How do you reconstruct what a financial system observed at time T, years later?

3 pointsby denjuro6 месяцев назад2 comments

I’m working on a problem I keep running into when dealing with complex financial analysis systems (quant, risk, governance, infra). Markets are obviously uncertain, that’s not the issue. The issue is that, ex-post, it’s often very hard to reconstruct what the system actually observed at a specific point in time. Typical questions that become surprisingly hard to answer a year later: - What data was actually available at time T? - In what order was it processed? - Which transformations and constraints were applied? - What was explicitly known vs implicitly assumed? - Can this observation be replayed deterministically, without hindsight? In practice, dashboards change, pipelines evolve, inputs get revised, and explanations become narrative rather than evidential. I’m curious how people here deal with this in real systems: - Is this a problem you recognize? - Do you solve it structurally, or mostly through process and/or/with documentation? - Are there established patterns I’m missing?

Not pitching anything, I’m genuinely trying to understand how others approach this.

thank you!

2 comments

curious_curios6 месяцев назад

Generally they do it via bitemporal modeling in databases: https://en.wikipedia.org/wiki/Bitemporal_modeling

denjuro6 месяцев назад

Thanks for your reply. Yes, bitemporal modeling helps with data lineage. What I’m struggling with is reconstructing the process that produced an observation like ordering, constraints, fallbacks.. not just the data state. Curious if you’ve seen that handled anywhere.