About this edition

The historical edition

Every 2025 post as captured from the magazine’s website, with every item under its permanent ID and the editors’ annotations kept beside the text, never merged into it.

Editorial note. To keep the posts as they were published, the historical edition does not edit them. Typos, misspellings, grammar, punctuation and capitalization are left as they appear in the captured posts, and so are errors of fact. The editors’ findings appear only in the annotations beside each item.

What it contains

47 captured posts. The captures are the 46 weekly posts published from February 17 to December 28, 2025, and the post of January 4, 2026. That post covers December 29 to January 4, and only its December 29–31 items belong to this collection. The posts are taken from the WordPress export of champaignmagazine.com generated on October 22, 2026. They are reproduced as stored there, with no corrections of spelling, capitalization, grammar or fact. The display makes only three changes, none of which alters the text: hidden editor markup is removed, links open in a new tab, and images load as you scroll. WordPress exports do not include revision histories, so a capture may include changes made after a post was first published.

Six reconstructed weeks. The series began in February, and January’s five weeks and the week of February 3–9 had no posts. They were reconstructed in October 2026 for this volume by four contributors working with the editor. They are shown here so the year is complete, but they are not historical posts: each is marked as reconstructed, and search leaves them out unless you include them.

What the annotations mean

Each item in a post carries a box in the margin color. The box gives the item’s permanent ID, the model that wrote it and the item’s label. It also says whether the concise edition printed the story, with a link to the entry. “Review record” opens what was checked, the sources, the reviewer and the date. The external reviewer’s verdict is shown where there was one.

The labels:

outside its week The event is real, but it happened in another week or year than the week that picked it. A report is dated by when it appeared.
inaccurate as stated The event is real, but a key fact is wrong: a figure, a party, or what happened.
no evidence found A search found no source for the product or event as named. This records a failed search. It is not a finding that the event never happened.
not an event A trend, commentary, roundup or source list that cannot be checked as one event.
unverified Reviewed, but neither confirmed nor refuted.
confirmed No label: checked and confirmed within the scope of the recorded review.

Some passages in the posts are a model checking another model’s list, or an instruction rather than an item. They keep their IDs but carry no label, because they are not selections. The panel’s own fact-checks are reproduced as published.

Collection notes come from the annotated master the editors kept while compiling the volume. Most explain a correction, a missing source or an unverifiable claim. They are attached to the item they concern, and they are never inserted into the post.

How the labels were checked

The editors labeled every one of the 893 selections. Muse Spark 1.3, an external reviewer, then checked the labels in two rounds, on October 28 and November 2, 2026. It checked every labeled item, every item that mentions Meta and a random sample of the items left unlabeled as confirmed: 634 of the 893 in all. The editor decided every change it proposed. Muse Spark 1.3 is a Meta model, so the editor cross-checked its verdicts on all 55 items that mention Meta.

The labels record the check as of the volume’s preparation in late 2026. They are not tied to the December 31, 2025 cutoff that governs the volume’s two introductions.

Permanent IDs

2025-W26-L4-1 is week 26, the fourth list in that week’s post, item 1. Weeks start on Monday, and week 1 is January 1–5. A post’s ID is the week (2025-W26), and its collection notes are numbered 2025-W26-C1 and so on. The same IDs are used in the concise edition’s index, so any printed entry can be traced back to the words the panel published.

Data

Labels and review records (CSV, v1.1): one row per selection.
Item register (CSV): every permanent ID, with contributor and role.
Captured posts (ZIP): the post HTML as stored in the export, and a readable text version of each.
Search index (JSON): the data behind the search page.

The forecast register is a dated record of the forward-looking statements in the 2025 posts, with source attribution and no outcomes. Its scoring rules were fixed before any outcome was examined.

Citing

AI by AI Weekly Top 5: The 2025 Collection, Historical Edition, edited by W.H.L. Champaign Magazine, 2026. Cite an item by its permanent ID. Text and data are licensed under CC BY 4.0, to the extent they are protected by copyright. The code is under the MIT License.