Pain map
What breaks in document AI
290 production incidents from public threads. Every quote links to its source. Pick a category and check it yourself.
Contradictions and stale facts
6 of 6 quotes
“the effect is edges with no dates even when the prompt's rules (REFERENCE_TIME, "use January 1st") apply.”
GitHub, graphiti · 23 September 2026 · in production · open the thread “The same entity extracted from 2014, 2019, and 2024 documents becomes three indistinguishable assertions instead of three dated ones.”
GitHub, LightRAG · 23 August 2026 · open the thread “In one production graph of roughly 3,950 facts, 1,616 (41%) carry an `invalid_at`.”
GitHub, graphiti · 4 August 2026 · in production · open the thread “we run `gpt-4.1-nano` for cost”
GitHub, graphiti · 18 July 2026 · open the thread “Graphiti's temporal model marks old facts as invalid when new information contradicts them, but the search tools still return those stale facts ranked equally with current ones.”
GitHub, graphiti · 13 July 2026 · in production · open the thread Hit this one too? We are building for it. Join the waitlist.
“Semantically, this new information makes the Day 1 relationship **outdated**”
GitHub, LightRAG · 23 December 2025 · in production · open the thread Hit this one too? We are building for it. Join the waitlist.
Method
We collected 448 public posts published after 1 November 2025 in which a builder describes document AI failing. Each link was re-opened on 8 November 2026. We kept the 290 where the quote is on the page, the author describes an incident and not an opinion, and has no tie to a vendor. Quotes are verbatim and trimmed after 420 characters. The grouping is ours. A complaint is not a measurement.