Sports & Fitness AI agent · Officiating and odds
Officiating & Odds-Adjustment AI Agent
Two agents, separately deployed and separately tested: one measures for match officials, the other prices in-play markets inside an operator's approved system, and a person decides in both.
- Typical delivery
- 4–6 weeks
- Deployment
- Your stack
- Human decides
- Two deployments
- After launch
- Agent Care
What this agent does
Two pipelines, and no join between them
In
- 01
Where the deployment is officiating, ingest venue tracking and the competition's own definition of the line or zone.
- 02
Where it is odds adjustment, ingest market state and the feeds the licensee has approved.
Reason
- 03
When an output is produced, carry the tolerance band its tier permits with it.
- 04
Where a market is new, it is filed and approved by the licensee, not the agent.
- 05
Where the rules give the official the final word, the output is advisory, not determinative.
Decide
- 06
When an automated decision affects one customer, they can contest it.
- 07
Where separation is claimed, it is a test case, not a promise.
Out
- 08
Retain the output, the band it carried, the version and who acted, inside its own deployment.
- 09
Execute write actions only inside the approval boundaries agreed during implementation.
Product statement
The agent measures or prices; a match official or a licensee's trader decides, and the competition's rules say which of them is bound.
Example workflow
One event, in one deployment
- 1Event input receivedVenue tracking, or the licensee's approved market feed
- 2Rules assembledThe competition's line or zone definition, or the operator's filed house rules
- 3Output proposedA measurement or a price move, its band and confidence
- 4Controls appliedTolerance-band checks, separation checks, filed-rule checks and the confidence threshold
No human action required
Stages 1 to 4 run unaided inside one deployment, and nothing is decided at any of them — the official's or the trader's lane opens at the gate.
Goes to the official or the trader on duty.
Arrives marked unresolved, with no value.
Official or trader
The output arrives with its tolerance band, its version and the confidence.
Accept · Override · Suspend the feed
Accepted — acted on by a person
- 6Match or market record updatedOnly where write access and approval policy allow it
- 7Outcome evaluatedOverride rate, band breaches, suspensions and what the person found
Overrides
Every override is counted in the evaluation.
What should not run autonomously
Human approval stays in control
Outside the boundary — human approval required
8 items
- Making the call the competition's laws give a person.
- Setting the line, the zone or what counts as out.
- Bringing a new wager type to market.
- Deciding whether a wager stands or is voided.
Automation boundary
Agent acts unaided
- ✓Produce the measurement or the price move, and the band it carries.
- ✓Run in one deployment, and test for a path to the other.
- ✓Apply the definitions or house rules already filed.
- ✓Hand it to the official or the trader on duty, and log who acted.
Any write happens inside the boundaries agreed at implementation, never ahead of a person.
- Deciding one customer's limits, offers or restrictions.
- Ruling that a match or a market was affected.
- Reporting suspicion to a regulator or a governing body.
- Changes to tolerance bands or approval boundaries.
Example output
One measurement, annotated
Everything the agent outputs carries the definition it was measured against.
Officiating output · single call
Illustrative example- Competition
- Tennis, live line calling
- Measured event
- Ball impact rebuilt from the calibrated camera set
- Output type
- Fitted path
- Definition used
- Governing-body line
- Confidence
- 91%
- Authority
- Chair umpire, determinative
As received
Taken from the officiating deployment — the trading deployment keeps its own, and a join is a tested case.
What was used
- Calibrated camera set
- The filed line definition
- Per-frame positions
Why it carries a band
It is a trajectory fitted to camera data, not a photograph of the impact.
What the score decides
Inside the configured band the agent returns unresolved, and the person calls it unassisted.
Action
AcceptOverrideSuspend the feedValue
Where AI adds value
The same four claims, placed at the point in the workflow where each one applies.
Where the value lands
Value 01 – 04
Every event
Inside one deployment
Work to the filed definition
Work from the definitions the governing body set, or the house rules the operator filed.
Two deployments, never one pipe
The officiating agent and the trading agent are built, deployed and run apart.
Send the close ones to a person
A call inside the band arrives unresolved rather than as a value, so attention goes where it is close.
Build an evidence trail
The measurement, its tolerance and the official or trader who acted on it stay on the record.
Integrations
Typical integrations
Five system groups connect to the same agent. Which of them are in scope is decided in discovery.
