How a claim earns its badge

The five grades

Every assertion on this site — a birth, a marriage, a parentage, a death — carries one of five grades, informed by the Genealogical Proof Standard — which governs how research is conducted and written up rather than defining grades of its own. The grade travels with the claim; nothing below Proven is presented as settled fact.

One qualifier turns up on a number of pages and is not a sixth grade: Probable–Strong. It marks a Probable claim whose compiled joint is corroborated from a second, independent direction — a répertoire naming a soldier's wife where a sighted act names the same woman, or a clerk's extract whose stated age lands exactly on a Proven birth date. It says the claim is unlikely to move; it does not say it has been seen. Everything that applies to Probable applies to it, including that nothing at this level is presented as settled.

What "sighted" means

A record is sighted when someone on this project has actually read it — the original document or a direct image of it — rather than relying on someone else's transcription or summary. A parish register entry is an original; a photograph or microfilm scan of that entry is a derivative of it, but still counts as sighted, since the content hasn't passed through anyone's retelling. A family group sheet, a published genealogy, or a database abstract is authored — someone has already done the reading and interpreting for you, which makes it useful as a lead and unreliable as a foundation.

Separately, information within a record is either primary (recorded by someone with direct knowledge, at or near the event) or secondary (recorded later, or at a remove). A death certificate is an original document, but the birthdate on it is usually secondary information — supplied by a grieving relative, from memory, decades after the fact. Original documents can and often do carry secondary information. Both distinctions matter, and both travel with every claim graded on this site.

Two copies of every act — and, sometimes, a copy of the copy

Quebec's parish registers were kept in duplicate. The curé wrote each act out twice: one copy stayed at the presbytère, the parish house, and the other was deposited each year with the civil authorities and now forms the registres d'état civil series at BAnQ — the greffe copy, a greffe being the registry office of a court. Both are originals, written out by the same priest within days of each other, and damage to one implies nothing about the other. When a band of one copy is unreadable, the remedy is to pull the other. This project has recovered a godfather's given name from one copy of a 1664 baptism and the godmother's surname from the other, each supplying what the other had lost.

What that leaves is a trap, and it took until 29 August 2026 to name it. Not every volume filed under the greffe label is that second original. A marriage act at Trois-Rivières, read that day, turned out to be neither copy but a nineteenth-century clerk's transcription of the register — and it announced itself only to someone looking for it: a uniform copperplate hand running across many years of entries, machine-ruled paper, sequential pagination in a series that should have none, and a spine reading registres photographiés au GREFFE where the volumes on either side of it read au PRESBYTÈRE. Under this site's standard a later clerk's transcription is compiled — whatever it is bound with, and however old the ink looks. It is someone else's reading of the act, which is the whole of what Probable means here.

The rule, adopted the same day: read the spine label and look at the hand before grading any image sighted. The label is what tells you to look; the hand is the test. A page of acts in several hands, crowded and irregular, each signed by the priest who performed it, is a register. A page of acts in one even hand is a copyist.

This is load-bearing rather than fastidious. It is the single reason the closest kinship this project has found — Philorum Oligny and Marie-Azilda Bessette, fourth cousins once removed — is graded Probable and not Proven. Seven of the eight joints on that descent are sighted originals. The eighth, a baptism at La Prairie in 1713, survives only in this form, and a chain is only as strong as its weakest joint.

Register 03 §3.75(f), standing rule adopted 29 Aug 2026 · Register 03 §3.32(iv) on the duplicate registers · Register 01 §3.32 for the application to the 1713 act · the copies compared are of the baptism of Jacquette Clément, Château-Richer, 4 Mar 1664

Why no generated image is ever a source here

On 31 July 2026 an image of a 1721 baptismal act from Chambly — faint, hard to read, and standing behind one of the more important findings on this site — was put through a generative model and asked to be restored. What came back was legible, in a convincing period hand, and correct on every name that mattered. Paul Laporte. Catherine Savary. Prudent Robert. Marie-Josèphe Besset. All four were later confirmed against the actual paper.

It is not a source, it has no source number, and it never will have one.

The reason isn't that it was wrong. It's that the same image rendered the child's own name as "Chuleverne" and the closing formula as "Lorineuce" — nonsense, invented, in the two places nobody happened to be looking. Nothing in the picture separates the parts it read from the parts it made up. Its accuracy was only knowable by going to the original, which means it added nothing the original didn't, and would have added error if the original hadn't been checked.

A thing that can be simultaneously correct and fabricated within a single line, with no internal marker telling you which is which, cannot enter an evidence chain — no matter how often it happens to land.

