The evidence changed the commitment we could make

In 2018 I approved a simulator-mount geometry after an initial static check appeared satisfactory. The design held its position in that demonstration, and the programme was preparing 240 preproduction units. A later repeated-load review of forty rigs produced a different account: twenty-eight held position and twelve tilted under the reviewed sequence. The first check had not represented the condition that exposed the weakness.

The mechanical team wanted to understand whether the result could be addressed through assembly adjustment. The commercial team was preparing for the planned release date. I had approved the geometry, so I needed to explain why the later evidence changed my conclusion rather than describe the hold as a precaution imposed by somebody else. We stopped all 240 preproduction units before shipment and reopened the mount design with the team responsible for the repeated-use evaluation.

The redesign cost ₹62 lakh and added seven weeks to the programme. The case does not claim a patient incident, because these were teaching rigs and the units had not been shipped. It also does not treat the successful hold as evidence that the original approval was sound. The additional engineering work and delay were real consequences of giving too much weight to the first stable position. The account retains that responsibility alongside the value of the later review.

I changed the approval discussion so that the team had to state which use conditions a demonstration represented and which remained to be examined. A single stable result could still be useful, but it could no longer silently stand for repeated-use behaviour. The lesson was not that one particular loading sequence answers every question about a teaching rig. It was that the scope of the conclusion must match the scope of the evidence. I wanted the programme to make that relationship explicit before another attractive demonstration carried more authority than it had earned.

A 2022 positioning review included seventy-two participants: forty-eight right-handed and twenty-four left-handed. The initial geometry met the reviewed positioning criterion for all forty-eight in the first group and eighteen in the second. Six left-handed participants did not meet it. Reporting only that sixty-six of seventy-two succeeded would have concealed the concentration of the difficulty in a group whose experience mattered to the design.

Meghna Qervesi asked the engineering team to examine the positioning limitation rather than treat it as a training issue for the six participants. Some colleagues preferred to retain the geometry and explain a different setup sequence. I wanted to understand whether the mechanism itself was constraining the user before asking an educator to work around it. The sample was not a census of every possible student, but it was enough to challenge the claim that the first arrangement served the reviewed groups equally.

We funded ₹84 lakh of adjustable-mount work. That changed the engineering scope and deferred the broader suitability claim while the revised arrangement needed its own evaluation. The case does not assert a measured improvement in learning, clinical skill or later patient care. Meeting a positioning criterion is an engineering and use-condition observation; educators retain responsibility for teaching and assessment. The six results remained part of the record rather than disappearing behind the larger passing group.

My contribution was to make the population inside the result visible to the investment discussion. A total success rate can be useful, but it should not hide who was excluded by the geometry. I also avoided making the small sample prove more than it could: the decision justified further design work and evaluation, not a universal statement that every user's needs were now understood. The trade-off was additional cost and a less complete commercial claim in exchange for addressing a concrete limitation that the first product account had not adequately represented.

In 2023 we checked an instrumented teaching rig's force display against twelve specified reference points. Nine were within the agreed bench-engineering specification and three were outside it. The display itself looked precise, with orderly numerical output. That appearance did not establish that the reported values met the required comparison at every tested point. The three exceptions mattered to the account of what the sensing arrangement could support.

The sensing team proposed describing the result as broadly satisfactory while completing adjustments. The programme sponsor wanted to preserve the release schedule. I asked for the failed points to remain explicit and for the calibration setup to be examined before a broader acceptance statement was made. We were discussing engineering behaviour under stated conditions, not a recommended clinical force or a judgement about a student's competence. The clarity of a display should not allow those different questions to become mixed together.

We funded ₹53 lakh of calibration-fixture work and repeated the twelve-point check under the stated bench conditions. All twelve then met the agreed comparison. That was a bounded result for the repeated check, not proof of every value across every operating condition or of the equipment's educational effectiveness. The original nine-and-three result remained in the development record, alongside the work needed to reach the later result. It was not rewritten as an early success merely because the follow-up was satisfactory.

The decision required additional engineering expense and a more precise account of acceptance. I wanted colleagues to distinguish the measurement, its reference and the claim being made from it. In a leadership profile, the case is useful because it shows how a seemingly small exception can change the scope of a senior approval. My responsibility was to ensure that the evidence supported the engineering statement and that its boundaries remained visible, rather than let numerical neatness or a majority of passing points substitute for the comparison the programme had agreed to meet.

A 2024 controller review compared an isolated bench demonstration with the conditions of a teaching room containing thirty-six stations. The isolated setup had shown an eight-millisecond response. Under the reviewed shared-room workload, the median was eighteen milliseconds and the 95th-percentile response was 110 milliseconds. Those measures came from different conditions and were not interchangeable proof of the same performance. The long tail was particularly important to understanding the uneven response participants encountered.

The controller team initially emphasised the median, which made the room result appear relatively close to the demonstration. Educator feedback concerned the slower episodes rather than the typical one. I asked the team to preserve the distribution and the workload instead of selecting a single attractive number. The question was what the shared system did under the reviewed conditions, not whether we could reproduce the fastest isolated observation or infer that every participant experienced the median.

We funded a ₹96 lakh local-controller programme and repeated the thirty-six-station review. The 95th-percentile response was then thirty-two milliseconds under the stated test conditions. We reported that change as a result of the reviewed configuration and workload. It did not establish that all delay had disappeared, that every future room would behave identically or that students had achieved better clinical outcomes. The isolated eight-millisecond observation remained a different test context rather than a benchmark silently substituted for the classroom result.

The cost bought engineering work on a specific response problem, not a universal claim of immediacy. I wanted the commercial and teaching discussions to retain what the distribution showed and what still required evaluation. A median is a useful statistic, but it can leave a consequential tail outside the sentence used to describe the product. My contribution was to make that tail part of the resource decision and to keep the later improvement attached to its tested conditions, rather than turn a better percentile into a promise the evidence could not support.

Our 2025 tool-free jaw-cassette review recorded 1,200 installation attempts, of which 1,092 were correctly seated and 108 were mis-seated. These were attempts, not 1,200 distinct people. The proposed mechanism was intended to make changing a teaching component straightforward, but the observed failures showed that the installation state could not be assumed from the simplicity of the intended action. We needed to understand the latch and the user's interaction with it.

The mechanical team proposed a revised latch. Commercial colleagues wanted the replacement to move quickly into a broader release, and some early demonstrations appeared encouraging. I asked for the redesign to be reviewed as a new configuration rather than inherit acceptance from its intended purpose. We spent ₹58 lakh on the work and ran a separate 200-attempt review. That review recorded 196 correctly seated installations and four unresolved results.

We kept broader release paused. The second set was smaller and differently conducted, so the two success proportions were not presented as a controlled demonstration of universal improvement. The four unresolved attempts still needed explanation. I did not ask an educator to compensate for the uncertainty through instructions that the engineering team had not shown to be adequate. Nor did I claim a clinical consequence from a teaching-equipment installation review. The case concerns a product-use question whose remaining exceptions prevented the wider conclusion the programme wanted to make.

The immediate consequence was continuing development effort without the broader release. That is a legitimate endpoint for an executive case. The redesign expenditure did not require us to announce success merely because most of the later attempts were correct. My contribution was to keep the unfinished question specific: which conditions led to the unresolved installations, and what evidence would justify the next decision? A leadership account should make that open work visible rather than round four exceptions down to a finished product story. The value lies in the judgement about the next commitment, not only in launches that can be described as complete.