Dossier evidence rules (hard rule, no exceptions)

Set by Justin Smith on 2026-10-11. Applies to every dossier in dossiers/ and to anything the site says about a named person.

The rule

Every claim about a person carries a public link to the exact evidence for it. "Exact" means a reader who clicks lands on the statement, vote, or page itself: the right page of the PDF, the right second of the video, the right agenda item. A link to a site's home page or a meeting list is not evidence.

If a dossier says person X supports or opposes Y, the link goes to the place where X said it (their words, their signed letter, or the minutes recording what they said). A position is never inferred from votes, co-sponsorships, party, or associations.
Local files don't count. Paths like raw/…, video/granicus-3165/…, transcript.json or speaker_evidence.json are working files. They can sit in an HTML comment for maintainers, never as the citation.
No public link, no claim. A claim without one is deleted, or moved to the section "Unverified / pending public source", written as an open question ("Not yet confirmed: …"), never as a fact.
Each claim gets an ID (K-01, …) so it can be audited, and every claim line must contain at least one https:// link. tools/check_dossier_links.py enforces this and re-checks every link (HTTP 200 and right document type; for #page=N, the PDF has at least N pages).

What each kind of claim links to

Claim type	Required evidence link	Label
Vote / motion / outcome (minutes exist)	Approved minutes PDF at the page: `MinutesViewer…#page=N`	none needed (official record)
Vote / outcome (minutes not out yet)	Granicus video at the moment of the vote: `MediaPlayer…&entrytime=S`	"machine transcript, unofficial; replace with minutes when approved"
Sponsorship / authorship	Docket/agenda PDF at the page with the `BY:` line or the signed letter: `AgendaViewer…#page=N`	none
Quote from a document	The PDF page (`#page=N`) or web page (optionally with a `#:~:text=` fragment) where the words appear	none; must be verbatim, `[sic]` for errors
Paraphrase from minutes	The minutes page (`#page=N`)	[paraphrase from minutes]
Quote from a meeting video	Video at the timestamp (`&entrytime=S` or YouTube `&t=Ss`)	"machine transcript; check against the video" (wording is ASR)
Machine speaker attribution (who said it)	Video at the timestamp of the remark plus a link to the moment that identifies the voice (for example, the chair being addressed by name)	"Speaker identified automatically by voice matching, not by official record"
School board vote	BoardDocs item link (shows "Motion & Voting" with mover, seconder, Yea/Nay by name)	none
Self-description (bio, questionnaire, campaign site)	The page, ideally with a `#:~:text=` fragment	"her/his own description, not independently verified"
Secondary source (news)	The article URL	"secondary source"
Election result	Cuyahoga BOE certified results PDF at the page	"unofficial" if it is not the certified canvass

Verified link formats (tested 2026-10-11 ET)

Source	Format	How it was verified	Status
Granicus video at a timestamp	`https://onelakewood.granicus.com/MediaPlayer.php?view_id=2&clip_id=<clip>&entrytime=<seconds>`	Clip 3165: HTTP 200 (redirects to `/player/clip/3165?view_id=2&entrytime=720`). The server writes the value into the player config (`start_time: 720`; `start_time: 0` without it), and Granicus's own share/embed code builds links with `&entrytime=`. Player chapter cuepoints (13:52 item 3, 14:08 item 4) line up with our ASR times (13:52, 14:10), so ASR seconds = player seconds for Granicus clips.	Works. Caveat: actual playback couldn't be watched from the box (the video CDN refuses the box's headless browser), so the seek itself was confirmed from the player config, not on screen.
Granicus minutes PDF at a page	`https://onelakewood.granicus.com/MinutesViewer.php?view_id=2&clip_id=<clip>&doc_id=<uuid>#page=<N>`	200, 302 → `DocumentViewer.php?file=…pdf`, `application/pdf`, served inline (no attachment header). Browsers keep the `#page` fragment across the redirect. N is the PDF page index; for Lakewood minutes it equals the printed page number.	Works in Chrome, Edge, Firefox, desktop Safari. iOS/Android viewers may ignore `#page`, so always also write "p. N" in the text.
Granicus agenda/docket PDF at a page	`https://onelakewood.granicus.com/AgendaViewer.php?view_id=2&clip_id=<clip>#page=<N>` (or `&event_id=<id>` before a video exists)	Same as minutes. N ≠ the docket's printed "pg.": Oct 5 and Sept 8 dockets have 3 cover pages, so PDF page = printed page + 3 (printed pg. 161 = `#page=164`). Always compute N from the PDF itself.	Works (same caveat)
BoardDocs agenda item	`https://go.boarddocs.com/oh/lwcsoh/Board.nsf/goto?open&id=<ITEM_UNIQUE>` (item ids from `BD-GetAgenda`); meeting: same with the meeting id	`goto?open&id=DXWL925545A1` loaded in a real browser shows "Agenda Item Details … Motion & Voting … Motion by Lisa Dopman, second by Nora Katzenberger … Yea: …". 200 with a browser User-Agent.	Works
YouTube at a timestamp	`https://www.youtube.com/watch?v=<id>&t=<seconds>s`	Video id `w4HUbUbaYdE` confirmed public via YouTube oEmbed (200, title "09-28-2026 Lakewood City Schools Board of Education Meeting"). `&t=` is YouTube's standard start-time parameter.	Use it, but the box can't load the watch page itself (YouTube returns 429/"sorry"), so the link checker verifies the id via oEmbed, not the seek. YouTube auto-caption timing matches the video timeline.
City / other web page	Plain URL; for a quote, optionally add a text fragment `#:~:text=<exact words>`	`lakewoodoh.gov/city-council/sarah-kepple/` etc. 200; quoted text confirmed in the fetched HTML.	Works. Text fragments highlight in Chrome/Edge/Safari and Firefox 131+; older browsers just open the page, so the quote must also be in the dossier text.
Cuyahoga BOE results PDF	`https://boe.cuyahogacounty.gov/elections/GetDocumentById/<uuid>/#page=<N>`	Certified "NOV25GEN – Official Results by Contest" (2025-11-19): 200 `application/pdf`; Lakewood Council at Large is on PDF page 17.	Works

Formats that did NOT work or must not be used

https://onelakewood.granicus.com/videos/<clip>/captions.vtt: 200 but empty for clip 3165. Not usable as evidence.
Granicus DocumentViewer.php?file=… direct file URLs: they work today, but they're the redirect target, not the canonical link. Cite MinutesViewer/AgendaViewer instead.
The raw HLS stream (archive-stream.granicus.com/…m3u8) needs a Referer header and has no timestamp UI. Never cite it.
Local paths (raw/, video/, data/), our ASR JSON, and our speaker-label files are never evidence.
Lakewood Observer election-night tallies are unofficial. Use the BOE certified PDF once it exists, which it does for Nov 2025.

Procedure for every new or updated claim

Find the public document or video moment. Compute #page=N from the PDF itself (pdftotext + search), or the second from the ASR segment start (link about 2 s early).
Put the link on the same line as the claim, with the label from the table above.
Run python3 tools/check_dossier_links.py dossiers/<name>.md. Every link must return 200, and every claim line needs a link.
When minutes are approved, swap machine-transcript links for minutes links and drop the label.
When something can't be linked, move it to "Unverified / pending public source" as a question, or delete it.