Correcting reading order without declaring the collection perfect
A 24,000-page collection had passed its file-completeness checks, but reviewers found search extracts that combined text from the wrong parts of a page.
A useful data system lets a reader see what the source supports and where interpretation begins. I would rather retain an honest uncertainty than make a record look complete by adding a fact nobody can substantiate.
A diagnostic review and a representative quality estimate serve different purposes. A useful finding should not become an unsupported percentage.
Records and data work
26 years
Career begins July 2000
Data colleagues
38
Three functional heads
Annual operating budget
₹9.2 crore
Separate ₹2.6 crore platform programme
Keep the uncertainty that belongs to the source
I lead data work at Nerzhavi Archive Software in Mysuru. We help institutional archives and specialist collections organise, search and inspect their digital records. My thirty-eight colleagues work through three heads, with a ₹9.2 crore annual operating budget and a separate ₹2.6 crore platform programme. Our responsibility is to make the software's treatment of records understandable. The collection owner decides historical interpretation, access and disposition; a convenient field in our product does not give us authority over those decisions.
I began as a digitisation analyst in 2000, checking the relationship between a scanned page and the text a user would search. Later roles brought me into descriptive metadata and the way a collection's meaning survives a move between systems. Since becoming CDO in January 2023, I have had to connect those details with product promises, client expectations and the cost of correcting a misleading result. Data leadership in this setting means asking what a transformation preserves, what it changes and who can explain the difference.
The cases on this site concern misordered text from complex page layouts, similar descriptions that did not necessarily identify duplicate records, dates that could be known only to a year or range, and a search-answer experiment that remained withheld after unsupported claims appeared. Each case starts with a particular collection or test. I resist turning a local correction into an assurance about every archive, because the material itself is often diverse and the consequences of a mistaken simplification are not evenly distributed.
52 pages exposed a reading-order problem
A deliberately mixed review of 520 pages from a 24,000-page collection found fifty-two with misordered text. We funded ₹46 lakh of correction and review work, preserving the scan as the source instead of presenting corrected search text as a replacement.
800 candidate pairs needed a human decision
A similarity review across 27,000 descriptions produced eight hundred candidate pairs. Collection specialists accepted 210 as duplicates and retained 590 as distinct. The result supported linked groups with provenance, not automatic erasure of apparently redundant records.
A year remained a year
Among eight thousand dated records, 2,400 supported an exact date, 3,100 only a year and 2,500 a range. We revised the model to preserve those distinctions instead of filling missing day and month fields with an invented first day of January.
Current focus
I am improving how a user moves from a search result to its underlying source and sees the certainty of each statement. That includes the record's origin, any transformation applied and the person or process responsible for a later interpretation.
Conversations I welcome
I am exploring a first external board contribution where the reliability and meaning of records materially affect the service. I can provisionally allocate eighteen days a year, subject to employer agreement, relationship review and the preparation the actual appointment requires.
Beyond the page
The record behind the perspective.
Formation
BSc in MathematicsRavsheli Science College · 2000
Nilofer kept the unsupported answers in the release discussion when the demonstration looked persuasive. Her recommendation made the cost of waiting explicit without pretending that the remaining work had already succeeded.
Deepak NorvethiChief Executive Officer, Nerzhavi Archive SoftwareThe full record ↗
Contact
I am exploring a first external board contribution where the reliability and meaning of records materially affect the service. I can provisionally allocate eighteen days a year, subject to employer agreement, relationship review and the preparation the actual appointment requires.
