I'm Filip, the creator of FilingLantern. It is primarily a plugin for ChatGPT, with a companion website for inspecting the figures, calculations and source documents.
The problem I wanted to solve
A fluent explanation of a company's results can sound convincing while leaving an important question unanswered: can I check the evidence behind it?
FilingLantern's workflow starts in ChatGPT. You can ask follow-up questions about reported company figures, examine matching fiscal periods, follow SEC filing sources and see where missing or incompatible inputs prevent a calculation from closing.
Try it in ChatGPT
- Open the FilingLantern installation page.
- Connect FilingLantern using the controls shown in ChatGPT.
- Start with a concrete question, for example:
Analyze FIX's 2024 and 2025 annual results. Show the evidence card, explain the calculations, then challenge the explanation.
Then ask: Which reported figures support this comparison? What remains missing? Which filing supports that claim?
Explore the companion website demo for FIX. The website interface supports 10 languages, selectable from its language menu; original SEC documents remain in their source language.
Three boundaries that matter
- A closed calculation is not a causal explanation. Reconciling operating income, nonoperating totals and tax to net income does not establish why the business changed.
- Missing inputs stay visible. An unresolved gap should not become a plausible story or an invented number.
- Automated checks and reviewed explanations are different. Full reviewed Evidence & Challenge cases currently cover FIX and CME only.
This is a research tool, not live prices, forecasts or buy/sell recommendations. The public website demo is free; the full website research plan is $9 per month.
Feedback I'd find useful
For anyone building tools around language models: how would you present the boundary between retrieved facts, arithmetic and interpretation?
For anyone trying FilingLantern: is the path from a question in ChatGPT to checking the source clear, or is there an onboarding step you would change?
Disclosure: I'm the maker of FilingLantern. This post was drafted with AI assistance.
Top comments (2)
I'd make the explanation cite the exact calculation version, not just the filing. If a source figure is restated or a period mapping changes, keep the old evidence card readable and mark the old explanation as superseded until it is reviewed again. A test with two filings for the same fiscal period would show whether the challenge follows the figure actually used or quietly switches to the latest one.
Thanks, Kent — tying an explanation to the exact inputs and calculation version is a useful distinction. For now, I'm keeping the free demo focused rather than adding a version-history feature to it.
To clarify what is already available: FilingLantern is primarily a ChatGPT plugin, with a companion website. The $9/month website research plan provides access across 2,122 company cards, quarterly and annual views, nine reported financial metrics where available, source filings, reported history and CSV exports. Its automatic accounting check uses operating income, the reported nonoperating total, tax and net income across two periods; missing inputs and unresolved differences remain visible. Full read-and-reviewed Evidence & Challenge currently covers FIX and CME only. A calculation that closes does not establish the business cause.
Your proposed immutable calculation history and automatic superseded-explanation status are not features I'm claiming to offer today. I see them as a possible extension to the paid research workflow, rather than something the free demo needs. The fact that the underlying SEC reports are public doesn't remove the value of reproducibility. Your two-filings/same-period test is a good acceptance criterion for that work.