How this site is made, and where its biases come from

This site asks you to check where claims come from. So here is where ours come from, including the parts we'd rather you didn't have to worry about.

Who makes it

Sidenote is written by an independent editor working together with an AI model (Claude, by Anthropic). The AI drafts entries, searches for sources, reads archive scans and checks quotes; the editor decides what the site is for, reviews, and corrects. Entries drafted by the AI carry the tag provenance: llm in the site's data and count as drafts until they're reviewed; as of October 2026 almost all entries are still drafts, and the pages themselves do not yet show that on screen, so read every entry as a draft unless it says otherwise. Sources marked ✓ have been checked against the original document; ⚠ means they haven't been yet.

Bias in the sources

Everything online about this conflict has been fought over, and that includes the reference works. Wikipedia's Israel–Palestine articles have been under special arbitration restrictions for years, and organised editing campaigns from both sides have been documented and sanctioned. That doesn't make Wikipedia useless, but it does make it a pointer, not a source. Here we follow a claim to the document it rests on: the law, the ruling, the dataset, the original newspaper page.

Reports that cite each other count once. A conclusion repeated by five organisations that share their inputs is one finding, not five confirmations (see the citation loop). This applies on both sides.

Old newspapers have biases too. The archive scans on this site are mostly from the New York Times: one outlet, one country's view, and an index that contains its own errors. We use them to show how events were reported at the time, not as the final word on what happened.

Reading the conclusions

The "Working it out" notes end by sorting what follows from the evidence. Each label says how a conclusion stands to the evidence, not how comfortable it is:

  • Holds on every reading: it survives every reading of the disputed inputs on the page, including the one most hostile to it.
  • Holds on the evidence here (or "on these figures", "on these cases"): the evidence cited supports it, but not every reading has been tested against it.
  • Holds only if … (or "only case by case"): it is true if the named assumption or condition holds, and not otherwise.
  • Open, needs … / Open until …: the evidence that would decide it is not on this site, does not exist yet, or waits on an event such as a court ruling; the label names which.
  • Arguable: a judgement the evidence leaves to the reader; the page gives the strongest case each way.
  • Could happen …: a statement about the future and its conditions, not a conclusion from data.
  • Not supported / Not supported by any reading: the evidence here does not carry it, or no reading of it does.

Until 12 October 2026 the labels were "Safe", "Possible" and "Not supported". "Possible" covered four different things (missing evidence, a condition, a judgement, a forecast), so it was split.

Source cards can carry an evidence weight: incident (one event or statement: shows it happened, not how often), pattern (many incidents documented case by case: shows repetition; how typical depends on how the cases were chosen), aggregate (a count or statistic: shows scale; what was counted and who counted matter) and finding (a conclusion drawn from evidence by a court, commission, inquiry or published analysis: as strong as its method and inputs). A general claim that rests only on incidents is a warning sign.

An argument is labelled "Common" only when three or more different speakers are shown making it in public; with one or two, the page names them ("Made publicly by …").

Reading how a page changed

Pages are edited after the headlines are written. Organisations add notes, corrections and addenda to a page after it has been quoted around the world, and the later text rarely travels with the first. Where it matters, we record what a page says now, what was added later, and in what order (for an example, see the famine entry).

How we tell. First, dates in the documents themselves: addenda, corrections, "updated" notes. Second, archived copies in the Internet Archive's Wayback Machine. Third, the page's own structure: many websites are assembled from content blocks that carry increasing ID numbers, so on a saved copy of the page those numbers show the order in which the blocks were created, whatever order the page now displays them in.

"Removed" is a claim too. A page that no longer loads may have moved, or may never have been at the address we tried. We only say something was removed when an archived copy shows it at its old address.

Bias in the AI

It recalls; it doesn't look things up. An AI model writes from patterns in its training data. It's usually right about the shape of a story and less reliable about exact numbers, dates and wording. It can also be confidently wrong in a very fluent way.

Its training data mirrors what's been written. A misquote that circulates widely online is more familiar to it than the verbatim original. So the AI reproduces popular versions: the famous wording rather than what was actually said, or a summary's mistake rather than the source's fact.

It surfaces the most-written-about version first. Asked about a topic, an AI reaches first for the arguments that appear most often in its training data, usually the consensus framing and the older, familiar arguments on each side. Newer or less-covered arguments, sometimes the strongest ones their advocates actually use, can stay out of view unless someone looks for them on purpose. So before an argument is stated here, we search for what that side's current leading advocates and legal organisations argue, rather than relying on what comes to mind first.

