Extension docs

Equipe Startlist AI

Startlist AI helps show organizers turn rider PM notes into structured start-list wishes, test those wishes against Equipe's prepared start slots, and export a proposed order back to Equipe.

What The Extension Does

Imports Equipe data

Pulls competitions, starts, riders, horses, horse numbers, and PM notes from the current meeting.

Parses rider wishes

Uses AI to extract wishes such as late starts, near another rider, before or after a break, and spacing or order between a rider's own horses.

Evaluates the order

Moves starts within the existing imported slots and shows which wishes are honored, unhonored, or unenforceable.

Exports changes

Writes the proposed position, start number, and time back to Equipe when those values are available. After a successful export, fulfilled PM todos can also be marked completed in Equipe. Export stops if affected competitions contain ridden or ranked starts.

Before You Start In Equipe

The extension works from Equipe's existing start slots. Prepare the meeting in Equipe first:

1. Draw every competition. The extension does not create a start list from scratch.
2. Add planned breaks. Breaks should already exist in the Equipe schedule before import.
3. Run Equipe's automatic start-time conflict resolver. Use the magic wand in Equipe so the start slots and estimated times are ready.
4. Verify the order and estimated times. Startlist AI moves starts between the imported slots; it does not recalculate the whole day plan.
5. Make sure PM notes are entered. Rider PM notes and horse PM notes are imported and parsed as the source wishes.

Workflow

Step 1

Import

Pull the prepared meeting data from Equipe.

Step 2

Parse wishes

Use AI to turn PM notes into structured wishes.

Step 3

Evaluate wishes

Move starts within existing slots and check wish status.

Step 4

Export order

Write the proposed order back to Equipe and complete fully resolved PM todos.

Parse Throughput And Rate Limits

Wish parsing fans out to one background job per PM note. The app limits concurrent OpenAI parse requests per organizer/API key, so parsing several meetings at the same time does not multiply traffic past the same organizer's OpenAI rate limit.

Parse concurrency: PARSE_CONCURRENCY=2
Job threads: JOB_THREADS=10
Job processes: JOB_CONCURRENCY=1

Effective parallel parsing is bounded by the lowest of parse concurrency and available job worker capacity. If OpenAI returns rate-limit errors, the per-note job is rescheduled with backoff instead of continuing to occupy a worker thread, and the meeting page shows an OpenAI rate/spend limit warning while parsing waits. Sustained rate-limit errors mean the concurrency should be lowered or the OpenAI project limit, billing, or spend limit should be raised.

Supported Wish Types

The AI parser turns PM text into structured wishes. Only wishes with real Equipe ids for the rider, horse, competition, and target rider can be acted on by the resolver.

PM notes can be written in Swedish, Dutch, French, or English. Politeness wording and abbreviations — such as Dutch “a.u.b.”, “aub”, “ajb”, “graag”, or “gelieve” and French “s.v.p.” or “svp” for “please”, and greetings or sign-offs like “alvast bedankt” and “mvg” — are understood as politeness, not as wish content. They do not need to be removed or rewritten before parsing, and they never block a note from being parsed.

Placement

Requests to start in a part of a competition, in a numeric range, or before/after a break. “Last”, “late”, “at the end”, and similar wording is treated as the back of the competition. Placement wishes must be bound to one horse in one competition.

Zone: front, middle, or back
Range: min and max start positions
Break: before or after, optional offset range

Starting Number

Requests for an exact start number, around a number, or within a start-number range. A class number, such as “competition 1”, is not treated as a requested start number. Neither is a time of day, such as “around 9.00”, or a horse label, such as “Paard 2”, which riders use to point out which of their own horses a line is about. If a PM note says a horse is “crazy”, the parser treats that as a hard request for that horse to start first in every competition where the horse appears. around means plus or minus one start number. These wishes must be bound to one horse in one competition.

Near Rider

Requests to start near, before, or after another rider. The parser must resolve the other rider with the rider lookup tool and store target_equipe_person_id; a plain name is not enough. The lookup ignores common accent differences, so a note can mention “Léa” even if the imported rider name is stored as “Lea”. Overall support/travel/coaching requests can be rider-level. Start-list wording inside one concrete horse/competition segment is competition-horse level. Family or group requests such as “the Van Es and De Swert families together” are treated as rider-level near-rider wishes: the names are looked up separately, every clear matching other rider becomes a target, and the current PM rider is not used as their own target.

Own Horse Order And Spacing

Requests about the rider's own horses, either as an explicit horse order (ordered_equipe_horse_ids) or as a time gap. A gap such as “30 starts between” is converted during parsing with the rule of thumb 1 start = about 2 minutes, so it becomes gap_minutes: 60. We store this as time instead of a start count because several arenas and competitions can run in parallel, so “20 starts” is easier and safer to evaluate as about 40 minutes between the rider's own starts. A gap can also be a range with gap_min_minutes and gap_max_minutes.

Info And Unparseable

Non-actionable text is stored as info or unparseable so it can still be reviewed, but the resolver marks it as unenforceable because no start-slot move can satisfy it. Use the Info / unparseable filter in the rider wishes view to focus on these notes. If the PM text is clear to the organizer but hard for the parser, use Edit for parsing to save a local parsing override and re-parse that note. This does not change the PM text in Equipe. If Equipe later imports a changed PM for the same note, the local override is kept and marked for review. An operator can also remove an extracted wish that should not be enforced; removed wishes are ignored by the resolver and completed-note export until they are restored.

