When it does ask, you will usually see a short list of options to tap. Here are the most common questions.

Question you may see	When it appears	How to answer
Which events should I include?	You asked for a "highlights package" or "key moments" without naming the events.	Choose an option, such as "Goals and penalties only", or name the events you want.
Should I include substitutions, off-target shots or other extra events?	The agent is not sure whether these count as meaningful for your request.	Say "yes" to include them, or "no" to keep the package focused.
Which match did you mean?	The team names or match-week code match more than one match.	Give the match-week code, the date, or both team names.
Which player did you mean?	A name matches more than one player, or it is spelled differently in the archive.	Use the full name as it appears in the archive, or choose from the options shown.
Should I include the build-up or celebrations?	Your request mentions a clip style but not the context around the moment.	Say "with build-up", "with celebration", or "moment only".
Do you want the full match footage?	Only if you ask for the whole match. The agent does not add this on its own.	Say "full match" explicitly. Otherwise the agent creates clips only.

What the agent will not ask: whether to create the collection. It builds the collection and then shows it to you for approval.

5. Examples of good prompts

These examples show requests that work well. Replace the placeholders with your own match, player and event details.

Goals by one player in one match

PromptCreate a collection of goals scored by [player name] in [team A v team B], [date or match-week code]. Include the build-up and the celebration. Order them by minute.

Why it worksIt names the player, the match, the event and the extras. The agent does not need to ask anything.

What to expect: One sub-clip per goal, with the build-up included and the celebration after each goal, in match order.

Highlights package for a match

PromptCreate a highlight package from [team A v team B], [match-week code]. Goals, penalties, saves and cards. Between 10 and 12 clips. Order by match time.

Why it worksIt lists the event types and the volume. This avoids a question about which events to include.

What to expect: A chronological package with one clip per meaningful moment. If there are more than 12 meaningful moments, the agent will say which ones it left out and why.

Set-piece goals across a matchday

PromptCreate a collection of set-piece goals (free kicks, penalties and corners) from [match-week codes, for example MD10 to MD12]. About 12 clips. Include the build-up to each set piece and the celebration. Exclude penalty shootouts. Group the clips by match.

Why it worksIt defines the moment type, the scope, the volume, the extras and the grouping.

What to expect: Clips grouped by match, with the set-piece type shown in the description. The agent uses the scene information to identify set pieces, so these clips may be found through the footage descriptions rather than the event data alone.

Defensive moments

PromptCreate a collection of defensive blocks and last-ditch clearances from [team A v team B], [date]. 8 to 10 clips. Show the opposing attack, the defensive action, and the recovery of possession.

Why it worksIt describes the sequence you want to see, not only the event name.

What to expect: The event data may not tag every block or clearance. If it does not, the agent will search the footage descriptions and label these clips as content-based, or it will ask you whether to continue with what it found.

Score-related moments

PromptCreate a collection of the goals that changed the result in [team A v team B], [date]. Include the equaliser and any winning goal. Include the celebration.

Why it worksIt describes the story, not a score line the agent would have to calculate.

What to expect: The agent works out the score at each goal from the match footage descriptions. If it cannot confirm the score for a goal, it says so rather than guessing.

Player moments across several matches

PromptCreate a collection of [player name]'s goals and key saves or shots on target across [match-week codes, for example MD10 to MD12]. Around 10 clips. Group by match, then in match order.

Why it worksIt sets the scope across several matches and asks for grouping, so you get a per-match breakdown.

What to expect: A summary per match, such as "Match 1: 2 goals, 1 shot on target". Clips from every match are in one collection.

6. When things go wrong

Most issues come from the request rather than the archive. The table below lists common symptoms and what to do.

