Where we were wrong.

What we said. That a human-rights analysis of Massachusetts “family court cases found that in 21 of 39 reviewed cases (54 percent), judges granted physical custody to fathers with documented histories of intimate partner violence”. No year was given.

What we got wrong. Three things. The study is Silverman and colleagues, 2004, and we never said so. It is not a review of 39 court files — its own abstract says “Battered women involved in child custody litigation in Massachusetts (n = 39) were interviewed”. And the 21-of-39 figure covers one or more state actors granting or recommending custody, a category wider than judges alone. We could not obtain the full text to confirm the narrower reading, so the page now makes the weaker claim, which cannot overstate.

The caveat that was missing entirely. The 39 women were recruited through domestic violence services. These are proportions within a group who already believed the court had failed them, not the share of Massachusetts custody cases that go this way. Published without that, 54 percent reads as the second thing.

How long it was live. Since the page was published.

What we did. We re-derived every comparative claim on the site against the figures printed beside it.

What we found in Ohio. The page named Mahoning County as the cheapest at $250 and Delaware County as the dearest at $485, then called that “more than triple the price”. It is 1.9 times. The same sentence said “same paperwork” while comparing a plain divorce to a divorce with children. A second claim, “more than twice as much”, was arithmetically right but compared a filing without children in one county to a filing with children in another — two different cases, not one case in two places.

What we found in Louisiana. The page said “the same divorce can cost three times as much one parish over”, reaching three by comparing $200 in Tangipahoa to $600 in Jefferson. The sentence immediately before it says the $600 figure is a divorce filed together with a rule and a TRO. So it was not the same divorce.

What we changed. Both now compare like with like and name both figures. Ohio: $200 in Cuyahoga against $385 in Delaware, both without children — nearly twice. Louisiana: $200.00 in Tangipahoa against $475.00 in West Baton Rouge — about two and a half times. The wider Ohio pair is still shown, labeled as two different cases. A Louisiana curator fee described as “more than triple the cost of the entire filing” now names which filing cost it is triple of.

How long they were live. Since each page was published. We have not tried to establish a shorter window, because we cannot show either sentence was ever right.

The rule we are adopting. A comparison must name both numbers and both must describe the same thing. Every multiplier on a cost page is now checked to sit within sight of the two figures it is derived from.

What we did. We swept all 861 pages for stated ratios, multipliers and comparisons, and re-derived each one we could check against the numbers printed beside it.

What we found. The custody page counts the states twice. A summary tile near the top said “Presume nothing: 24”. The detailed section below said “Nothing is presumed, in either direction: 26 states” — with the same explanatory sentence attached to both. Same category, two numbers.

What we changed. The detailed section divides the country five ways — 5, 3, 9, 7 and 26 — and adds to exactly 50. The summary tile listed three of those five categories and added to 38. The detail is complete and consistent; the tile was wrong. Corrected to 26. A check now runs on that page that fails if the five categories stop adding to 50.

How long it was live. The page was published on 21 August 2026 and we cannot show the tile was ever right, so assume the whole time.

What we did. We re-read the original sources behind the statistics on this page, rather than the summaries of them.

What we found. The page said intentionally false allegations were identified in 12% of maltreatment investigations involving a custody dispute against 3% without — “four times the rate, and still one in eight”. Both figures are exactly right. They are also from the 1998 Canadian Incidence Study, and the page gave no year and used the present tense.

What we changed. The 2008 cycle of the same study, published by the Department of Justice Canada, reports malicious reports in 10% of investigations noting a custody dispute against 5% where none was noted. Roughly double, not four times; about one in ten, not one in eight. Both cycles now appear, with their years and their source. The direction of the finding is unchanged and both cycles agree on it.

Why this one matters more than its size. This page is read by people who have just been accused. “Four times” overstates the thing such a reader most wants to be true, which is the direction of error to be most careful about.

How long it was live. Since the page was published on 4 September 2026.

What we did. We audited the destination of every citation link on all 861 pages — 2,098 of them — and sorted them by who publishes the page at the other end.