There's a further reason this is written down rather than just quietly avoided. Every previous mistake on this project came from a source that was visibly second-hand: a broadsheet, a printed répertoire, a website's narrative, someone's online tree, a memorial page. The kind of thing it was could always be seen at a glance, and that is exactly what made the corrections possible. A generated image doesn't announce itself. Filed in a dated folder six months later, it is indistinguishable from a scan. It is the first artefact this project has handled with no external tell, and the ordinary care that caught every earlier error would not have caught it.

It happened a second time, on a 1733 burial act, where the model produced precisely the sacramental formula being searched for. That is the trap in its purest form: a model filling illegible strokes will produce the most probable phrase for that position, and the phrase you are hoping for is the phrase it will invent. Accepting it would have "confirmed" a reading using an artefact generated out of the very ambiguity it was meant to settle.

So the rule, fixed:

• Permitted — contrast, levels, channel separation, sharpening, inversion, magnification. Operations on data already in the file: reversible, reproducible by anyone working from the same original.
• Permitted with care — asking a model to propose candidate letterforms for a difficult hand, where the output is a hypothesis about an image still in front of you and every proposal is checked against the actual pixels.
• Prohibited absolutely — any generated or in-painted image, in any capacity. Not as evidence, not as corroboration, not as a lead. Its output will always agree with whatever it was primed with, so it can never disconfirm anything, so it can never be evidence of anything.

The rejected file hasn't been deleted. It's been renamed so it can never be mistaken for a scan, and kept — on the same principle as everything else this project got wrong and left visible. The record of the near-miss is worth having.

Register 03 §3.48 and §3.63 · the underlying act was subsequently sighted as an original and all four names confirmed

Which documents this site can show you

Most of the acts behind these pages are read as digitised images in the Drouin Collection, and those images are licensed. This site is not free to republish them and does not. The restriction is worth stating precisely, because it is narrower than it sounds: it sits on the digitisation, not on the seventeenth- or nineteenth-century act underneath it. Anyone may say what a parish act says, and this site does, at length and word for word where it matters. What it may not do is hand you somebody else’s photograph of the page.

So the test for showing a document here is not how old it is. It is who made the copy in hand. Where the copy came from a commercial archive, the act is described and quoted and never shown. Where it was obtained directly from a state or federal agency — a certificate ordered from a city clerk, a case file pulled from a county court, a form released by the Social Security Administration under the Freedom of Information Act — the paper itself can go on the page. That happens to put the visible documents at the American end of both lines, which is also where this site’s narrative lands.

Two rules travel with any document published here. An identifier still usable to impersonate someone is removed from the image — a Social Security number, most often — and removed destructively, painted into the pixels before the file is written rather than covered with a box over live data, which is the ordinary way this goes wrong. And the removal is disclosed in the caption, because an unexplained blank on a site built around showing its work reads as concealment. Nothing else is retouched: a document is published with its stains, its scanner seams, its clerical pencil and its period language intact, and the caption says which marks are the applicant’s and which are a later hand’s.

Licensing position adopted 5 Sep 2026 · applied first to Roland Rock’s 1936 Form SS-5, Register 01 §3.26a · this section states a policy and grades no claim about the family

When the ceiling belongs to the record, not to the research

A grade of Probable usually names a missing document — go and find this act, and the badge moves. Occasionally it names a document that was never written. The difference matters, and a reader has no way to tell the two apart from the badge alone, so this site says which one it means.

Three men on these pages are identified as soldiers of the Carignan-Salières regiment: Jean Besset dit Brisetout, Mathurin Banlier dit Laperle, and Germain Gauthier dit Saint-Germain. All three are graded Probable–Strong rather than Proven, and each page gives the same reason: no act sighted by this project calls the man a soldier. Stated three times about three men, that reads like three separate shortfalls. It is one fact about the documents.

On 29 August 2026, twelve seventeenth-century marriage acts were read against a fixed checklist — at Notre-Dame de Québec, Notre-Dame de Montréal and La Prairie, every one of them the marriage of a man named on the compiled regimental lists to a fille du roi. Not one names a regiment, a company, or any military status for the groom. Not even miles, soldier, in the acts written in Latin. Twelve of twelve. A seventeenth-century Canadian marriage act records who a man's parents were, where he came from, and who stood beside him; it does not record what he did before he married.

So Probable–Strong here is a ceiling, not a stopping point. It is where the evidence ends rather than where the searching did, and going on looking for a marriage act that styles a groom a soldier is not a productive line of work — there is no such act to find. What raises these identifications as far as they go is a bridge of a different kind: a published regimental répertoire records each soldier's first wife, and the name it gives matches the wife a sighted act names. That is a match on two facts rather than a surname floated across a gap, and it is worth a great deal. It is not worth Proven.