What you see	Likely cause	What to do
No results	A player name, team name or match-week code does not match the archive.	Use the spelling and code shown in the archive, or ask the agent to list the matches for that week first.
Fewer clips than you asked for	There are fewer meaningful moments than requested, or some moments have no usable footage.	Read the list of excluded moments and the reasons. Widen the scope (more matches or more event types) or accept the smaller collection.
Some moments were excluded	The footage has no usable frame range or media reference, or the moment is a duplicate.	Check the exclusion reasons. Duplicates are normal when the same match is stored more than once.
Clips show the wrong player	The request used a nickname or a shared surname, or the penalty record names the player who conceded.	Use the full name as it appears in the archive. The agent uses the event data to decide who did what.
Clips are very short or cut off	The editorial points were tight, and no build-up or celebration was requested.	Ask again and add "include the build-up" or "include the celebration".
Test or colour-bar footage appears in the scope	Test patterns sometimes appear at the start or end of a file.	The agent should exclude these automatically. If one gets through, report the collection name and the clip so the team can review it.
The agent could not work out a score	The footage descriptions do not state the score at that point.	Ask for the goals by minute instead, or accept the moments without the score.
The agent asked too many questions	The request was broad or left out key details.	Answer once with the four details from the recipe (match, moments, volume, extras). The agent should not need to ask again.
The package has more moments than you wanted	"Highlights" includes all meaningful events by default.	Name the events you want, or ask the agent to narrow it to the top moments by significance.

7. Adjusting your prompt

If a result is not what you wanted, change the request rather than repeating it. Here are some common adjustments.

First attempt	Problem	Better version
"Some Saudi League goals."	No match, no volume, no event detail. The agent has to guess the scope.	"Create a collection of goals from [team A v team B], [match-week code]. About 6 clips, in match order."
"Ronaldo highlights."	Too broad. The agent may ask which matches and which events.	"Create a collection of [player name]'s goals and key saves in [match-week codes]. 10 clips, grouped by match, with the celebration included."
"Make the clips 2 to 3 seconds before and after each goal."	The agent does not use fixed padding. It uses the editorial points, so fixed padding is not applied.	"Include the build-up and the celebration for each goal."
"Exclude test footage."	Not needed. The agent already excludes test and colour-bar footage.	Leave it out, or name the things you do want excluded, such as penalties or bench reactions.
"Add social captions."	Captions are only added when you ask for them, and they are not part of the default output.	"Add a short social caption to each clip, under 120 characters." Be specific about the tone and length if you need it.

Checklist before you send

Have I named the match, team or match-week code?
Have I named the events, or said "not sure"?
Have I given a number or range of clips?
Have I said whether I want build-up, celebration, or the moment only?
Have I said how to order the clips?
Have I listed anything to leave out?

8. What the agent will not do

It will not invent footage. If a moment has no usable annotation, the agent reports it as missing rather than substituting a full match.
It will not use whole-match footage unless you ask for it. Say "full match" if you need it.
It will not add fixed seconds of padding. Clip boundaries come from the editorial points and clip windows in the data.
It will not add social copy unless you ask.
It will not show internal system details. If you ask how it works behind the scenes, it will describe the editorial process only.
It will not confirm before creating a collection. You approve or reject the collection afterwards.

9. Glossary

Term	Meaning
Collection	A set of clips created from one request. It can cover one match or several.
Clip (sub-clip)	A time-coded segment of match footage with a start and an end point.
Record	The archive entry for a match or programme. Each record holds the match-level details, such as teams and match week.
Match-week code	The code for a round of fixtures, for example MD12. Use the code shown in the archive.
Clip window	The start and end points that the event data suggests for a moment. It includes some footage before and after the event.
Editorial points	The in and out points set by editors for a moment. Where they exist, they are preferred.
Build-up	The play leading to a moment, such as the passes before a goal.
Celebration	The reaction after a goal or key moment.
Set piece	A restart such as a free kick, corner or penalty.
Content-based clip	A clip found from the description of what happens on screen, rather than from a recorded event.
Meaningful event	An event that counts toward a highlight package by default. Core events include goals, shots on target, saves, yellow cards and red cards. Substitutions and shots off target are included only when requested or confirmed.