Show HN: We had AI write a wiki from 45k Enron emails, every fact cited

A prototype called this+that used an AI to build 214 pages about organizations Enron dealt with, drawn from 45,047 Enron emails. The site reports 49,401 facts, each citing its source email, and says facts are expired on the date the mail shows they stopped being true. A held-out audit by a second AI reportedly found 93% of sampled facts fully supported by their email and 0 of 500 sampled facts wrong at their core. Entity selection used a threshold of 25 messages.

Key takeaways

• The site reports 214 pages and 49,401 facts, each citing the email it came from.
• A second AI audit reportedly found 93% of sampled facts fully supported by their email.
• Entity pages were chosen for organizations named in at least 25 messages, plus five internal Enron units.
• Facts are expired on the date the mail shows they stopped being true, rather than edited.

Why it matters

It illustrates a concrete approach to grounded, citation-backed company knowledge bases built by LLMs, and how temporal expiry of facts might be handled. The method's reliability rests on the reported self-audit figures.

Read with care

Source claims not independently checked; the 93% and 0-of-500 figures come from an audit the source describes but whose methodology is only summarized.