Where a complete-looking dataset needed a narrower claim

In 2019 I approved a freshness account for 120 road segments using their recent update dates. A subsequent review separated fifty-six segments with recent field observations from sixty-four whose underlying observations were older but whose records had recently been edited. I had treated the date of maintaining the file as evidence of observing the road again. The records were useful, but the description I approved gave them a different meaning.

The product sponsor wanted to retain the single freshness status because customers needed a simple way to judge the release. Another suggestion was to remove the sixty-four older observations from the product altogether. I rejected both responses. Keeping the status would preserve the mistaken inference; deleting the older evidence would discard material that could still be useful within its actual date and purpose. The distinction had to survive in the product rather than be solved by pretending one group did not exist.

We committed ₹41 lakh to provenance work that separated observation dates, record revisions and release dates. I took the correction to the recipients of the original account and accepted responsibility for my approval. Peter's survey team helped identify what the source records established, while qualified owners retained decisions about further observation and actual use. We did not set one expiry period for every kind of feature or infer that all older observations were wrong.

The outcome was a more accurate description of the evidence behind the release. It did not resurvey all 120 segments, change a road, establish vehicle clearance or demonstrate a reduction in incidents. The cost included the programme expense and the loss of an uncomplicated freshness message. Commercial colleagues had to explain why a recently maintained product could contain evidence of different ages without treating that fact as a defect to conceal.

I would now ask which event a date records before approving a quality status based on it. The useful question is not simply whether the database was touched recently, but whether the evidence needed for the intended decision has been obtained. A product can be carefully maintained and still require a user to recognise that some observations are older. Keeping that history visible allows a responsible next question; replacing it with a recent timestamp only makes the uncertainty harder to find.

The 2022 review covered eighty-six selected crossings in a planning dataset. Fourteen were grade-separated without a connection between the roads, while seventy-two were actual junctions in the reviewed evidence. The flat drawing made the lines appear to meet in all eighty-six places. A proposed network account treated those visual intersections as connections, creating relationships the physical arrangement did not support.

The delivery sponsor preferred to release the network and qualify it as indicative, arguing that experienced users would recognise the limitations. I accepted that planning data are not a substitute for operational approval, but that boundary did not excuse an avoidable error in the relationships we were selling. A general qualification could not make an unconnected crossing into a junction. I recommended holding the affected network account while the selected relationships were reviewed with survey colleagues.

The ₹53 lakh programme retained the fourteen separate-level cases and the seventy-two junctions in the model, with the source relationship inspectable through the authorised process. We did not simply remove all crossings, which would have broken genuine connections. Nor did we infer that every turn was permitted wherever a junction existed. Connection, turn restriction, vehicle suitability and current road conditions remained different questions for the relevant evidence and responsible owners.

The immediate result was a corrected relationship account for the selected population. No claim was made that every road in the wider product had been verified or that a particular real journey was approved. We supplied no driving instructions from this fictional exercise. The commercial consequence was a narrower release and the review cost, rather than a claimed saving in distance or fuel that the work had not measured.

I would place the relationship question in the initial data specification instead of checking only whether the line geometry looks plausible. A visually clean map can contain an unusable network, and a technically connected network can still be unsuitable for an actual movement. The data team's responsibility is to represent the evidence faithfully enough for the intended analysis, then make the remaining decisions explicit. That is more demanding than attaching a general caution to a finished picture, but it gives the customer a product whose basic relationships can be examined rather than assumed.

In 2023 a selected register contained eighty road-restriction records. Twenty-six were described by their sources as temporary and fifty-four as continuing. The product proposal would remove temporary records when their expected end dates passed, keeping the displayed network uncluttered. The date represented an expectation in the source account, however, not a later observation or authoritative confirmation that the restriction had ended.

The product team wanted a predictable maintenance rule. An unresolved record could remain visible indefinitely and make the product look stale. I agreed that an unreviewed backlog was not a satisfactory answer, but automatic removal would replace a known uncertainty with an unsupported implication that the restriction no longer applied. Treating the fifty-four continuing records as permanent would create a similar error in the other direction. The source status described what we knew, not an eternal property of the road.

We allocated ₹44 lakh to separating the reported condition, expected period, source and later review status. The programme established a route for obtaining and evaluating updated evidence through the appropriate owners. It did not prescribe a universal legal rule for when a restriction ends or allow our data team to overrule the authority responsible for it. Expected dates remained useful for prioritising review without becoming automatic proof of a physical or regulatory change.