What we found. 229 links across 54 pages went to a commercial republisher rather than to an official source: 148 to Justia, 60 to FindLaw, and the rest to WomensLaw, DivorceNet and LawServer. The text those pages carry is, as far as we can tell, accurate. The problem is the promise printed above it. This site says every figure is traced to a primary source. Of a link to a republisher, that was not true.

What we changed. Each of the 102 unique URLs behind those links was researched against the official publisher — the state legislature or judiciary, the eCFR, uscode.house.gov, or the Library of Congress’s scan of the U.S. Reports — and the replacement was opened and read before it was accepted. 121 links have been repointed at the official source. No wording changed; only the destination of a link moved.

What is still wrong. 121 links remain on a republisher, and they are the same short list that blocks everything else here. Georgia, Tennessee, Oklahoma, Mississippi and Alabama license their codes to a commercial publisher, so there is no free official text to point at. Indiana and New Mexico serve a JavaScript shell that returns nothing to a plain request. And roughly a dozen pre-2000 state appellate decisions have no free official archive at all. We would rather say that than pretend.

What we did. The fifty “Divorce in [state]: what to expect” pages were compiled on 31 August and published on 2 September. On 3 September we re-read every row of every page against the statute or court source it cites — 350 rows in all. The pages had been live for one day.

What we found. 58 rows across 35 states were wrong, overstated, or cited the wrong section, and have been corrected. A further 84 rows that had said “we could not confirm this at a primary source” have now been filled from the official source — mostly filing fees, waiting periods and the alimony statute. Eleven of the “what is unusual about this state” notes were rewritten. In sixteen states our tools could not reach the official statute site for some rows; those rows are marked not re-checked on the page and still rest on the 31 August reading, and the page’s “sources last checked” date was left at 31 August rather than moved.

The ones that could have changed a decision: Montana said there was no post-filing waiting period; a decree cannot be entered until 21 days after service (MCA 40-4-105(3)). Pennsylvania gave the unilateral separation period as two years; it has been one year since December 2016 (23 Pa.C.S. § 3301(d)). Maryland said the court “does not physically transfer title”; Family Law § 8-205(a) expressly lets it. Utah said “3 months” residence; the statute says 90 days in the county of filing (§ 81-4-402). Iowa cited a section repealed in 2005; the rule now sits in § 598.5(1)(k). Florida still listed “permanent” alimony as a type; the 2023 reform removed it from § 61.08. West Virginia said there was no custody presumption; since 2022 § 48-9-102a presumes equal (50-50) time. New Hampshire omitted that RSA 458:16-a presumes an equal division of property. Wyoming described a “fault” ground; the state has none — the only ground besides irreconcilable differences is incurable insanity.

The rest, by state: Arizona — the custody “presumption” was actually the § 25-403.03 bar on joint decision-making where there is significant domestic violence; restated. Delaware — the six-month separation rule and its misconduct exception are in the definition at § 1503(8), not where cited; note rewritten. Hawaii — four no-fault grounds, not one. Idaho — the “substantially equal” rule is statutory (§ 32-712), the five-year separation ground is § 32-610 and not fault, and § 32-717B(4) presumes joint custody. Illinois — the “equal parenting time” presumption was misquoted; restated from §§ 602.5 and 602.7. Kansas — the note wrongly said Kentucky divides separate property; it does not. Kentucky — the 180 days run before filing, not before the decree. Louisiana — the separation periods are in art. 103.1, and the protective-order ground added in 2018 was missing. Maine — residency alternatives misdescribed; the 60-day minimum is in M.R. Civ. P. 113; the guardianship ground was missing. Massachusetts — residency wording and the section for the imprisonment ground; note rewritten. Mississippi — no “both spouses resident” waiver exists in § 93-5-5; irreconcilable differences is not joint-complaint-only; the statute’s no-presumption rule names only the mother. Missouri — the statute does not “expressly” disclaim community property. Nebraska — the three-step division test is case law, not § 42-365; the 60-day bar is § 42-363. Nevada — the two-year insanity ground was missing. New Jersey — the 18-month separation ground was missing. New Mexico — residency cited to the grounds section. New York — the two separation-based grounds, § 170(5) and (6), were missing. North Carolina — one year’s separation is not the “sole” ground; § 50-5.1 (incurable insanity) exists. North Dakota — residency cited to the wrong section and applies to the plaintiff. Ohio and Oklahoma — fault-ground lists were incomplete (two and three grounds missing). Oregon — the “married in Oregon” exception was misread; it applies only to certain grounds. Rhode Island — the fault list omitted three of the statute’s eight causes and the three-year separation ground. Utah — two statutory grounds were missing. Vermont — both residency rules are in § 592(a) and are statutory, not “noted by some sources”. Wisconsin — residency was cited to the grounds section. Alaska — the note hedged the residency rule as “approximate”; the court’s own guidance is definite: no minimum period.

