How to choose a point-in-time fundamentals API

Institutional platforms offer long history, global coverage, and point-in-time data that keeps figures as originally filed. They are generally sold through a sales process and licensed per seat or by agreement, and they bundle fundamentals with prices, estimates and other datasets.

Developer APIs are self-serve, with an API key in minutes and published prices. What they keep when a company restates, and whether they can answer what was known on a past date, varies from one to the next, and many keep only the latest figures.

Arche sits between them: the point-in-time and as-filed history of an institutional platform, for SEC filers, through a self-serve API with a free plan and published prices. The questions below are the ones that separate providers in practice.

A backtest or model that reads today’s figures for a past period sees corrections that were not public at the time. That is look-ahead bias, and nothing in the result shows it.

Across providers: Institutional platforms generally offer point-in-time history. Among developer APIs it varies, and many return only the current figures for a past period, so the same query run months apart can give different answers.

Arche: Yes, on every plan. An as-of query returns statements built only from filings the SEC had accepted by that date, and as_reported=true returns each period as originally filed. See Time series and as-of queries.

If a restatement overwrites the original, the value a model saw at the time can no longer be retrieved, and a result produced then cannot be reproduced now.

Across providers: Many datasets overwrite the previous values when a company files an amendment, keeping only the most recent figures. The SEC’s own aggregated company facts API works the same way: it returns one answer per period, the current one. Few providers send a notice naming each restated line with its old and new value.

Arche: A restatement adds a version. The original is never overwritten and stays queryable, with per-metric deltas and a materiality classification. Signed restatement webhooks are available from the Developer plan. See Versioning.

A number you cannot trace to its source cannot be checked, cited or defended in a review.

Across providers: Normalized datasets often present a value for a company and period without the filing it was read from, and figures the provider calculated can look the same as figures the company reported.

Arche: Every value carries the accession number of the SEC filing it was read from, and provenance endpoints return the XBRL concept. Values Arche derived rather than read are marked as derived. See Provenance.

Research run only on companies that still exist is biased toward survivors, in a direction that is invisible from inside the results.

Across providers: A dataset assembled from the companies that are listed today leaves out every company that was delisted, acquired or taken private, which biases backtests toward the survivors.

Arche: Yes. Arche covers 17,000 companies that file with the SEC, from 2009 to today, including registrants that were delisted, acquired or taken private. See Coverage.

Reproducibility is what lets a colleague, an auditor or your own future self confirm a result rather than reconstruct it.

Across providers: In many datasets, historical values change silently when filings are amended or reprocessed, and few providers let you hand a reviewer something that re-runs a query against exactly the data you saw.

Arche: Yes. Schemas, ordering and error semantics are stable, each snapshot is fingerprinted with a hash, and every query returns a replay token that reproduces the response exactly. See Concepts.

Inconsistencies found after the fact end up in your pipeline first.

Arche: Yes. Values taken from a filing are checked against what that filing reported, and statements are reconciled for accounting consistency before they are published.

An agent that reads different data than your code cannot be checked against it.

Across providers: Where providers offer an interface for AI agents, it is often reserved for paid or enterprise plans; institutional platforms generally do not offer one on a free tier.

Arche: A REST API with 81 endpoints and a public OpenAPI document, a typed Python SDK, an MCP server on every plan including Free that calls the same contracts, and bulk snapshots from the Scale plan. See MCP server.

Testing against companies whose statements you already know is the fastest way to judge a dataset.

Across providers: Institutional platforms are typically sold through a sales process and licensed per seat or by agreement. Developer APIs are usually self-serve with published prices, though what they offer for point-in-time history and restatements varies.

Arche: Yes. The Free plan covers any 25 companies you choose, with no card and no expiry. Paid plans are $149, $600, $2,500 a month, priced per organization rather than per seat, and Enterprise is per agreement.

• SEC filers only, from 2009 onwards. If you need longer history or companies that do not file with the SEC, an institutional platform is the better fit.
• Filed fundamentals only: financial statements, segments and narrative disclosures. Arche is not a market-data or price feed, and has no estimates or other asset classes. If you want those from the same vendor as your fundamentals, look elsewhere.
• If you want raw filing text, full-text search or the XBRL facts of individual filings to normalize yourself, a filing-level EDGAR API fits better than a normalized fundamentals API like Arche.

New to the terms? Start with what point-in-time fundamentals are, or see what teams build on Arche.