# Core concepts

Understanding Arche requires understanding how financial data evolves over time.

## Historical drift

Financial statements do not stay fixed forever.

They change because of:

- amended filings
- accounting corrections
- new disclosures
- vendor-side data revision pipelines

When a dataset exposes only the latest corrected state, the same historical query can return a different answer later.

## Statement lineage

Arche keeps each SEC filing's statement as filed. When a later filing restates
a period, the earlier filing's statement is kept alongside it rather than
overwritten.

Axon's fiscal 2024 balance sheet, for example:

| Filing (`accession_id`) | Event            | Accepted at          |
| ----------------------- | ---------------- | -------------------- |
| 0001069183-25-000019    | Original 10-K    | 2025-02-28T21:22:06Z |
| 0001069183-25-000075    | Restating 10-K/A | 2025-05-07T20:38:11Z |

That lets you ask both:

- what the latest filing reports for the period
- what was known on a specific historical date

## Point-in-time retrieval

A point-in-time query returns the state that was knowable at the specified boundary.

A date includes every filing the SEC dated on or before it. EDGAR dates a filing accepted after
5:30 pm Eastern the next business day, so a filing made after the close is not known on the day it
was accepted. For a tighter cutoff, pass a timestamp with an offset, such as
`2023-11-02T16:00:00-04:00`: it includes every filing the SEC accepted at or before that instant.

```http
GET /v1/edgar/as-of/2024-01-01/financials/0000320193
```

This is different from asking for the latest record today.

> **Tip:** Point-in-time correctness is a retrieval property, not just a timestamp field. The request must
> constrain the historical boundary directly.

## Deterministic research

Research is deterministic when the same request yields the same historical answer later.

That matters when you need to:

- reproduce a backtest
- audit an analyst workflow
- compare research runs across time
- explain why a company was included in a model at a historical point

## Next steps

- [Data model](https://docs.arche.fi/data-model): Learn how companies, filings and statements relate.
- [Versioning](https://docs.arche.fi/versioning): Work with latest, version history, and as_of semantics without historical drift.
- [Request & response conventions](https://docs.arche.fi/request-response-conventions): Response envelopes, timestamps, nullability, ordering, and field conventions.