Officiating deployment
- Optical ball tracking
- Video review
- replay desk
Competition rules
- Laws, playing conditions
- Match and event records
Odds deployment
- Trading platform
- Approved market feeds
Agent
Two agents, deployed apart
- Reads its own side
- Proposes with a band
- Hands to a person
Compliance and integrity
- Regulator filings
- Integrity reporting
Observability & evaluation
- OpenTelemetry
- Langfuse
- Supported monitoring/evaluation sources
Integration availability depends on the client's existing systems and API access.
Agent controls
Six layers between the model and the call
The layers nest, and the two deployments do not share one. What escapes is named below.
L6 · Outermost — last line of defence
Inward → L1 · closest to the model
- L1 – L2
- Test whether an output may stand
- L3
- Puts the act in a person's hands
- L4 – L5
- Hold the measurement and the tolerance it carried
- L6
- Reverts to unassisted operation when signals degrade
How Nestack evaluates it
Evaluate the whole assist — not only the number at the end.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the call or the price that goes out
- E1Final-output evaluationDid the output sit inside the band its tier permits?
- E2Step-level evaluationDid the agent use the right definition, version and filed rules?
- E3Tool evaluationDid it read and write only its own deployment's systems?
- E4Confidence calibrationDo low-confidence outputs actually attract more overrides?
- E5Slice evaluationHow does performance change across competitions and market types?
- E6Business outcomeHow many outputs were overridden, and how many were suspended?
Floor — what the league or the licensee answers for
Failure modes
Where each failure originates in the agent
Seven failure modes, each at the stage it starts.
Agent lifecycle
Direction of processing →
01 · Retrieval
1 mode
- OD-03
Superseded definition
The zone or line definition is read from a stale version.
Stage gathers
Tracking or market state, and the definitions on file
02 · Reasoning
2 modes
- OD-04
Certainty overstated
A fitted trajectory is reported as though it were observed.
- OD-06
Unfiled rule applied
A price moves on a rule the operator never filed.
Stage proposes
A measurement or a price, its band and confidence
03 · Tool / write
2 modes
- OD-02
Acted without a person
Agent completes an act a person owed the decision on.
- OD-05
Cross-deployment path
Venue knowledge reaches the trading side.
Stage writes
Only where write access and approval policy allow it
04 · Output
1 mode
- OD-01
Undetected system state
Nobody is told the assist is off, degraded or unreset.
Stage returns
The call the official makes or the price that posts
05 · Change / Version
1 mode
- OD-07
Silent band regression
A model or rule change widens what the agent will assert.
Stage tracks
Model, prompt, tolerance bands and filed definitions
- Sev-1 · a path opens between deployments
- Sev-2 · a wrong value reaches the decision
- Sev-3 · input degrades, the assist stands down
Affected slices
One cohort carries most of the overrides
Take an aggregate override rate apart before trusting it: split it by cohort, and by deployment, because a small number of hard cohorts carry most of the overrides in both.
Slice performance — reported separately, not only in aggregate
Illustrative example| Slice | Failure rate | Lift | Lift vs. threshold | Status |
|---|---|---|---|---|
| Occluded and crowded incidents | 9.1% | 3.7× | Review | |
| Single-action micro-markets | 5.4% | 2.2× | Review | |
| Restarts after an interruption | 3.7% | 1.5× | Watch | |
| Clear, settled cases | 1.5% | 0.6× | Normal |
Bar: override-rate lift vs. clear-case baseline · scale 0–4.0× · tick marks the 2.0× review threshold
2 of 4 slices over threshold
Evidence-linked improvement
A cycle ends in a test, not a meeting
Nothing closes on a review meeting. It closes on a case the next release has to pass, and that suite is what the next output either side is measured against.
Improvement cycle · five stages
Switchback — the path turns at Improve and returns at Learn
- 01
Detect
Override rate rises in one cohort.
- 02
Diagnose
If the band held, the fault is in the definition, the version or the hand-off.
- 03
Improve
Every change ships against a version with the incidents attached.
- 04
Verify
The release is blocked while an affected case fails.
- 05
Learn
The case becomes permanent, and the separation controls are re-tested.
Learn → Detect
The return edge. The next detection runs against a suite one case longer.
Typical build scope
Twelve workstreams across six weeks
The build scope read against the delivery timeline. Week structure follows the six-week plan — discovery, sources, output workflow, evaluation, integration, then production validation and handover.