What these acts do carry, now and then, is the military status of a witness rather than a groom — and that grades in three tiers. A named regiment (enseigne d'une compagnie du régiment de Carignan) is worth most; a named company (de la compagnie de Loubias) next, since it can be checked against a roll; a bare soldat is worth little, because a colonial soldier in the Québec of the 1660s is at least as likely as a regimental one, and the word alone cannot tell you which.

Register 03 §3.75(a) and (b), from twelve acts read against a fixed checklist on 29 Aug 2026 · the three-tier rule for witness attributions adopted the same day · applied on Catherine Fourrier and Jean-Baptiste Saurette · the finding is about the class of document, and does not move any individual grade

How a cousin degree is computed here

Cousin degrees on this site are computed, never estimated. That sounds like pedantry until you have published a wrong one, which this project has — see Corrections for the degree that had to be withdrawn and recounted.

The method is a set intersection. Trace every ancestor of each of the two people as far as the record goes; take the two lists and find the names on both; then, for each shared ancestor, count the generations back to them on each side. The pair of counts is the relationship — six generations on one side and five on the other means fourth cousins once removed — and the closest pair in the whole intersection is the answer to "how were they related." One further distinction decides how thick that relationship is: if both lines run back to an ancestral couple, it is a whole relationship; if they run back to one person by two different partners, it is a half one — just as real, and half as thick.

The obvious approach, asking the genealogy software for every relationship between two people, does not work on a tree like this one. It was tried; it timed out, and then failed outright. These are endogamous parishes — a population where the same few families intermarry generation after generation, as the St. Lawrence valley did for two centuries — so the number of distinct paths between any two of their descendants grows explosively, and an algorithm that walks the paths one by one never finishes. Intersecting the two ancestor sets sidesteps the problem entirely, because a set does not care how many routes lead to a name in it.

The part worth trusting is that the method was validated before it was used. On 29 August 2026, before computing anything new, it was re-run against the 1912 couple — whose figures this project had already recorded weeks earlier. It returned 842 ancestors traced for Joseph Philorum Oligny, 984 for Marie-Anna Saurette, and 103 individuals in common across 50 couples: the 2 August figures to the digit. It then independently reproduced the corrected Doucinet position from 14 August — the exact count this site had already had to publish a correction about. A method that reproduces both the old count and the old correction can be trusted with a new one. That run is what several pages here cite as the set-intersection re-run of 29 Aug 2026.

One caveat travels with every figure of this kind. The ancestor sets are drawn from this project's own research database — a GEDCOM file, the standard export format for genealogical data, which held 22,299 individuals in 10,626 families when these sets were computed. A computed degree is therefore an exact statement about that tree, and not by itself a statement about the documents underneath it. The grade on the descent is what governs what a page may claim: where one joint along the way is compiled, the kinship is Probable however cleanly the arithmetic runs.

Register 01 §3.32, method validated before use, 29 Aug 2026 · Common Ancestors §1 (the Doucinet correction, 14 Aug 2026) and §5 · the file is tree1-working-master.ged, which held 22,299 individuals in 10,626 families when these sets were computed · a computed degree is graded no higher than the descent it runs along

A note on "S244" and "R2 §3.31"

Some citation lines on this site carry a short code alongside the plain description — something like S244, or R2 §3.31. These point into this project's own working research log: a set of Evidence Registers, kept separately, where every source is entered, numbered, and cross-referenced as it's found. S numbers a single source; R with a section number points to the register's own running narrative of how a finding came together, including the wrong turns along the way. There are four registers, and each numbers its own sources starting from S1 — so S244 in one register and S244 in another are two unrelated documents that happen to share a number. Where a citation on this site points into a specific register rather than the one its own page's material mostly comes from, it carries that register's own prefix instead — R1-S244, R2-S244 — so the two never get read as the same source.

That log isn't published, so those codes aren't a link you can follow — they won't resolve to anything if you go looking. What they're meant to signal is narrower and more honest than a citation you can click: that the claim they're attached to has a fuller paper trail behind it than fits on this page, tracked somewhere specific rather than asserted from memory. The description that comes with each code — the document, the date, the archive — is the actual citation, and it stands on its own whether or not the code is there.

Where the records come from

This section lists the actual archives, published répertoires, and scholarly works this site draws on — organized by kind, since that determines how each is graded rather than what it happens to be about. A source appearing here doesn't mean every claim it touches is Proven; see the five grades above for what each one is doing on a given page.

Parish, civil, and judicial archives

Most of the parish acts behind this site — baptisms, marriages, burials — come from the Drouin Collection (Institut Généalogique Drouin) and from PRDH-IGD (Programme de recherche en démographie historique, Université de Montréal), which together cover Quebec's Catholic parish registers back into the seventeenth century. Notarial acts — marriage contracts, estate inventories, land transactions — and court records come from BAnQ, the Bibliothèque et Archives nationales du Québec, including the Événements de 1837-1838 rebellion files (fonds E17) behind the material on Isaac Oligny's own file. British military and colonial administrative records — militia rolls, governors' correspondence — come from Library and Archives Canada (the former National Archives of Canada, RG-8 and MG-24 series). American vital records, federal and Quebec census sheets, naturalization files, and Holyoke-area newspaper obituaries fill in the nineteenth- and twentieth-century end of the line, once the families crossed into New England.