How The Resolver Works

The resolver does not build a new start list. It keeps Equipe's imported start slots and moves movable starts inside each competition to satisfy as many wishes as possible. Only starts that a wish refers to are placed by the optimizer; every other start keeps its imported relative order and just slides a few slots to make room. A rider without a wish is never thrown to a completely different part of the class.

Movable Starts

Only normal starts move. Pause rows stay fixed and are excluded from moves and gap analysis. The proposed start receives the target slot's position, start number, time, estimated time, and timestamp.

Per-Competition Search

The resolver optimizes one competition at a time. It starts from the imported order, seeds direct placement/starting-number wishes, then tries improving swaps for up to 80 iterations.

Scoring

A hard wish is worth 10 points and a soft wish is worth 1 point, unless a wish has an explicit custom weight. Each moved start costs 0.001 points, so unnecessary moves are avoided.

Statuses

Honored means the proposed order satisfies the wish. Unhonored means the wish was actionable but the chosen order does not satisfy it. Unenforceable means required data is missing or the wish cannot be checked by moving start slots.

Resolver Thresholds

Front, middle, back

The resolver divides each competition into thirds using the start's center position: (position - 0.5) / total. Values up to 1/3 are front, values up to 2/3 are middle, and the rest are back.

Near rider

near is honored when the starts are at most 5 starts apart in the same competition. before and after require the correct side of the target rider; they do not currently require a maximum distance.

Breaks

Breaks are inferred inside each competition from longer gaps between imported start times. The resolver uses est_stamp when present, otherwise the displayed estimated/start time. A gap is treated as a break when it is at least the greater of median gap + 5 minutes and median gap × 2. If no break can be inferred, before/after-break wishes are unenforceable.

Riders with several horses

The resolver protects the spacing your Equipe preparation gave riders with several horses in the same competition. It will not propose an order that puts a rider's own starts closer than 6 starts apart — or closer than the imported list already had them, with a small allowance of 2 slots for starts that slide when other wishes are placed. Creating such a squeeze costs 2 points, so a soft wish is left unhonored rather than giving an uninvolved rider back-to-back rides. Riders whose own notes ask for exact start numbers, exact positions, or a specific gap between their horses are exempt — their wish decides instead.

Own-horse gaps

Own-horse gaps are checked in minutes using the proposed slots' displayed times, measured between the rider's starts in time order. gap_minutes is a minimum gap. gap_min_minutes and gap_max_minutes define an accepted range. If times are missing, the wish is unenforceable. When the source note says combinations or starts instead of minutes, the parser converts it with 1 start = about 2 minutes before the resolver sees it.

What Export Writes Back

Export writes the proposed start-list values back to Equipe. It writes the new position for every moved start, and also writes start time when it exists in the imported start slots. Start numbers are read from the competition's current startlist in Equipe at the moment you export: the number stays with the slot, so a start that moves into a slot takes over that slot's number and the list stays numbered in running order. This also works when the competition was numbered after the import. If a competition has no start numbers at all, none are written and you can number it in Equipe afterwards as usual.

Start order

Starts are moved into existing Equipe slots. The extension does not create new slots or rewrite the whole schedule. Export stops before writing if affected competitions contain ridden or ranked starts.

PM todo status

Equipe PM notes can use [] for an open todo and [x] for a completed todo. After the start order has been exported and verified, Startlist AI changes [] to [x] only when every parsed item in that rider or horse PM note is fully resolved.

PM notes are completed conservatively

A PM note is left unchanged if any actionable wish in that note is unhonored or unenforceable, if the parser marked part of the note as info or unparseable, if the note has no actionable wishes, or if the live PM text in Equipe has changed since the last import. Horse PM notes are also left unchanged unless every imported copy for that horse is fully resolved.

Privacy And Data Responsibility

The organizer chooses to use their own OpenAI API key. That means the organizer is responsible for deciding whether the meeting's PM notes and related start-list context may be sent to OpenAI for processing.

What is sent to OpenAI

When parsing wishes, the extension sends the PM note text, the current rider name, the rider's imported competitions and horses, and any needed lookup queries/results for riders, horses, or competitions.

What the organizer must check

Before parsing, the organizer should make sure they have the right permission/legal basis to process this data with OpenAI and that their OpenAI project has acceptable data retention and data sharing settings.

Retention settings matter

OpenAI data handling depends on the API endpoint, project settings, organization settings, and whether Zero Data Retention is enabled. Check OpenAI's current Data controls in the OpenAI platform and the organization's data sharing settings before using real rider data.

OpenAI API Key

Startlist AI needs an OpenAI API key to parse PM notes. The key is stored encrypted for the organizer and is used only when parsing rider wishes.

How to create the key

Go to platform.openai.com and sign in.
Open the API keys area for the project or organization that should pay for the usage.
Create a new secret key.
Use All permissions / full permissions for the key.
Paste the key into AI access in this extension.

Current parser model: gpt-5.4-mini
Permission level: All / full permissions

Keep the key private. Do not paste personal ChatGPT credentials here; this must be an OpenAI API key.

Important Limits

The AI parser can misunderstand ambiguous notes. Use the rider wishes view to inspect original text vs. extracted data. Individual PM notes can be parsed again from that view; the note card updates in place when the new result is ready.
The resolver only moves starts inside imported slots. It does not invent new slots or redesign the schedule.
Export is blocked when affected competitions contain ridden or ranked starts, so live competition data is not overwritten.