Where the audience account changed the engineering recommendation

In April 2024, we reviewed revised video-encoding recipes for forty selected titles. The development team expected the new treatment to reduce the bitrate needed for acceptable presentation. The initial investment account assumed that one revised approach could move across the whole selected set. Colin Merrowby's content colleagues wanted to understand what happened to motion and grain, particularly where those qualities were part of the intended look rather than noise to be removed.

I authorised an eight-week, CAD410,000 study with comparative viewing alongside the engineering measurements. We defined the chosen rendition used for the numerical comparison and retained the existing recipe as the reference for each title. The reviewers identified thirty titles for which the revised recipe was acceptable in the examined conditions. Ten remained on their existing treatment while concerns about motion or grain were investigated. The finding was not a judgement that those ten titles were defective; it concerned the suitability of the proposed processing.

The development sponsor preferred a single migration because it simplified preparation and the operating account. I supported a split decision. The thirty approved titles could proceed through their defined release checks, while the remaining ten kept their existing recipes. That gave up some simplicity and accepted additional compute and review work. It also meant that the programme could not present the whole forty-title group as a completed conversion.

For the approved thirty, the unweighted median bitrate of the chosen rendition moved from 4.8 to 3.9 megabits per second. That comparison did not represent traffic-weighted catalogue delivery, every rendition or a measured financial saving. Device checks remained attached to the defined release path. We did not use the median to claim that every scene looked identical or that every title would produce the same reduction.

The decision made exceptions a legitimate part of the implementation plan instead of an embarrassment to be averaged away. Content and engineering colleagues could agree where the new treatment was useful and where further work had a specific purpose. I would now identify the likely difficult material earlier, so that the initial budget includes the cost of investigating it. A better processing method should earn its place through the programme it carries, rather than through the convenience of applying one recipe everywhere.

A September 2024 live-event trial compared two delivery settings across 12,000 eligible playbacks on selected client versions. The product team wanted a shorter delay from the event because viewers could otherwise encounter developments through another route first. Engineering wanted to understand the effect on continuity under the connections represented in the trial. We agreed to examine both the delay and the occurrence of rebuffering, rather than treat the latency figure as the complete experience.

Eligible playbacks were randomly allocated between two groups of six thousand. The lower-delay group recorded a median event-to-display delay of 3.2 seconds, with 480 sessions experiencing at least one rebuffering interruption. The more buffered group recorded 6.1 seconds, with 180 such sessions. Those counts corresponded to eight and three per cent within their respective groups. They were not measures of interruption duration, and the latency medians did not describe every playback.

The product sponsor preferred the lower-delay setting for the initial release. It produced a clearer proximity claim and seemed closer to the way the event was promoted. I recommended retaining the more buffered design while continuing targeted work on the faster option. Colin Merrowby supported examining the consequence for the programme as well as the promotional message. We accepted a longer observed delay in exchange for the lower incidence of interrupted sessions in this selected comparison.

The nine-week investigation cost CAD680,000. Its outcome was a release choice for the tested context, not a statement that six seconds is the correct delay for every live service. We did not claim improved retention or revenue, because the trial had not established either. The lower-delay work continued with a clearer question about the receiving conditions and client behaviour that required attention.

The case changed the way our investment discussions described live delivery. A movement toward the event could be valuable without being the only measure that mattered. I would bring programme teams into the definition of acceptable trade-offs earlier, before a latency objective becomes attached to a launch narrative. That would help us distinguish settings where immediacy is central from those where an interruption is the more damaging experience, and would make the next engineering experiment answer a specific service question.

In February 2025, a studio programme was approaching handover with two IP transport fabrics. The duplicated paths were an important part of the design, but they did not answer every timing question. During a seven-week, CAD920,000 investigation covering six selected fault scenarios, degradation of the primary clock exposed endpoint behaviour that the handover team had not expected. Under the then-current selection policy, affected endpoints did not move to the healthy reserve as intended.

The project sponsor argued that the transport redundancy had passed its checks and that the remaining timing concern could be monitored after handover. The engineering team wanted the selection behaviour understood before programme operations depended on it. I supported a three-week hold and authorised CAD260,000 of additional work. The cost and delay were explicit; describing two networks as redundant could not resolve a different dependency shared by the receiving equipment.

