Method

Living people are omitted from this build entirely — not hidden, not gated, not present in the output. 10 people were removed on that ground before this site was generated.

One exception, and only one. Lynn Derrick Luwinski is named, with his birth year and his parents, and nothing else — no occupation, no address, no photograph, no descendants. An archive about a man's ancestry that would not say whose ancestry it was would be useless; an archive that published him would be a different kind of failure.

Where the source flags someone as living but they were born in 1920 or earlier, and so would now be 106 or more, they are presumed dead and published with the absence of a death record stated plainly. Several of those are Holocaust-era people whose fate was never recorded. Saying so is the entry's purpose.

A removal request is honoured within days, without argument and without requiring a reason.

A plausible number from a check that could not see the thing 21 September 2026

The worst failure this archive meets is not an error. It is a reasonable-looking answer about the wrong object, because nothing about it asks to be checked. Six instances landed in a single day, four of them inside an hour, and they share one shape: a test ran, found nothing, and reported the nothing as a fact about the world rather than a fact about itself.

• The scan that stopped before the name. The Le‑ block was measured on Schöneberg's registers and carried to Kreuzberg's without rechecking. Kreuzberg 1980 reported no Leipholz with its last word at Leipelt, and 1981 with its last at Leimbach — both of which sort before the name being looked for. The scan had never reached it.
• The volume with two alphabets. Kreuzberg 1982 is 226 pages where its neighbours are 125, and holds two registers. Reading it as one sequence covers half the year and calls the whole of it empty.
• The spelling the search could not contain. A scan for LEIBHOLZ read the 1975 Schöneberg index and declared it clean. Widening the pattern to Le[iy][bp]h found ERNA FRIDA ELSE LEIPHOLZ in the same volume on the first pass.
• The surname settled by its neighbours. A telephone book prints “Leibholz Erna”. She is a Leipholz, and no better scan shows it — the line sits between Leipert and Leipner, where Leibholz cannot be.
• The dates that were there all along. A visiting session reported this archive's atlas as carrying no dates. It has no date key; fifteen of its sixteen places carry a year in prose. The check looked for a field and reported a world.
• The index nobody fetched. The search page fetched a file that existed only between builds. Every gate passed, because no gate opened what the page asks for.

The rule that comes out of it is not “be careful”. It is that a negative must carry proof that the instrument reached the place. This archive now requires a scan to return the nearest name on each side of the target, and treats a result with an empty side as inconclusive rather than negative. A gate must derive what to check from what the page actually asks for, not from a list somebody maintains. And when several negatives share a property — an umlaut, a suffix, a blank parameter, a spelling — suspect the property before believing any of them.

It is the tidy negative that should frighten us 21 September 2026

A sister archive spent an afternoon on record searches that returned zero, and was one step from publishing a finding that explained all of its evidence and was built on nothing. Its own summary of what went wrong is better than the incident: noise looks like noise; it is the tidy negative that should frighten us.

Three failures here the same week had that same shape, and none of them looked like an error:

• A clean year-by-year scan of the 1942 death index, reporting no Leibholz. The volume has no text layer at all. The search was matching against an empty string and returning a tidy, orderly nothing.
• Four complete-looking register pages. The browser pane had quietly resized, so each page was captured half-height. A cropped page of a ruled table is indistinguishable from a whole one.
• A scan that stopped before the name. It reported an absence at Kreuzberg 1980 with its last surname at Leipelt — which sorts before the name being looked for.

Every one produced a result with internal structure. That is the tell, and it is the opposite of the intuition: a garbled answer gets checked, a neat one gets published. In each case the instrument was describing itself while appearing to describe the world.

And a healthy neighbour proves nothing. Throughout that archive's dead afternoon the image host kept serving plates perfectly, in the same browser. A day of successful image reads is not evidence that the search beside it is alive. Here, the ZLB page images and the ZLB text index are likewise separate instruments and can fail apart.

The cause turned out to be narrower than first reported and the correction is worth keeping too: the account was never signed out, one browser context was, and it recovered on its own. Each session holds its own cookies, so a probe run from a different session would have come back clean and proved nothing. A check on a shared service has to run in the same context as the work it is vouching for.

The decisive fact was in a third file 21 September 2026

Two sessions worked on this archive on the same morning and both got Georg Leibholz wrong, in opposite directions, within an hour of each other.

A visiting session read the status field — DOCUMENTED — but NOT a brother — and rendered exactly that. It was accurate when written and it is a summary of research at a moment, so the page stated the half that was in the data. This archive then corrected it with the newer finding: Sally Leibholz, father of the three, was also born at Hammerstein, so a Hammerstein birth no longer puts a man outside this family — and wrote that the reason for expecting Georg to be separate had gone.

