Media AI agent · Transmission
Broadcast-QC & Compliance AI Agent
Measure a finished master against its delivery spec, check the access, classification and scheduling requirements, and hold anything that fails — a named compliance officer reads the result and signs the transmission release.
- Typical delivery
- 4–6 weeks
- Deployment
- Your stack
- Transmission
- Named officer
- After launch
- Agent Care
What this agent does
Measures the file against the delivery spec
In
- 01
Take the finished master, the delivery spec version and the transmission order it is booked against.
- 02
Read the classification, content warnings and scheduling restrictions already issued for the programme.
Reason
- 03
Measure loudness, true peak, flash and pattern risk, freeze, black and level against the spec.
- 04
Read the structure — line-up, clock, part timings, programme start and the audio track layout.
- 05
Check caption presence, timing and accuracy score, and whether a description track carries content.
Decide
- 06
Hold the file on any failed check and name the measurement that failed.
- 07
Route the whole programme to the compliance officer with what was measured and what was not.
Out
- 08
Hand over the fault list, the measured figures and the spec version each was read against.
- 09
Retain what was checked, what failed, what a person overturned and what went out anyway.
Product statement
The agent measures and reports. It does not pass anything for transmission: no programme goes out until a named compliance officer has signed the release.
Example workflow
One master, end to end
- 1Master receivedA finished file with the delivery spec version and the transmission order it arrives against
- 2Structure readLine-up, clock, black, part timings, programme start timecode and the audio track layout the spec asks for
- 3Measurements runIntegrated loudness and true peak, flash rate and pattern risk, freeze, black, levels and timing
- 4Access and content checkedCaption presence, timing and accuracy score, description track, classification, warnings and scheduling restrictions
No human action required
Stages 1 to 4 run without a person in the loop — the file is measured and the whole fault list is written before anyone is asked to read it.
Reaches the compliance officer clean.
Held with the failing measurement named.
Compliance officer and transmission controller
The compliance officer reads the fault list against the programme; the transmission controller confirms the version, the restrictions and the slot before the release is signed.
Accept · Re-check · Send back
Signed — handed back
- 6Transmission releaseOnly after a named compliance officer has read the fault list and signed it
- 7Outcome evaluatedCheck coverage, disagreement with the compliance viewing, faults found on air and the corrections logged
Overturned calls
Calls a person overturned are counted, by check and by genre.
What should not run autonomously
Human approval stays in control
Outside the boundary — human approval required
8 items
- Passing a programme for transmission.
- Clearing or suppressing a photosensitivity flag.
- Issuing or changing a classification or content warning.
- Overriding a failed delivery-spec check.
Automation boundary
Agent acts unaided
- ✓Measure the file against the named delivery spec version.
- ✓Read the structure and check caption and description carriage.
- ✓Flag photosensitivity risk, classification mismatch and restriction conflicts.
- ✓Hold the file on a failed check and name the measurement.
Write actions run only inside the approval boundaries agreed during implementation. Transmission is not one of them.
- Signing off a territory or compliance version.
- Moving a programme across a scheduling restriction.
- Changing a loudness, flash or caption threshold.
- Notifying a regulator after an on-air fault.
Example output
One master, measured before transmission
Everything the agent measures is attached to the file and the spec version it was read against.
QC output · single master
Illustrative example- Programme
- Documentary, one hour
- Delivered to
- Channel spec v4.2
- Slot
- Pre-watershed
- Checks
- Two failed — held
- Confidence
- 91%
- Transmission
- Not signed
As received
The master, the spec version it was booked against and the classification already issued.
Evidence used
- Integrated loudness read
- Flash sequence at 12:04
- Caption gap, 40 seconds
Why it is held
A sequence at 12:04 crosses the flash threshold; forty seconds carries no caption.
What the score decides
It decides how much the officer re-checks by hand, not whether the programme may go out.
Action
AcceptRe-checkSend backValue
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 master delivered
From the delivery and the transmission order
Measure what a viewing cannot
Integrated loudness, true peak, flash rate and pattern, caption timing and part structure are figures. They are read off the file rather than judged in the room.
Send the viewing a measured file
The measurements, the structure read and the carriage checks are finished before the compliance officer sits down, so the hour goes on the editorial call rather than on the meters.
Catch it in the file, not on air
A loudness, caption or flash fault found after transmission is a complaint, a correction and a regulator-facing record. Found in the file, it is a re-deliver and a slot that still holds.
Build an evidence trail
Retain the spec version, each measurement, the flags raised, who signed the release and what was found on air afterwards — on both paths.