Specialists examined the clock-selection policy and the monitoring needed to recognise degraded timing. They revised the arrangement and repeated the selected scenarios. The repeat showed the intended behaviour within that test set and supported the next handover decision. I did not prescribe a universal configuration or substitute my judgement for the engineers responsible for the design. My role was to ensure that the unanswered question had the time, authority and funding needed for a useful answer.

The result did not establish a probability of future availability. Six selected scenarios were an investigation plan, not a representative sample of every fault the studio might encounter. Operational monitoring, documented responsibilities and wider programme preparation remained necessary. The project also retained a clear distinction between a transport-path failure and a timing-source problem, so later reports would not use one successful test as evidence for both.

The useful organisational change was a more precise handover account. Programme operations could see what had been examined, what had changed and which responsibilities continued after acceptance. I would now ask for the timing-degradation scenarios at the beginning of the studio plan, alongside the transport tests. That would expose the relationship between duplicated infrastructure and shared selection behaviour before a fixed handover date made the finding look like an avoidable late objection.

An August 2025 programme examined an automated captioning approach through 120 selected clips from live material. Caption specialists and editors worked with twelve deaf or hard-of-hearing reviewers. The engineering sponsor initially emphasised the proportion of clips carrying a text track. The review asked a different question: whether that text conveyed the intended meaning at a usable point in the programme. A caption's presence was necessary but did not settle its usefulness.

Twenty-seven clips contained a meaning-changing caption error, eighteen had a display-timing issue and nine appeared in both groups. The unique affected set was therefore thirty-six clips. We kept the overlap explicit rather than adding the counts and implying forty-five different failures. The sample was selected for investigation; its result was not a percentage estimate for every programme or every viewer using the service.

The automation sponsor wanted to move toward a broad release while tuning the remaining cases. Caption and content colleagues argued for retaining human review in selected formats where a changed name, number or qualification could materially alter understanding. I supported that narrower approach and funded a timing repair. The ten-week programme cost CAD530,000. It accepted continuing review effort instead of booking a general saving from removing that work.

We repeated the thirty-six affected clips after the changes. Seven still contained a meaning issue and four a timing issue, with two in both categories, leaving nine unique clips unresolved. The repeat showed progress within the examined set. It did not establish that all other programmes were accessible or that a regulatory requirement had been satisfied. The remaining nine stayed in the exception and review process, and the release recommendation retained those limits.

The decision helped the team distinguish an automated output from a service that people could rely on to follow the material. It also made the reviewers' contribution part of the engineering decision, rather than an endorsement sought after the release had been chosen. I would now involve those perspectives when selecting the first clips and defining the reporting categories. That would make the initial account more informative and prevent a simple coverage measure from concealing the two different problems we needed to solve.

In May 2026, we examined twenty-four selected programme sequences combining material from different production sources. The development proposal sought a more consistent conversion path for HDR production and SDR delivery. The transport checks showed valid outputs, but creative and engineering reviewers rejected the appearance of six SDR sequences on the reference chain. A technically accepted signal had not preserved the intended look closely enough for those reviewers to approve it.

The initial eight-week review cost CAD610,000. The pipeline sponsor preferred to keep the common conversion placement because it simplified operations. Content colleagues wanted the affected sequences examined in relation to their source treatment and the final output being reviewed. I supported further work on conversion placement and monitoring, while keeping the creative decision with the people responsible for the programme. Engineering could explain the alternatives; it could not declare an appearance acceptable solely because the transport format was valid.

In July, I authorised a four-week extension at CAD190,000. Specialists revised the relevant treatment and presented the six sequences again through the reference chain. By September, those revised sequences had been accepted in that setting. Follow-up on receiving devices remained unfinished. The programme therefore had a completed reference review and an incomplete wider delivery assessment, which were recorded separately in the recommendation.

We did not claim identical appearance on every television or approve a full-channel rollout from those six revisions. Display capabilities, settings and delivery conditions could affect the experience beyond the reference environment. The value of the completed work was more specific: it resolved the identified creative objections on the examined chain and clarified what the device assessment still needed to establish.

The case changed the language used around conversion readiness. A successful technical output became one part of the account, accompanied by the intended look, the review conditions and the remaining receiving-path work. I would now agree the creative review points before the pipeline architecture is presented as a finished operating choice. That would allow engineering and content teams to investigate differences while the arrangement remains open to change, and would reduce pressure to treat a late visual objection as merely subjective resistance.