That was wrong too, and this archive's own atlas said so. The Hammerstein pin reads “Georg born 1884 — which is why he cannot be Jacob's brother.” Georg was born 7 June 1884 and Jacob on 8 August 1884, two months apart, and two men born two months apart are not brothers. The point survives the one-year discrepancy in Georg's record, because 7 June 1885 still leaves ten months. He is excluded by calculation, which is a stronger and more final thing than an absence of proof: an absence invites somebody to go looking, a calculation closes it.

A visiting session reads the summary. A resident session reads the moment it happens to remember. The fact that settled it was in a third file that neither had opened. The resident failure is the worse of the two, because it was committed while explaining the visitor's.

The gate watched the pages and not the data Widened, 20 September 2026

check:living exists so that the promise about living people is tested against the built output rather than trusted to the generator. Until tonight it walked *.html only — and this site serves two JSON files that a reader's browser fetches directly: the atlas data behind the map, and the search index. A living person could have reached a reader through either without the build saying a word.

A peer session working across the shared tooling found the same blind spot and widened the kit's copy of the gate. This archive does not use the kit's copy; it runs its own tools/check-living.py, so the fix did not reach it and had to be made here.

Measured before it was changed: the leak was zero. A first crude search looked like three hits and every one was a published dead namesake — Joachim Luwinski born 1865, Sally Leibholz, Gertrud Leibke. The blind spot was real and the leak was not, which is the order those two things have to be established in.

And it was proved rather than assumed. A scratch copy of the build was given a served atlas-data.json naming a guarded living person, with nothing in the HTML. The old gate answered “ok” and exited zero. The new one named the file, the person and the matched text, and failed the build.

A common-name control cannot clear a rare-name negative 22 September 2026

Twelve Berlin telephone volumes were swept for the surname LEIPHOLZ. The search found it in four and nothing in eight, and the eight were false. The OCR prints the name Leiphotz, Lelpholz, Leipkslz and Leibholz — five of the eleven readable volumes mangle it, and a search for the correct spelling misses every one.

What licensed the error was a control that looked healthy the whole time: SCHMIDT, 133 to 191 hits a volume, in every book with a text layer. That proves the layer exists. It proves nothing about whether one broken glyph ate a rare string, and a volume can be richly indexed and still lose one line.

The instrument that works is the alphabet, because it does not depend on the glyph being right. The name falls in a known slot — Leipholz sorts between LEIPERT and LEIPIC — so the whole slot is printed and read, and an absence is recorded only when a name above it and a name below it are both in hand. A missing side is inconclusive, not negative. Eight unreliable zeros became six positives and two real absences, and two published claims turned out to be wrong.

A corollary found the same week, reading the Adressbuch: the slot method reproduced two counts already read at the image (seventeen householders in 1938, ten in 1940) and then failed predictably where the OCR loses the typographic dashes. A method that states where it stops is worth more than one that quietly under-reports.

The Landesarchiv’s own finder has a defect, and it is not consistent 22 September 2026

LABSA, the Berlin register finder, offers births, marriages and deaths as three checkboxes. Ticking more than one does not widen the search — it corrupts it:

Not the sum of anything, and not even reliably zero. A count from the combined form is not a count and a zero from it is not an absence. This archive’s existing work was then checked against the defect rather than assumed safe: the sixty-one Schöneberg death volumes it rests on are exactly the single-checkbox figure, so nothing published was disturbed.

Guessing at names is not enumerating them 23 September 2026

The Charlottenburg registry holdings were inventoried by querying the five office names the archive could think of: 143 volumes in five designations. The real figure is 207 in six. It had missed Standesamt Charlottenburg von Berlin entirely, because LABSA matches an office name exactly and the list of names is only obtainable from the form’s own autocomplete endpoint.

The same root produced a factual error on Kurt’s page, which said Jacob’s 1913 office was “one of Charlottenburg I to IV”. Charlottenburg IV did not exist in 1913 — its earliest volume of any kind is 1922. Correction 9500913.

A tenth of a catalogue went missing, and the hole was shaped like an umlaut 23 September 2026

With the office names finally in hand, the whole free holding was enumerated: 5,383 name-index volumes across 139 offices — 2,845 death, 1,595 marriage, 943 birth. That single list turns several jobs this archive had been costing by eye into finite ones: Reinickendorf’s remaining death indexes are thirteen volumes, and the whole of “she died in some other borough” is 296 volumes across 25 offices.

