Start with the ordinary delay
A dated examination of real-time systems in technical cultural records, and the operating choices that shape operating limits. The subject is easiest to see in a software-dependent cultural work being documented for display, acquisition and later reconstruction. A small error at that point does not stay small: it creates another message, another check, another journey or another version. Across 2,275 monthly records, even a single unnecessary return step becomes material operating work.
The 2020 files linked to this note are Meridian Road Services, Alder Civic Logistics. They are used as operating context, not as evidence for a market-wide claim. The purpose of the article is to ask how real-time systems can improve operating limits when source material, model dependencies, display configurations and object records have to remain usable despite changing technical dependencies and incomplete conventions.
PTD takes a deliberately strong position. Waiting that can be identified, classified and routed should be engineered out. Repeated judgment should be converted into a rule when the evidence is stable. Machines should inspect, reconcile, forecast and prepare the next action at a scale no small operating team can sustain manually. People should retain authority over safety, rights, capital and exceptions whose cost is genuinely asymmetric.
External evidence, not analogy
PREMIS 2.2 offers a durable starting structure through five connected entities: intellectual entity, object, event, rights and agent. It provides a way to record technical change without reducing a cultural work to one file.
The Library of Congress digital-format program asks which dependencies and technical characteristics affect long-term usability. That makes source inventories, format identification, fixity and event history part of the object record rather than later conservation notes.
The technosolutionist position is direct: hashes, timestamps, dependency manifests, display profiles and change events should be captured by the system at creation time. Human review can then concentrate on authorship, permissions, interpretation and conservation choices.
The arithmetic of a smaller queue
Consider an illustrative operation processing 2,275 items a month. This is a model, not a PTD company result. At a return rate of 12%, approximately 273 items re-enter the queue. If clearer data, automatic validation and guided decisions reduce the rate to 10%, the queue avoids about 45 returns each month.
At 14 minutes for each returned item, that change releases roughly 126 staff hours a year. At an illustrative loaded cost of €69 per hour, the direct capacity value is approximately €8,694. The more important gain may be outside payroll: shorter customer delay, fewer repeat visits, earlier rights clearance or less time with a physical asset unavailable.
The arithmetic is intentionally transparent: volume × change in return rate × handling time. It can be challenged line by line. A project should not be approved because the result sounds large. It should proceed only if a baseline sample of at least 220 records confirms the volume, return reason and time per return, and if quality or safety does not deteriorate when the step is removed.
Build the operating layer
The object record should be generated alongside the work: content hashes, source inventory, software and runtime identifiers, operator events, output versions, display dependencies, rights and preservation actions. The goal is a reconstructable chain, not an assertion of permanence.
The data model should be smaller than the organisation chart. Identity, state, evidence, owner and next action are the core. Everything else must justify its collection cost. A field that does not change a decision, satisfy a legal requirement or improve later learning should not be mandatory merely because software can store it.
Automation then operates in layers. First, validate format and completeness. Second, reconcile the item with the current master record. Third, propose a classification or next action. Fourth, execute low-risk routing inside a defined permission. Fifth, measure the result against the accepted baseline. This sequence prevents an automated recommendation from becoming an unexplained fact in the operating ledger.
Test before scaling
A credible deployment can begin in 9 weeks. Weeks one and two freeze definitions and sample the current process. Weeks three and four instrument waiting, rework and error states without changing the workflow. The next four weeks run the new rule on a limited cohort while keeping a control group or the preceding comparable period.
The release decision should use five measures: first-pass completion, elapsed time, staff time, error severity and the share of records requiring manual override. A faster system that creates more severe errors fails. A slower pilot can still pass if it exposes hidden work and produces a baseline the organisation did not previously have.
Scaling requires a second gate. The system must show that performance survives a different team, location or catalogue; that its fallback works during disconnection or source failure; and that operators can explain why an automated recommendation was accepted. Only then should the same architecture receive more volume.
Capital follows verified capacity
For an owner, a program built around real-time systems becomes investable when it changes repeat economics rather than creating a one-off presentation. The relevant evidence is a sustained reduction in avoidable work, a shorter asset or catalogue cycle, more capacity at the same fixed cost, or a lower requirement for corrective capital. Revenue growth is useful, but operating leverage is the stronger test.
Capital should be staged around proof. The first amount buys measurement and integration. The second buys deployment to a wider operating perimeter. The third buys automation or physical capacity only after the constraint has moved. This avoids installing technology around an unexamined process and then calling the resulting complexity transformation.
The same logic explains why PTD holds companies for several years. Data quality, work design and maintenance intervals improve through repeated cycles. The value is cumulative: each corrected identifier, accepted definition and captured exception reduces the cost of the next decision. A short ownership horizon can harvest a software launch; it is less suited to building an operating memory.
A clear next decision
The Practical Limits of Real-Time Operations: Special Projects Note 2020 argues for a narrow but forceful conclusion. Everyday efficiency is not an abstract social objective. It is the sum of queue length, elapsed time, first-pass completion, asset availability, energy and material consumed, and the number of decisions that reach the right person with enough evidence.
Technology should be expected to improve those quantities. The burden of proof belongs in the implementation: a frozen denominator, a dated baseline, a visible error budget and a comparison that survives a change of team. When the result is weak, the system should say so. When it is strong, the organisation should scale it quickly and redirect people toward work that requires judgment rather than repetition.
This position is recorded at 2020-06-13. The next review should replace the illustrative figures with an accepted cohort, retain the original model and show which assumption changed. That is how a technosolutionist thesis becomes an operating record instead of a slogan.
Sources and data notes
- Library of Congress — PREMIS Data Dictionary for Preservation Metadata, Version 2.22012 · Reference for a practical preservation record built around objects, events, rights, agents and the intellectual entity being preserved. This 2012 version is used for issues published before PREMIS 3.0.
- Library of Congress — Sustainability of Digital Formats2004 · Reference for evaluating formats, technical dependencies and the information needed to manage digital content through its lifecycle.
- Google — Site Reliability Engineering2016 · Reference for service-level objectives, eliminating repetitive toil, monitoring, automation, testing and learning from failure.
- World Intellectual Property Organization — Electronic Commerce and Copyright — Creation Identifiers and Metadata1998 · Reference for the role of identifiers and the relationships between descriptive and rights metadata in licensing, transactions and protection.