Broadcast report · 23 August 2026
Miss Universe Jamaica 2026
Report and post mortem
We carried the national final live and free to the world. Over five and a quarter hours, the broadcast reached 3,698 viewing devices in 45 countries — close to two thousand hours of attention, three quarters of it on a phone, with the audience strongest after midnight.
3,698
Unique viewing devices
Mux viewer identities that started or attempted playback across 45 countries
691+
Observed concurrent
Direct Mux observation; 1,044 active watch pages at the event peak
1,878hrs
Total watch time
78 continuous days of viewing, delivered in a single night
45
Countries
Jamaica, the diaspora and the wider Caribbean
Part one · Performance
The audience built through the night
This was not a crowd that arrived for the opening and drifted away. It grew for four straight hours and peaked at the crowning — the signature of an event being shared while it happens.
How peak concurrency was established
Three measurements describe different parts of the same audience. 1,044 watch pages were open at 1:25 AM, reconstructed from the playback endpoint’s request volume and known 30-second poll cadence. Mux recorded 1,878 hours of actual playing time across the reporting window, while its live counter was directly observed at 691. Because the historical Mux maximum was not retained and an open page is not always playing video, this report states 691+ observed concurrent and the 1,044-page operational peak rather than presenting a modelled estimate as a measured maximum.
Reach
Two thirds at home, one third abroad
A Jamaican national event is never only a Jamaican audience. Just under a third of the people watching were in North America and the UK, and the wider Caribbean turned out in numbers of its own.
Unique viewing devices by country
+ 37 more: Guyana, Panama, United Kingdom, Turks & Caicos, Colombia, Belize, Mexico, Trinidad & Tobago, Dominican Republic, Japan, Germany, France, Switzerland, Netherlands, Spain, UAE and others.
Mux counts unique viewing devices here. First-party request activity, which is weighted toward time spent rather than reach, was 66.3% Jamaica, 26.4% United States and 3.0% Canada. The two sources agree on the market shape and suggest the home audience contributed a slightly larger share of sustained activity.
How they watched
Three in four watched on a phone — a second-screen audience with the event in one hand and a group chat in the other.
Attention held
- Mux average completion*
- 94.5%
- Playing time per unique device
- 30.5 min
- Chat messages
- 4,238
- Chat participants
- 119
- Reactions
- 1,000
- Browser passes issued
- 1,677
- Email participation attempts
- 224
- Verified accounts from event
- 0
- Cost to the viewer
- Free
* Mux completion is average playing time divided by the average content duration of individual live views. It does not mean that the average viewer watched 94.5% of the full five-hour programme. Seats are browser access credentials, not people; the 3,698 Mux devices are the stronger reach measure.
Community conversion gap
The event-era shortcut created 224 account rows from typed email addresses, but none was inbox-verified, claimed, password-backed or accompanied by valid marketing consent. Those are participation attempts, not acquired members. The post-event verified-account flow has removed that shortcut.
Part two · Post mortem
The night as it happened
The venue's uplink could not sustain the encoder's bitrate, and the first half of the broadcast fought it. Recording the sequence honestly is how the next one goes better.
Advertised start. Doors had been open an hour; 426 viewers were already waiting.
Source connects — 19 minutes late. Waiting rooms convert to live automatically, with no action from any viewer.
The worst hour: 223 disconnect intervals totalling 28m 55s as the venue uplink saturates.
Playback authorisation rate-limits under load — 2,394 rejections, concentrated on one widely shared link.
Encoder bitrate reduced to 1500 kbps on our recommendation. Disconnects fall away and the second clean recording begins.
Peak chat minute — 73 messages in sixty seconds.
Peak watch-page activity — 1,044 pages open. The separately retained Mux live-counter observation was 691.
Audience exits sharply as the main programme concludes — an ending, not a drift.
Source disconnects for the last time after 5h 16m. The full record is rebuilt automatically and published as a replay.
Root cause · source ingest
The encoder disconnected and reconnected 270 times, losing 41 minutes 30 seconds of source in total. The longest single outage was 3m 08s; 223 of those intervals fell in the first hour alone. Both recordings finished clean at 720p29.97 with no asset-level errors. Server responses were generally fast, but playback authorisation still failed or rate-limited 1.16% of requests. The repeated missing source media remains the primary evidenced cause of buffering.