The first pass reported 4,838 volumes and said twenty-nine offices held nothing — and all twenty-nine have an umlaut or an eszett in the name. Schöneberg, Neukölln, Köpenick, Weißensee. The autocomplete hands names back double-encoded and the search needs them that way, but the results table prints them correctly encoded, so the loop queried in one form and filtered the returned rows on the other. Every query succeeded; every row was then discarded. A quiet, confident zero, twenty-nine times, worth 545 volumes.

It was caught by a control and by nothing else. Schöneberg has sixty-one death volumes — a figure this archive already rested on and had pulled by hand an hour earlier — and the new catalogue said zero. The fault was not in the archive’s instrument but in the code reading it, which is the harder kind to see: nothing errored, and 4,838 looked plausible.

What the slot method then bought 23 September 2026

Applied to Reinickendorf’s thirteen outstanding death-index volumes, read through pdf.js over HTTP range requests — 807 pages of 2,120 out of some 700 MB, nothing downloaded — it closed a borough: with the run already read to 1981, twenty-one consecutive years of Friedrich Leipholz’s own borough, 1974 to 1994, and his death is not registered there.

And the first pass of that sweep was discarded. It had tested whether each volume’s alphabet was open on both sides of LEIBHOLZ and then reported thirteen negatives covering LEIPHOLZ too. Those are two different slots. In the 1990 volume the nearest name above Leibholz is Leipe, which sorts below Leipholz — so that slot had never been shown to be read at all. The claim would have been true and unearned. Two volumes also failed mid-read with network errors and were not counted as absences: a failed fetch is not an absence.

The sweep’s one positive is on the Leibholz page — Marie Leipholz, the only Leipholz in all twenty-one years, dying in the borough Friedrich lived in.

And an exception, made deliberately 15 September 2026

Two more people are now named who may be living: Linda A Wear, born 1933, and Bruce W Wear, born 1936 — two of the six children of Henry Robert Wear and Martha Maud Castle. They were withheld for a few hours this evening under the rule below, and then restored at the family's direction.

The reasoning is worth setting down, because the rule is otherwise absolute. The sibship is the argument: what identifies Derrick's biological father is that the couple had exactly two sons and the younger was eleven years old in 1948. Withholding the younger son's name while relying on his existence and his age is a strange kind of privacy — it publishes the fact that matters about him and hides only the label.

So they appear as the General Register Office index has them and in no other way: a name, a quarter, a district. No addresses, no descendants, no dates beyond the quarter of registration, and nothing inferred about either of them. Neither is suggested to be anybody's father. If either of them, or anyone acting for them, would rather not be here, write and they come out the same day.

One living person is named, and that is all that is published about him

Ten living people are omitted from this archive entirely. The eleventh — Lynn Derrick Luwinski — is named, because the archive is about his ancestry and cannot be written without saying whose it is.

Named is the whole of it. No birth date, and — since 14 September 2026 — no birth year either. A year is a date of birth with the precision filed off, and paired with a full name it is the half of an identity that is worth stealing. It was removed from the pages, from the generated register, and from the builder that writes the register, so that a rebuild cannot quietly restore it.

Facts that gave the year away by arithmetic went with it: his age at his father's death, his age when the first divorce action was filed, the interval between the wedding and the birth. The events are still here — a child was born, and it was after the wedding — because they carry the argument. The dates attached to him are not, because they carried nothing but him.

A false claim of kinship, caught before it was sent Corrected

A request for Ivone's Porto death record had been drafted months ago and never sent. Read again before sending on 15 September 2026, it contained the line “sou neto dela” — I am her grandson.

It is not true. Ivone's son is Derrick. The researcher is the son of Derrick's former wife by her second marriage, and is no relation to Ivone at all — a fact his own DNA test states plainly, his father being that second husband. The sentence would have put a false statement of relationship in front of a government registry.

It was also unnecessary. A Portuguese certidão de óbito is a public document and needs no claim of kinship at all. The line now says only that the purpose is genealogical and non-commercial. Check who you are claiming to be before you claim it — an archive that writes to registries under a borrowed relationship is worth nothing, however good its intentions.

And the tool was corrected, so this page should not be read as a standing complaint 21 September 2026

The refusals above stand — those two particular folds are still wrong for this archive's content. But the tool that proposed them has since changed, and the change answers the objection. Where an archive spells a name several ways it now names the label it used and says it is only a label, and the map is applied to the search query and to the stored row alike, so every form finds every other and nothing claims any one of them is correct.

That matters for a distinction this archive leaned on hard on 20 September. A fold of this kind is a search rule with no concept of identity. It says these letters find those rows, not these people are those people — which is exactly the line between LEIPHOLZ reaching Willi and Jenny, who are ours to consider, and LEIPHOLZ reaching an eighteenth-century Protestant family at Orzysz-Arys in Masuria, who are certainly not. The variant list widens the search. It must never widen the family.