Integrations
Typical integrations
Five system groups connect to the same agent. Which of them are in scope is decided in discovery.
Playout & scheduling
- Playout automation
- Scheduling
- As-run logs
- Traffic systems
Media & transcode
- MAM
- Transcode farms
- File QC engines
- Delivery watch folders
Access services
- Caption files
- Description tracks
- Signing files
- Access logs
Agent
Broadcast QC & compliance
- Measures the file
- Checks the carriage
- Names the fault
Compliance records
- Classification records
- Warnings
- Territory versions
- Complaint logs
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 transmission
Each control wraps the one inside it. A file clears every layer before it reaches a compliance officer, and the release itself sits outside all six.
L6 · Outermost — last line of defence
Inward → L1 · closest to the model
- L1 – L2
- Decide whether the result may stand
- L3
- Decides who signs the transmission release
- L4 – L5
- Keep the spec version and the trail honest
- L6
- Pulls automation back when signals degrade
How Nestack evaluates it
Evaluate the measurement, not only the pass or the fail.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the fault list a compliance officer opens
- E1Final-output evaluationDoes the fault list match what the compliance viewing found?
- E2Measurement evaluationDo the loudness, flash and timing figures match a reference meter?
- E3Tool evaluationDid it read the right master, version and spec, and write nowhere else?
- E4Access-service qualityDoes the caption score agree with a person reading the same hour?
- E5Slice evaluationHow does the result change across genres, eras and languages?
- E6Business outcomeHow many faults were found after the programme had gone out?
Floor — the faults that reached air anyway
Failure modes
Where each failure originates in the agent
Seven failure modes plotted against the five stages of the agent lifecycle. The ones that matter are the ones a meter reads as clean.
Agent lifecycle
Direction of processing →
01 · File & spec
1 mode
- BQ-01
Measured against a superseded spec
The channel moved the spec after the last delivery.
Stage gathers
The master, its spec version and the transmission order
02 · Measurement
2 modes
- BQ-02
Loudness passes, dialogue buried
Integrated level passes, speech sits under the mix.
- BQ-03
Stage lighting read as flash risk
A clean concert master is held on a false flag.
Stage measures
Loudness, true peak, flash and pattern, structure
03 · Access & content
2 modes
- BQ-04
Caption score clears accented speech
Errors dilute across the hour and the metric passes.
- BQ-05
Description track present, silent
Carriage is measured, not what is on the track.
Stage checks
Captions, description, classification and restrictions
04 · Report
1 mode
- BQ-06
A clean report read as clearance
Nothing failed is taken to mean it may go out.
Stage returns
The fault list a compliance officer reads
05 · Change / Version
1 mode
- BQ-07
A moved threshold re-reads the past
Yesterday's passes disagree with today's limit.
Stage tracks
Model, prompt, threshold and delivery-spec changes
- Sev-1 · a fault goes out on air
- Sev-2 · the file is measured against the wrong thing
- Sev-3 · a transmittable programme is held
Affected slices
The wrong calls sit in the speech and in the old masters
Accented and multilingual speech, re-mastered archive and as-live shows carry most of what a compliance officer overturns and most of what reaches air with a fault. Nestack reports performance by slice, not only in total.
Slice performance — reported separately, not only in aggregate
Illustrative example| Slice | Failure rate | Lift | Lift vs. threshold | Status |
|---|---|---|---|---|
| Accented and multilingual speech | 6.8% | 4.5× | Review | |
| Archive and re-mastered masters | 4.1% | 2.7× | Review | |
| As-live and live-to-tape | 2.7% | 1.8× | Watch | |
| Returning-series episodic deliveries | 1.1% | 0.7× | Normal |
Bar: wrong-call rate lift vs. episodic-series baseline · scale 0–4.0× · tick marks the 2.0× review threshold
2 of 4 slices over threshold
Evidence-linked improvement
A check is only as current as the spec behind it
Delivery specs, thresholds and scheduling rules are revised, and a check nobody re-reads keeps measuring against a document that has been replaced.
Improvement cycle · five stages
Switchback — the path turns at Improve and returns at Learn
- 01
Detect
On-air faults, overturned calls or held-file counts move on one channel.
- 02
Diagnose
Traced to the spec version, a measurement, a carriage check, a threshold or a restriction rule.
- 03
Improve
The threshold, spec mapping or check is changed with your compliance officer, and version-linked.
- 04
Verify
Re-run against masters already transmitted, including the ones that went out with a fault.
- 05
Learn
The fault is written into the delivery spec your suppliers and the agent both read.
Learn → Detect
The return edge. Each cycle re-reads the spec and the scheduling rules as they now stand — a threshold that passed last quarter may not be the one in force.