Added 4 September. Connecticut’s own statute site refuses automated readers, so on 3 September all seven of its rows were left marked not re-checked. They have now been read by hand in the General Assembly’s published text, and Connecticut was the worst page on the site — four errors on one page, live for two days. An entire no-fault ground was missing: living apart by reason of incompatibility for a continuous period of at least eighteen months (Conn. Gen. Stat. § 46b-40(c)(2)). The waiting-period row said no post-filing waiting period had been found, when there are three separate clocks — 30 days after the return date before a default judgment on personal service, 60 days and a hearing where service was made another way, and no trial of a contested divorce until 90 days after the return date (§ 46b-67(a)). The custody row said “up to 16 statutory factors” and no presumption; there are seventeen, and there is a joint-custody presumption — it just only applies where the parents have already agreed to joint custody (§§ 46b-56(c), 46b-56a(b)). The filing fee and the alimony factors, both previously blank, are now stated: $360 under § 52-259(a), and the thirteen factors in § 46b-82. Thirteen other states still carry rows marked not re-checked for the same reason.

What caused it. Two things. The pages were built from a research pass that relied on secondary legal sites for section numbers, which is where most of the wrong citations came from — seventy such links have been removed and replaced with the official source or with no link at all. And several rows were paraphrased from memory of the rule rather than from the text; the overstated presumptions and the missing grounds are all that kind of error. The fix for both is the same: the row is now the statute’s wording or a close paraphrase of it, with the section that actually contains it.

What was wrong, part one. Our page builders truncated quoted material at fixed character counts, so sentences stopped mid-word. The prenup table was worst: 31 of its 45 state cells were cut at exactly 400 characters, losing the substance. New York’s cell stopped at “Second, GOL § 5-311 caps what y” — cutting off the entire explanation of what you cannot waive.

All 31 prenup cells have been restored in full from the underlying research, and the cap has been removed from the builder so it cannot recur. A further 89 truncated passages across 33 divorce-kit and cost pages have been restored the same way.

Still outstanding: about 33 truncated passages remain, mostly on the child-custody pages. Their original research data is not available to us in this environment, so we cannot restore the full wording yet. We would rather record that here than quietly leave it.

What was wrong, part two. Internal research notes had been published as prose on sixteen pages — first-person voice and raw field names. Readers of the Wyoming cost page saw “THE TASK PREMISE THAT WYOMING IS ‘ONE OF THE CHEAPEST STATES’ IS CONTRADICTED by the statute I read.” The Hawaii page said “headline_fee is null and headline_confirmed is false.” South Dakota had “NEGATIVE FINDING — the task asked me to confirm…”

The content of those notes was right, and is exactly what this site is for — Wyoming really is contradicted by its own statute, and Hawaii really does publish no total. Thirty of them have been rewritten in plain language, in our own voice, keeping every finding intact.

Why it matters: when an honest disclosure is written in the voice of an internal memo, it reads as an unfinished draft rather than a deliberate statement. That undermines the one thing this site is actually for.

What we changed so this gets caught next time: the weekly automated audit now also checks for visible markup, text cut off mid-word, internal research vocabulary in published prose, and numeric claims that disagree across pages. All four of tonight’s defect classes were invisible to us until we went looking.

What was wrong: A full copy audit of all 502 pages found three classes of defect, all of them ours.