It tends to agree with whoever it's working with. Language models are known to lean towards the views of the person in the conversation. Our counterweights are rules the site holds itself to: every argument is stated so that its own best advocates would sign it, every argument is examined in both directions, and every claim needs a public source.

It rounds toward the point it is making. Checking its own drafts against their sources, we found the errors lean in a direction: they make each sentence say a little more than its source, toward whatever the paragraph is arguing. In one check of public figures' quotations, 7 of the first 18 were not worded as on the page they linked to — each wrong version sharper than the real one. In an audit of how entries end, most endings that claimed more than their evidence leaned toward the Israeli reading, including on pro-Palestinian entries, where the ending assumed the harsher meaning of the other side's words. As of 28 September 2026, of 645 over-firm endings found, 344 leaned toward the Israeli reading, 191 toward the Palestinian and 110 toward neither. Both are corrected as they are found, and the audits continue; the pattern is stated here so readers can watch for it too.

The verification record

These numbers are generated automatically from the site's database each time the site is published: they count what is actually marked in the data, not what anyone says was checked. As of 2026-10-15 (database version 9n05ttal).

Sources cited in entries

912 checked against the original ✓44 marked to verify ⚠of 965 sources

Newspaper archive

94 flagged items confirmed26 corrected107 resolved by rule, not by eyeof 441 automatically flagged, 2208 clippings; 40 fully verified where cited

Examples from the record

ItemOutcomeRecorded note
archive clipping
The Jewish Question (1852-05-08)
correctedDate corrected from the date printed on the scan; the archive index had it wrong.
archive clipping
Russia and the Jews (1882-04-22)
correctedThe scan shows a different article than the index describes; the index entry is wrong.
archive clipping
Jewish Hardships in Russia (1891-04-15)
correctedText-recognition error corrected by reading the scan.
archive clipping
Mohammed and the Medina Jews (1895-09-13)
correctedDate corrected from the date printed on the scan; the archive index had it wrong.
archive clipping
Russian Jews Massacred (1904-08-28)
correctedDate corrected from the date printed on the scan; the archive index had it wrong.
archive clipping
Alleged Outrages on Jews (1905-08-31)
correctedThe scan shows a different article than the index describes; the index entry is wrong.
archive clipping
Minimizes Favor to Jews (1913-08-21)
correctedDate corrected from the date printed on the scan; the archive index had it wrong.
archive clipping
nazis said to arm palestine arabs (1941-05-29)
correctedDate corrected from the date printed on the scan; the archive index had it wrong.
archive clipping
“Displaced” Children (1948-09-05)
correctedDate corrected from the date printed on the scan; the archive index had it wrong.
archive clipping
Socialists Strong in Israel Regime (1948-09-12)
correctedDate corrected from the date printed on the scan; the archive index had it wrong.
source
James Dorsey, “'Wij zijn alleen Palestijn om politieke reden'” (interview with Zuheir Mohsen), Trouw, 31 March 1977, p. 7
confirmedChecked against the original document.
source
Al-Monitor, 'Arab-Israeli city Umm al-Fahm figures in Liberman's land-swap plan' (January 2014)
confirmedChecked against the original document.
source
The Algemeiner, 'Israel Arabs Prefer to Remain Israelis Despite FM Lieberman's Call for Territory Swap (VIDEO)' (10 January 2014)
confirmedChecked against the original document.
source
Freedom House, Freedom in the World 2025: Gaza Strip
confirmedChecked against the original document.
source
Freedom House, Freedom in the World 2025: West Bank
confirmedChecked against the original document.
source
Rafael Medoff, 'Mein Kampf in Gaza—and Beyond', Jewish Journal (26 November 2023), republished by JNS
confirmedChecked against the original document.
source
NATO Strategic Communications Centre of Excellence, 'Hybrid Threats: Hamas' use of human shields in Gaza'
confirmedChecked against the original document.
source
Dissenting opinion of Vice-President Sebutinde, ICJ Advisory Opinion on the Occupied Palestinian Territory (19 July 2024), UNISPAL copy
confirmedChecked against the original document.
source
Revue Conditions, 'Correspondance entre Yusuf Dia Khalidi et Theodor Herzl (1899)' — full French text of al-Khalidi's letter of 1 March 1899 and Herzl's reply
confirmedChecked against the original document.
source
UN Security Council Resolution 2803 (2025), 17 November 2025 — text via 'The World and Japan' database
confirmedChecked against the original document.

Quotes in the 'Where it's been said' sections carry their own ⚠ until checked; those aren't counted here yet.