Workstream
Week 1Week 2Week 3Week 4Week 5Week 6
- 01Deployment separation and boundary definition.Week 1
- 02Venue and market-feed source assessment.Week 1, Week 2
- 03Definition, tolerance band and house-rule mapping.Week 1, Week 2
- 04Input ingestion and normalisation.Week 2
- 05Output logic and tolerance binding.Week 2, Week 3
- 06Confidence scoring and unresolved routing.Week 3
- 07Official or trader workflow.Week 3, Week 4
- 08Per-deployment system integration.Week 5
- 09Separation and tolerance cases.Week 4, Week 5
- 10Guardrails and approval controls.Week 4
- 11Decision-trail instrumentation.Week 5, Week 6
- 12Deployment, documentation and Agent Care handover.Week 6
12 workstreams · 6 weeks · bar shows the weeks a workstream is active — several run in parallel
Final scope and sequence confirmed in discovery
Engagement tiers
What each tier includes
Rows are the capabilities named in each tier's scope. Higher tiers include everything below them.
| Capability✓ in scope · — not at this tier | PilotOne competition or book | ProductionProduction deployment | AdvancedMultiple competitions / books |
|---|---|---|---|
| Introduced at Pilot | |||
| Output to your filed definitions | ✓ | ✓ | ✓ |
| Separation, and a human decision | ✓ | ✓ | ✓ |
| Measurement-consistency baseline | ✓ | ✓ | ✓ |
| Introduced at Production | |||
| Reporting by competition and market | — | ✓ | ✓ |
| Decision workflow in your systems | — | ✓ | ✓ |
| Approved write-back | — | ✓ | ✓ |
| Trading-platform integration | — | ✓ | ✓ |
| Introduced at Advanced | |||
| Multi-competition rule sets | — | — | ✓ |
| Multi-desk trading approvals | — | — | ✓ |
| High event volume | — | — | ✓ |
| Multi-jurisdiction filing controls | — | — | ✓ |
| Build price | From $5,000 | From $8,000 | Custom quote |
- Final build price
- Confirmed after discovery based on integrations, workflow complexity, transaction volume, approval controls and deployment requirements.
- Separate from build
- Build pricing is separate from recurring Agent Care, which covers managed monitoring, evaluations, incidents and verified improvements after launch.
What we need from you
What you bring, and what we build with it
Each input maps to a piece of build scope and a week in the delivery timeline.
You bring
We build with it
01Your filed definitions or your house rules
→Input ingestion and definition mapping
Week 102Representative events from one deployment
→Output baseline, band binding and version stamping
Week 203Your tolerance bands and who signs them
→Definition, tolerance band and house-rule mapping
Week 104Access to relevant APIs, feeds or exports
→Venue and market-feed assessment, then integration setup
Week 205Calls and prices you would not want repeated
→Separation cases and failure-mode testing
Week 406What a measurement is never allowed to decide
→Confidence scoring, unresolved routing, guardrails and separation controls
Week 307Named officials or traders to act on it
→Decision workflow, then pilot and production validation
Weeks 5–6
Nothing else is required
Deployment, documentation and Agent Care handover are ours.
Delivery timeline
Four phases across six weeks
The bands sit on the weeks the work occupies, so the fifth carries evaluation and launch together.
Phase
W1
W2
W3
W4
W5
W6
- DiscoveryW1
- BuildW2 – W3
- EvaluateW4 – W5
- Pilot & LaunchW5 – W6
Week focus
W1Deployment separation, definition mapping and the automation boundary
W2Source integration and the output baseline
W3Output workflow, tolerance logic and decision controls
W4Evaluation suite, separation tests and failure-mode testing
W5Per-deployment integration, pilot events and targeted corrections
W6One competition round run under the officials or the trading desk, then handover
- Reading the band
- Each bar covers only the weeks its work is named in. The fifth week carries two kinds of work.
- At the end of W6
- The round closes validation and Agent Care picks up monitoring.
- Duration
- Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.
Next step · Sports & Fitness agent
Build one deployment, on its own side of the line.
Tell us which side you are on — a competition's officials, or a licensed book's desk. Your governing body signs the definitions, or your licensee files the house rules; we build to one of them.
- Typical delivery
- 4–6 weeks
- Deployment
- Your rules or licence
- Human decides
- Two deployments
- After launch
- Agent Care available
Nestack Agents · Officiating and odds adjustment
AGT-SF-13 · Agent Care available after launch