The story of one find
How it all comes together
This page was written the day after the find. It is about why a single line in an old book took three months of work, software trained to read this particular handwriting, and an index of an archival collection built from scratch.
Worked example
Act no. 664
The Podolia Noble Deputies' Assembly, 17 December 1802. A book of protocols, one thousand seven hundred and fifty-seven pages of Polish cursive.
What it cost: more than anything else in the project. Three months, a finding aid built from scratch, two different handwriting-recognition models — and in the end a human eye at eight times magnification. What had to be cleared away, one obstacle at a time, is set out below.
Names, ages and formulas checked against the original at up to eight times magnification. What we could not read is flagged separately below.
“Marszałek Gubernski y Deputowani z powiatów wysłuchawszy prośby Urodzonych Jana y Gabriela braci rodzonych, Synów Teodora, a wnuków Michała Doliszczyńskich, w powiecie Bałtskim mieszkających…”
“The governorate marshal and the deputies of the districts, having heard the petition of the well-born Jan and Gabriel, brothers, sons of Teodor, grandsons of Michał Doliszczyński, resident in the district of Balta…”
Then the list, name by name
| Who | Age in 1802 | Born |
|---|---|---|
| Jan, son of Teodor × Teresa | 44 / 28 | ~1758 / ~1774 |
| — son Piotr | 11 | ~1791 |
| — son Domian | 7 | ~1795 |
| — son Grzegorz | 5 | ~1797 |
| — son Terenty | 3 | ~1799 |
| Gabriel, son of Teodor, widower | 41 | ~1761 |
| — son Alexander | 20 | ~1782 |
| — son Hipolit | 10 | ~1792 |
And two words that change the picture: “mieszkają w powiecie Bałtskim we wsi Lipowenkim, DUCHOWNI PRZY CERKWIACH” — they live in the village of Lypovenke, clergy at the churches.
And the next sheet carries the decision: “…Dowody za dostateczne uznawszy… do CZĘŚCI DRUGIEJ SZLACHECKIEJ KSIĘGI GUBERNI wnieść… Deklarowali” — the proofs found sufficient, the family entered into part two of the noble pedigree book of Podolia Governorate.
Among the proofs submitted is Jan's baptismal record, entered on 24 March 1758 — parents Teodor and Paraskeviya. It is the earliest mention of a document of this family known to us; we have not seen the record itself, it is known only from the list of proofs.
The name of the church we could not read. The word is broken by a line wrap, and our scan simply does not have the pixels for it: it reads either Osic[z]kiey — which would be Osychky in the same Balta district — or Orsanskiey, and no parish of that name could be found. The first reading is the more plausible by letter shapes, but chronology undercuts it: the earliest surviving book of Osychky begins in 1765, and the baptism was in 1758. The question stays open, and we do not put a convenient answer in its place.
And now — why nobody found it
The act lay in that book for two hundred and twenty-three years. Five obstacles stood between it and a reader, and each had to be cleared separately.
First, what “the machine read the book” means. To a computer, a scan of an archival sheet is just a picture. To search it you need software trained to make out handwriting: it goes down the page line by line and writes out what it believes is there.
Such software has long read printed books, but our documents are handwritten, in faded cursive, and on top of that in several languages and alphabets. So for this project the software had to be trained separately, on the hands of our own scribes. It reads with errors: roughly every third word comes out wrong. Its text is not a translation of the document but a draft for searching: it says which page to look at, and the reading itself is done by a person, against the original.
| Obstacle | How it was cleared | |
|---|---|---|
| 1 | The case was not on our disk, and the name index of the collection alone runs to over seven thousand files | an index of the collection assembled from seven sources: what exists at all, and where to get it |
| 2 | 1,757 pages of manuscript, not one line of them in typed text | software trained on this cursive read the book from beginning to end |
| 3 | The text is Polish, and the software had been trained on Cyrillic | a second one, trained on Latin script, had to be used; the first produced forty-two pages of nonsense on this book |
| 4 | The surname was searched in its Cyrillic spelling — such a search never sees the Latin “Doliszczyński” | a rule: in a Polish case, search twice — in Cyrillic and in Latin script |
| 5 | The surname is written with the archaic long “ſ”, and the machine returned “Dolechynski” | the researcher's eye at ×8 magnification: D-o-l-i-ſ-z-c-z-y-ń-s-k-i-c-h |
The Latin-script search scored this page 97 out of 100 for similarity to the surname being sought — the highest score of that whole day's work. The Cyrillic search never saw it at all: the same surname written in two alphabets is, to a machine, two different words.
Look at the fifth row. Four obstacles were cleared by technology, the fifth by a person. The machine honestly offered a candidate but read the surname wrongly; what settled it was an eye on a magnified piece of the scan. That is how it should be here: the machine finds, the person decides.
What one act changed
- Social estate. Not odnodvortsi but clergy at the churches — the same estate in which Hryhorii Dolishchynskyi, the assistant from the 1821 confession list, stood on the previous page
- Place. Not the Olhopil district but Balta, the village of Lypovenke
- Generations. The pedigree moved up to Michał. His age is not given in the act: he appears only as the grandfather — “wnukow Michała”
- Earliest mention. The baptismal record of 1758, named in the list of proofs
And immediately the next step
The moment the village had a name, the our register of archival files answered in a second: for Lypovenke the central archive lists three parish registers — 1766–1794, 1766–1797 and 1767–1803. And all three were already on our disk, seventy-four frames.
Those books cover the years when the children of both brothers were born — from Alexander (~1782) to Terenty (~1799). The machine has already read them: some four hundred births are recognisable in the volume, along with hundreds of marriages and burials. No person has looked into them yet — that is the next piece of work.
This is how it works: one find points straight at the next. Not “search somewhere sometime”, but particular books for particular years — named, located and machine-read the same week.
Honestly about the limits of this find
The act contains no Antoniy and no Kyrylo — the men from whom the Miastkivka line is known. Which means the connection between the family of Lypovenke and the family of Miastkivka is not proved by this document.
It is likely: the surname is the same, the district is next door, the dates fit. But likelihood is not proof, and until the link is found, two branches remain two branches.
The same goes for the second coincidence. Lypovenke in the Balta uyezd is the address on the cover of file 159, examined on the previous page. Village, district and estate all match; what follows from that for the pedigree still has to be established from the documents themselves.
This is worth saying plainly, because the temptation runs the other way. When after months of searching you finally see your surname on a sheet from 1802, the easiest thing in the world is to supply the missing link yourself. That is exactly how pedigrees appear in which everything fits and nothing is true.
Why this project exists
Now it is clear what all the rest is for. Not for volume, and not to “digitise as much as possible”.
To find one line it took: knowing the case existed; having its scans; having software that copes with Polish cursive; realising the surname would never turn up in Cyrillic; searching again in Latin script; and having an index that, given a village name, says at once which parish registers survive.
Each of these on its own is useless. An index without the ability to read is a list of what you cannot read. The ability to read without an index is a tool with no material. And searching without a human eye produced forty-two pages of nonsense and a misread surname.
One last thing. This act concerns one family — but the book it lies in has one thousand seven hundred and fifty-seven pages, and almost every one of them is about somebody else's. The machine has now read it, and in our working database you can already search it by surname. Bringing that search here, onto the site, is the project's next big task.
That is why Knyha Rodu is being built as something shared: work done for the sake of one surname opens the book just as much for everyone else recorded in it.
Here the chain of worked examples ends. What comes after is what it was all assembled for: to see which books exist for your village, and to try your own surname against what has already been transcribed.
Search the surname as it is spelled today, or just the beginning of it
Search for a surname →