Connecting…
Each mark is an event assembled from the archive's own footage, or a person, place, facility or named perpetrator connected to one. Every link carries the reason it exists.
Links are graded. Solid means a deterministic signal — a matched video fingerprint, a resolved gazetteer id, a name in the text. Dashed means a lead worth checking, never a finding.
This is a pilot on partial extraction.
107,465 EVENTS · 134,428 CLIPS · 8,845 PEOPLE IN FOOTAGE · 72,785 IDENTIFIED DEAD · 724 ACTORS · 3,961 FACILITIES
Iris is currently very restricted, but we’re gradually lifting those restrictions. If you don’t get the answer you’re looking for, try again later. She’s updated several times a day.
Every node here came out of an event documented in the archive — the footage itself, the caption written by whoever filmed or posted it, or a daily report. No incident was authored by hand, and no coordinate was ever produced by a model — places resolve to OpenStreetMap rows by id, so each point on the map can be checked against its source.
Clips are grouped by measurable signals alone, each of which can be recomputed and checked, because merging two distinct massacres into one node asserts something that did not happen:
| video fingerprint | the same footage, after re-encoding — 3 of 5 perceptual frames within Hamming distance 10 |
| post_url | the same original post |
| date + place | only clips claiming the same day and the same resolved place are compared |
Established follows from one of those signals. Possible is a lead — a shared surname, the same facility on another date, fingerprints that matched 2 of 5 frames. Leads are drawn dashed and labelled unconfirmed everywhere they appear. They are offered so nothing is lost, not asserted.
Two clips of one event whose fingerprints do not match were filmed separately; if they also came from different source archives, they were published separately. That is the strongest corroboration this archive can generate from its own holdings, and it distinguishes footage reposted thirty times by one channel from an incident three people filmed.
| A | 2+ independently filmed clips from 2+ independent archives |
| B | 2+ independent archives, or 2+ independently filmed clips |
| C | single source, links to its original post |
| D | archive-only, with a date stated in the text |
| E | weak — the date is the post timestamp, not a stated event date |
A daily report existing for the same date is recorded but is not a tier anchor: it says the day is documented, not that this event is.
A casualty figure appearing in a caption is not automatically a count of people Israel killed at that incident. Every figure below was found, examined and rejected, with the reason recorded. They are listed here because an archive that only ever shows what it accepted is asking to be trusted, while one that shows what it discarded can be checked.
People and perpetrators are kept in separate classes and never merged. A person node exists only when the name is specific enough to be an identity (three tokens or more); shorter mentions are counted but never become a person, because one node for "Mohammed" would be a person who was everywhere at once.