1. Visible markup. On 63 pages, 122 places, a double-escaped HTML entity was printing as literal text mid-sentence — readers saw the characters “—” where a dash belonged. Worst on the alimony and protective-order pages. Fixed everywhere.

2. Pages contradicting each other about Alabama. When we resolved Alabama’s alimony tax position we updated some pages and not others. The money page still listed Alabama as unestablished while its own headline said only Maine was, its list of “36 states that conform” contained 35 names, and the Alabama page’s summary card contradicted the body of the same page. All reconciled.

3. Text cut off mid-word. Our page builder truncated some quoted material at fixed character counts, so sentences stopped dead. Nine instances on state pages have been re-emitted from the source, cutting at a sentence or word boundary instead. The prenups page still has about twenty-five of these and is not yet fixed — it is a different builder and we would rather say so than quietly leave it.

Also fixed: raw research notes had leaked into published prose on two pages. The Georgia money page carried an internal note reading “this session could not retrieve verbatim opinion text” where a statute quote belongs; the Wyoming child-representation page printed an internal field name, “see notable”, as if it were English. Both rewritten as plain statements of what we do and do not have.

Why it matters: this site asks readers to trust that when it says “we could not establish this,” that is a deliberate, checked statement. Leftover research notes make honest disclosures look like unfinished drafts, which corrodes the one thing the site is actually for.

What was out of date: Our stepfamily page described Papernow’s framework using her published vocabulary of stages. We wrote to her to check we had represented her work correctly. She replied that she has shifted the language from “stages” to “stations”, because most stepfamilies do not progress linearly and many move back and forth between them. In her words: “Becoming a stepfamily is a process not an event.”

What she confirmed: that our reading of the four-to-seven-year figure was right — “there is no set number of years” — and that “the findings you did pick out are absolutely correct and have been re-affirmed and clarified.” She also reports substantially more quantitative and longitudinal research since the work we cited, and a second edition of her book due November 2026.

What we added, and how we labeled it: she pointed us at stepmother-specific findings we had not covered, including a study in which most stepmothers expected to take a maternal role and no stepfathers did, and a qualitative study in which nineteen of twenty stepdaughters with good stepmother relationships said the role was not mother-like. She named some of these from memory. We have published them as correspondence rather than as citations, said so on the page, and will replace them with sourced entries as we verify each one.

Why this is here rather than buried: a researcher revising her own terminology is exactly the kind of update that never reaches the people still repeating her numbers. That is the whole problem this site exists to address, and it would be absurd to hit it ourselves and not say so.

What was wrong: All fifty state alimony pages described only the federal tax rule — that alimony under a post-2018 instrument is not deductible and not taxable — with no mention of state treatment. Federally that is correct. But a reader on the California, New York, New Jersey or Arkansas page would reasonably take it as the whole answer, and in those four states it is the opposite of the state answer. Alimony that is invisible on the federal return still moves money on the state one.

Why it matters: a settlement modeled on federal treatment alone is mispriced in those four states. The payer may have a state deduction they were not told about; the recipient may owe state tax on income the IRS does not see.

What changed: every one of the fifty pages now carries a state-specific paragraph. The four decoupled states get the mechanism and the citation. The thirty-five that conform say so, so the reader knows the federal answer is the whole answer rather than assuming it. The nine with no income tax say that. Alabama and Maine say plainly that we could not establish it.

Why we made it: the tax block was written once and reused across all fifty pages, which is the right way to keep a federal rule consistent and the wrong way to handle a rule that turns out to vary by state. Found while auditing the older pages against the fifty-state conformity survey.

Live for approximately one week before correction.

What was wrong: The page quoted the FTC’s February 2023 Data Spotlight as its most recent romance-scam figures — $1.3 billion in 2022, median loss $4,400. The FTC has since published 2023 figures: 64,003 reports, $1.14 billion, and a median individual loss of $2,000. The median we were quoting was more than double the current one.

Why: The 2023 numbers were published in an FTC business-guidance blog post rather than in a Data Spotlight or the Sentinel Data Book, so they did not surface where we were looking. We have also added what happened after: the 2024 Data Book does not break romance out at all, and the only 2025 romance figure the FTC has published covers social-media-originating scams only and carries a footnote that reports were not collected during the 2025 government shutdown. We now say that rather than reaching for a number.