Printed marriage répertoires

Several parish-by-parish marriage indexes, compiled by genealogical societies from the original registers, are used throughout: the répertoire of Saint-Joseph-de-Chambly (1706–1964, comp. Irénée Jetté), of l'Acadie and St-Luc (1785 and 1801, St-Jean county), and of the county of Napierville, covering St-Rémi and its neighboring parishes. A 1969 typescript répertoire of St-Jean-Baptiste-de-Rouville marriages is the source behind the marriage-pool ratio table. These are compiled indexes, not the registers themselves — useful for finding an act and for population-level counting, graded no higher than Probable on their own, and always superseded by the original when one is sighted.

Scholarly and reference works

One author carries most of what this site knows about the eight women before they reached New France, and that is worth saying once as a fact about the site rather than page by page. Peter J. Gagné's two standard references — King's Daughters and Founding Mothers, on the filles du roi of 1663–1673, and Before the King's Daughters: The Filles à Marier, 1634–1662, on the women who crossed before the Crown program — are the fullest published biographies of these women that exist, and six of the eight documented at Shared Ancestors cite him on their own pages for a French parish, a set of parents, a ship or an emigration story that no act read by this project supplies. His entries are compiled from French parish and notarial records that his footnotes name and this project has not read. That is what caps them: however good the book, a claim that rests on Gagné alone is graded Probable here, and the badge moves only when the act underneath is sighted. He has been checked against an act in three places, and the results are the measure of him. Catherine Fourrier's burial — the sighted act gives exactly the dates his account gives, to the day. Suzanne Bouvier's baptism — the sighted act's own dated headers put it at 24 June 1665 where his book has 1664, day and month agreeing and the year one off; it is logged in the registers as a correction to a published source, and the chart carries the act's year. Françoise Duval's parish and parents — the sighted marriage act and his account disagree on both, and her page leaves it standing as an open discrepancy rather than choosing. And where he hedges, the site keeps his hedge: Marie-Louise Lecoutre's origins are “unknown” in his own words, Mathurine Desbordes' parents are unknown even to him, and Jeanne Mechin's parentage is his hedge before it is this site's. Marc-André Bédard's Les protestants en Nouvelle-France supplied the Huguenot record behind Marguerite Doucinet's own origin. Pierre-Georges Roy's Hommes et choses du fort Saint-Fréderic and Cyrille Gélinas's The Role of Fort Chambly in the Development of New France provide military and settlement context for the Chambly-area generations. For the War of 1812 and 1838 Rebellion material on Isaac Oligny's page: Martin F. Auger's "French Canadian Participation in the War of 1812" (Canadian Military History, 2001), Michelle Guitard's Histoire Sociale des Miliciens de la Bataille de la Châteauguay, Luc Lépine's Les Officiers de Milice du Bas-Canada, 1812–1815, and Rev. J. Douglas Borthwick's History of the Montreal Prison from A.D. 1784 to A.D. 1886, compiled by the prison's own chaplain from its admission registers. These are authored secondary works, cited as themselves rather than as though their sources had been read directly — a distinction that matters most where one of them turns out to be wrong, as a few have.

Compiled trees and genealogical platforms

Steve's own research database, built over roughly twenty years and exported as a GEDCOM file, underlies a great deal of this site's early-stage findings — it's where a shared ancestor is first noticed, not where the claim stops. FamilySearch and Ancestry.com turn up leads regularly and host digitized images of the primary records above; neither is treated as a source in its own right until whatever it points to has actually been read. The same goes for Find a Grave and other user-contributed sites: useful for a lead, occasionally right about a detail no formal record gives, never a foundation on its own.

When something turns out to be wrong

Genealogical research corrects itself constantly, and that's not a flaw in the method — it's the method working. When a better source overturns an earlier claim, the earlier claim isn't deleted. It's marked as withdrawn and left visible, because the correction itself is part of an honest record: what was believed, on what evidence, and what changed it. A site that only ever shows its current best guess, with no trace of what it used to say, is easier to trust than it deserves to be. See Corrections for the record of what's been overturned or upgraded on this site, including a few of its own same-day mistakes.

If a claim on this site is graded Probable or lower, that grade is an invitation, not an apology — almost always it names exactly what document would settle the question, and where no such document was ever written, the page says so. If you're a cousin with a photograph, a family Bible, or a document tucked away in a drawer, it might be the piece that moves a badge from Probable to Proven.