Typical build scope
Twelve workstreams across six weeks
The build scope read against the delivery timeline. Week structure follows the six-week plan — discovery, specs and measurement, access-service checks, evaluation, the release workflow, then production validation and handover.
Workstream
Week 1Week 2Week 3Week 4Week 5Week 6
- 01QC discovery and automation-boundary definition.Week 1
- 02Delivery-spec and channel-profile mapping.Week 1, Week 2
- 03Master ingest and structure reading.Week 2
- 04Loudness, level and true-peak measurement.Week 2, Week 3
- 05Photosensitivity flagging and hold rules.Week 2
- 06Caption presence, timing and accuracy.Week 3
- 07Audio-description and access-service checks.Week 3, Week 4
- 08Classification, warning and restriction checks.Week 4
- 09Evaluation suite, slices and regression masters.Week 4, Week 5
- 10Fault report and compliance-review workflow.Week 5
- 11Playout, scheduling and as-run integration.Week 5, Week 6
- 12Observability, deployment 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 channel, one spec | ProductionProduction QC integration | AdvancedMulti-channel / multi-territory |
|---|---|---|---|
| Introduced at Pilot | |||
| Delivery-spec, loudness and true-peak checks | ✓ | ✓ | ✓ |
| Caption timing and accuracy scoring | ✓ | ✓ | ✓ |
| Audio-description carriage checks | ✓ | ✓ | ✓ |
| Classification, warning and scheduling checks | ✓ | ✓ | ✓ |
| Photosensitivity flagging and hold | ✓ | ✓ | ✓ |
| Named-officer transmission release | ✓ | ✓ | ✓ |
| Baseline evaluation | ✓ | ✓ | ✓ |
| Introduced at Production | |||
| Playout, scheduling and as-run integration | — | ✓ | ✓ |
| Additional channels, specs and master volume | — | ✓ | ✓ |
| QC observability and evaluation | — | ✓ | ✓ |
| Introduced at Advanced | |||
| Multi-territory versions and enterprise controls | — | — | ✓ |
| Build price | From $5,000 | From $8,000 | Custom quote |
- Final build price
- Confirmed after discovery based on the channels and delivery specs in scope, check coverage, access-service depth, volume of masters, release 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
01How a programme reaches transmission today, and who signs it
→QC discovery and automation-boundary definition
Week 102The delivery specs and channel profiles you deliver against
→Delivery-spec and channel-profile mapping
Week 103Masters that already transmitted, the clean and the faulty
→Master ingest, structure reading and the measurement baseline
Week 204Your loudness, flash and caption thresholds as they stand
→Measurement thresholds, photosensitivity flagging and hold rules
Week 205Classification records, warnings and scheduling restrictions
→Classification, warning and restriction checks
Week 406Faults that went out, and how long it took to find out
→Evaluation suite, slice reporting and regression masters
Week 407Named compliance officers and transmission controllers
→Fault report, review workflow and the release step
Weeks 5–6
Nothing else is required
Deployment, documentation and Agent Care handover are ours.
Delivery timeline
Four phases across six weeks
Phases are drawn over the weeks they actually occupy. Week 5 carries both the evaluation slices and the first releases signed by a named officer.
Phase
W1
W2
W3
W4
W5
W6
- DiscoveryW1
- BuildW2 – W3
- EvaluateW4 – W5
- Pilot & LaunchW5 – W6
Week focus
W1Programmes in scope, delivery specs and who signs a release
W2Master ingest, structure reading and the first measurements
W3Caption, description and access-service carriage checks
W4Classification, restriction and threshold rules, slices and evaluation
W5First releases signed under approval and targeted corrections
W6Production validation, on-air fault review and handover
- Reading the band
- The photosensitivity flag and its hold rule are in place in week 2, before any file reaches a compliance officer in week 5. The bars show that order, not a smooth ramp.
- At the end of W6
- Programmes have gone out under a named compliance officer's signature, with the spec version and every measurement behind each one on file, then Agent Care takes over monitoring.
- Duration
- Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.
Next step · Media AI agent
Build a broadcast-QC agent around your delivery specs and your compliance team.
Show us a programme you delivered last month, the spec and channel profile behind it, and a fault that reached air. We'll measure that master against your own spec and show you what it would have held, what it could only flag, and what no meter would have caught.
- Typical delivery
- 4–6 weeks
- Deployment
- Your playout and QC stack
- Transmission
- Named compliance officer, always
- After launch
- Agent Care available
Nestack Agents · Broadcast QC & compliance
AGT-ME-09 · Agent Care available after launch