Live for approximately four months before correction.

What was wrong: Not an incorrect figure, but a missing one. The page carried CDC’s 2024 provisional case counts without saying which way they were moving. All three major reportable STIs are now falling — syphilis down 22 percent and chlamydia down 8 percent for the second consecutive year, gonorrhoea down 10 percent for the third. Presenting large absolute numbers with no trend implies a rise.

Why: We had read the surveillance tables and not the announcement that accompanied them. Corrected, with the caveat that a falling population rate says nothing about any individual partner.

Live for approximately four months before correction.

What was wrong: The page described Nebraska’s protection order scheme as sitting in chapter 42 of the Revised Statutes, with a two-tier penalty ladder.

Why: LB 80 (2025) moved the entire scheme to a new chapter 26, and LB 965 (2026) added third and fourth-offense felony tiers with effect from 18 July 2026. Most published Nebraska guidance still describes the old structure. Found by a systematic re-check of every citation on every page.

Live for approximately from first publication until the sweep before correction.

What was wrong: The page cited chapter 14-07.1 for the domestic violence protection order scheme.

Why: S.L. 2025 ch. 145 § 17 repealed those provisions and replaced them with a new chapter 14-07.7, in force from 1 January 2026. We then made a second, smaller error in the correction itself — describing chapter 14-07.1 as repealed in its entirety, when only specified sections were. Both are now fixed.

Live for approximately from first publication until the sweep before correction.

What was wrong: The page cited AS 18.66.130 as the offense of violating a protective order.

Why: That section is “Specific protective orders.” The violation offense is at AS 11.56.740, in the criminal code. We had repeated a citation that appears in several secondary sources without reading the section itself.

Live for approximately from first publication until the sweep before correction.

What was wrong: The page did not mention Colorado’s firearms relinquishment scheme at all.

Why: C.R.S. § 13-14-105.5 makes relinquishment mandatory, attaches to temporary orders, and carries a 24 or 48-hour deadline with a seven-business-day affidavit and a mandatory arrest warrant for non-compliance. An omission rather than a misstatement, but a significant one — Colorado has one of the strongest schemes in the country and the page was silent on it.

Live for approximately from first publication until the sweep before correction.

What was wrong: An earlier version cited Ala. Code § 30-5-9 for the penalty for violating a protection order.

Why: That section was repealed in 2010. The current provision is § 13A-6-142. We had repeated guidance that is still widely published and still wrong.

Live for approximately corrected before this log began before correction.

What was wrong: An earlier version said Oregon’s firearms position was unverified.

Why: ORS 166.255 and the Oregon Judicial Department’s own bench sheet confirm that FAPA orders always qualify for the firearms prohibition, because the three findings required to issue one are exactly what ORS 166.255 requires. We had understated a protection that does exist.

Live for approximately corrected before this log began before correction.

The claim: Repeated as an established finding across essentially every stepfamily resource, usually without attribution.

What we found: It traces to a single source — Patricia Papernow, describing her own qualitative interview data in Becoming a Stepfamily (1993/1998). Her own words: “Preliminary data about the amount of time this process takes indicates that faster families complete the entire Stepfamily Cycle in about 4 years. Average-paced families take about 7 years… slower families remain in the early stages longer than 4 years, a few for as many as 12 years.”

What gets dropped: the word “preliminary,” and the twelve-year tail — which is the part that matters most to someone three years in who has been told they should be finished by now.

What the evidence says about the model itself: the one study designed to test whether stepfamilies develop in stages found that they largely do not. Baxter, Braithwaite & Nicholson (1999) tracked 53 families month by month across 566 coded turning points and found five trajectories rather than stages: 31.4% accelerated, 27.5% prolonged, 21.6% high-amplitude turbulent, 13.7% stagnating, 5.9% declining. They concluded the families “moved in and out of positive and negative turning points throughout their development,” a pattern belying reduction to a small number of stages.

Useful as a scale — this takes years, not months. Useless as a deadline. There is no clock you are behind on.