- Source window
- 5h 16m 17s
- Media recorded
- 4h 09m 12s (78.8%)
- Disconnect intervals
- 270
- Longest outage
- 3m 08s
- Playback authorisation requests
- 305,113
- Playback success rate
- 98.84%
- Authorisation failures
- 3,544 (1.16%)
- p95 authorisation latency
- 53 ms
Playback smoothness recovered
Fallback softened the later drops
Early viewers did see buffering and spinners: 57 chat messages mentioned buffering or freezing, and the incident screenshots captured it directly. Once a recorded segment was ready, the platform could serve that fallback during a later source drop and then return viewers to live without a new link. Mux attributes 63 of the 1,878 playing hours, or 3.4%, to fallback playback. It reduced later disruption; it did not make the source failures invisible.
What changes
Already fixed, and what comes next
Several event-era weaknesses were closed within hours of the broadcast ending. The remaining work is operational rather than architectural.
Shipped since the event
- Stable event URLs, with no reusable pass secret in the address bar
- Secure HttpOnly browser credentials in place of forwardable watch links
- Chat identities tied to a verified inbox rather than a typed address
- Event-specific audience counts, replacing the site-wide presence figure
- Explicit play-with-sound startup, fixing the silent-open problem
- A wider theatre layout and a much less frequent playback refresh
Before the next event of this size
Redundant ingest
A wired primary uplink with bonded cellular failover, plus a hot spare encoder configured identically. The primary evidenced source of lost programme time was one unstable uplink with no hot fallback.
Audience telemetry on by default
Session heartbeats verified before doors open, a privacy-safe viewer ID sent to Mux, and one-minute audience aggregates persisted — so a report never again depends on a screenshot taken at the right moment.
Release freeze
Production frozen two hours before doors through to event end. Eight deployments went out during this broadcast; emergency changes helped, but changing production mid-show makes every regression harder to attribute.
Load tested to 5,000
Playback grant, state refresh, chat and reconnect exercised at 2,000 and 5,000 concurrent sessions before the next event of this size.
Application error budget
Fix or disable side-chat until its 1,757 server errors are resolved, and stop signed-out clients polling buyer-only endpoints that returned 50,333 expected but wasteful unauthorised responses.
Verified community conversion
Measure verification requested, delivered, opened, account created, chat joined, followed and consented as separate stages. A typed email is an attempt, not a member or a marketing permission.
Method
How to read these numbers
Where a figure is inferred rather than counted, it is said so here rather than buried.
Measured. Mux viewing identities, views, watch time, completion, geography, device and playback-quality figures come from Mux Data, recorded on the event’s own live stream between 23 August 8:00 PM and 24 August 4:00 AM US Eastern. Watch-page concurrency, request volumes, error rates and latency come from production logs. Seat claims, chat volume and the timeline come from our own database.
Concurrent audience. The 1,044-page figure is a five-minute operational proxy, not a Mux viewer reading. Aggregate playing-time ratios can produce a directional estimate, but the Mux and first-party populations, time windows and view-finalisation rules differ. The report therefore uses 691+ as the directly observed video concurrence and presents 1,044 separately as the infrastructure-facing page peak.
Observed. Mux’s live counter was seen at 691 during the broadcast and 511 at 1:48 AM. Real-time monitoring data expires within a day, so these are point-in-time records rather than a retained maximum.
A viewer is an approximate device. The player did not send a durable customer-defined Mux viewer ID, so Mux used its SDK-generated identity. That usually joins reconnections in the same browser, but one person on a phone and laptop can count twice, cleared browser identity can count again, and a family around one screen counts once. Read 3,698 as distinct viewing devices/browsers, not a verified headcount.
Chat identities understate people. One widely forwarded link carried 51% of all messages, so 119 participants is a count of identities rather than humans. That link-sharing weakness is among the items already closed.
What we could not measure. Session heartbeats were not enabled, so there is no first-party record of who was watching minute by minute. That is why concurrency here is reconstructed rather than read off a retained chart. Vercel Web Analytics and the event-view funnel were also incomplete, so they are not used as reach totals.
This report covers the 8:00 PM–4:00 AM live-event reporting window; later on-demand viewing is excluded from these totals. Prepared by Jamaica Events · jamaicaevents.com.