The resulting product could show that a record required attention rather than silently treating it as resolved. That meant additional explanation for users and continuing work for the team. We did not claim that all twenty-six temporary restrictions had subsequently been lifted or that all fifty-four continuing records were still correct forever. No improvement in journey time, safety or regulatory compliance was inferred from the revised status design.

I would agree how a condition becomes known to have changed before accepting an expiry rule into a commercial data product. Removing a record is itself a statement when users interpret its absence as permission or availability. The enterprise needs a process for uncertainty that is operationally manageable, but ease of maintenance should not decide what the world is assumed to have done. A useful status can tell a user that the next answer is still required without claiming either that the old condition continues indefinitely or that it has certainly disappeared.

The 2024 product review examined 120 selected source items intended for a licensed planning release. The reviewed terms permitted redistribution for eighty-four items, while thirty-six were limited to internal research. All could be opened and processed by our team. A proposed combined release treated their technical availability as if it supplied the same commercial permission for every item.

The sponsor wanted the complete set because removing sources would reduce the product's apparent coverage. One suggestion was to alter the file format and merge the material into our own structure before release. I did not accept that a transformation of format settled the underlying rights. Qualified legal and commercial owners needed to interpret the actual terms, including any permitted derived use. The data function could preserve provenance and enforce the approved scope, but it could not create permission by making a new file.

We committed ₹35 lakh to a narrower product and the supporting source-rights account. The eighty-four items proceeded within the reviewed redistribution conditions; the thirty-six remained outside that release and available only within their permitted internal-research scope. We did not label the latter worthless or imply that all internal research could automatically be turned into an external product. Further rights discussions, if commercially justified, belonged with the authorised owners.

The result was a release whose content matched the permission established for it. Its smaller scope was a real commercial trade-off. We made no claim that the decision increased revenue, eliminated every rights risk or proved that all future uses were approved. The programme also did not transfer ownership of the underlying material to us merely because we had paid to organise it. Each intended use still needed to fit the relevant terms and review.

I would include source rights in the product brief before promising a coverage figure. Waiting until the final export makes a legitimate limitation feel like a late obstacle imposed on an otherwise finished offer. The scope of a data product includes what the enterprise can supply and what the customer is allowed to do with it, not only the volume the software can read. Keeping that boundary visible helps management decide whether to negotiate, narrow the offer or fund new evidence, rather than ask a technical transformation to answer a commercial-rights question.

The 2025 review concerned forty-eight location points in a customer-planning product. Thirty represented road approaches and eighteen were parcel centroids. The intended use required entrance positions. Both original categories could have precise coordinates while answering a different question: a centroid locates a calculated centre of a parcel, and an approach point does not necessarily establish the entrance through which access is possible.

The commercial sponsor wanted to release the points with a general location label because customers could investigate the final approach themselves. I accepted that the product did not approve a vehicle movement, but the agreed planning purpose still required us to distinguish the point types. Relabelling them all as entrances would have made the dataset look complete without obtaining the evidence the label implied. I recommended a defined survey programme and a release account that retained the unfinished work.

We allocated ₹62 lakh to the programme with Peter's survey team and the qualified owners of the assignment. Twenty-nine entrance positions were verified within the defined review; nineteen remained pending. Those are outcome groups, separate from the original thirty approaches and eighteen centroids. We did not invent a cross-tabulation showing which original category supplied each completed point. The actual record would be needed to make that connection.

The resulting account distinguished verified entrance positions from unresolved locations. Verification of a point did not establish a right of access, suitability for every vehicle, permission to use a road or an absence of changing conditions. Those questions remained with the relevant owners and current evidence. We also made no measured claim about shorter journeys, fewer failed visits or additional revenue. The immediate consequence was the survey cost and a partial result whose nineteen pending items remained visible.

I would establish the semantic purpose of a location before treating coordinate precision as product quality. A very exact answer to the wrong spatial question can be more misleading than an openly incomplete one. The enterprise needed to decide whether it was selling a property reference, an approach or an entrance record, then obtain evidence appropriate to that offer. The customer should receive a clear account of the twenty-nine completed positions and the work still required, rather than a complete-looking file in which a precise point silently changes its meaning.