Built for AI agents and backtests. Companies re-present their own history. Distill serves every value as filed, keyed to the accession it was read from, and screens the whole universe as it stood at any quarter-end from 2009. No lookahead, no survivorship, no restatement substituted in.
Use Distill MCP with Claude, Codex, or any other MCP client.
recreated from a real session · mcp.distillmarkets.com
Screen for profitable, low-leverage companies buying back stock. Then check whether any of them have re-presented their reported history.
37 match. Top 5 by net margin:
| Ticker | Net margin | Debt/EBITDA | Net dilution |
|---|---|---|---|
| INVA | 65.9% | 1.57x | −0.75% |
| APP | 60.8% | 0.81x | −0.51% |
| NVDA | 55.6% | 0.06x | −0.71% |
| BYD | 45.0% | 1.95x | −9.86% |
| VAL | 41.5% | 1.74x | −2.54% |
All five at once:
| Call | Restated facts | Last change |
|---|---|---|
| ✓INVA | 8 of 451 | 2017-02-28 |
| ✓APP | 15 of 214 | 2026-02-19 |
| ✓NVDA | 5 of 639 | 2022-03-18 |
| ✓BYD | 21 of 520 | 2025-02-21 |
| ✓VAL | 15 of 464 | 2023-02-21 |
BYD has the most restated facts, but 20 of its 21 were changed between 2014 and 2019. APP’s 15 all landed in one filing this February, and revenue is among them.
Revenue, as each filing stated it:
| FY | As reported | Filed | Accession |
|---|---|---|---|
| 2025 | $5,481M | 2026-02-19 | …26-000010 |
| 2024 | $4,709M | 2025-02-27 | …25-000018 |
| 2024 | $3,224M | 2026-02-19 | …26-000010 |
Revenue for a closed year moved 31.5% between filings. Either the numbers were wrong, or what counts as revenue changed. Net income across the same two filings settles it.
FY2024 net income, as each filing stated it:
| Filed | Accession | FY2024 net income |
|---|---|---|
| 2025-02-27 | …25-000018 | $1,579,776,000 |
| 2026-02-19 | …26-000010 | $1,579,776,000 |
Identical to the dollar. Revenue moved 31.5%; net income did not move at all. The basis changed, the results did not.
So FY2025 growth has one like-for-like answer: 70.0%, both years from the same filing on the same basis. Dividing FY2025 as filed by FY2024 as first filed gives 16.4%, a number no single filing supports.
Every value above carries the EDGAR accession of the filing it was read from. Factual information only, not advice.
As-filed figures carry their accession number. Follow it into EDGAR and check our work yourself.
Screen the whole universe as it stood at any quarter-end since 2009. As-filed values under the identities in force then. Broad universe coverage from 2012 onward, as the XBRL mandate phased in.
Refreshed nightly from EDGAR. Corrections are published, not patched quietly. Each correction says what was wrong, how many values changed, and the precision measured after.
The SEC already gives away every filed vintage of one company's facts, free, keyed to the accession that filed them. It won't resolve a delisted or renamed ticker to what was filed under it, and it won't say which of a filer's several tags for one line item is the one you want: GE's 2017 10-K carries three separate tags for net income, $436M apart. CELG resolves through its 2019 delisting to the numbers filed at the time, so you can reconstruct the fundamentals an investor could actually have used, as originally filed, with no survivorship or restatement bias.
The Beneish and Sloan detectors run on every filer with their components exposed, so you see why a name flags. Financials and Real Estate return null, since the ratios these models rest on do not carry the same meaning on those balance sheets; so does any name whose prior year is not on file. The same fields drive the screener: margins, leverage, forensic scores, the Piotroski F-score, cash conversion, across the whole universe in one call.
Run the same screen as of a past quarter-end and every row carries its filed date, for US domestic filers, at any quarter-end back to 2009, with useful breadth from around 2012 as the XBRL mandate phased in. The FAQ below shows what the SEC’s own frames endpoint leaves out.
| Ticker | Net margin | M-Score | F |
|---|---|---|---|
| INVA | 65.9% | — | 5 |
| APP | 60.8% | −2.78 | 8 |
| NVDA | 55.6% | −2.08 | 4 |
| BYD | 45.0% | — | 6 |
| VAL | 41.5% | −3.13 | 6 |
A 13F quarter fills in for about 45 days after quarter-end. Read it before that window closes and you catch a phantom collapse: an early read of 2026-06-30 showed Apple at $51.6 billion across 1,170 holders, against the prior quarter's completed $2,396.3 billion across 6,051. get_institutional_ownership returns the newest complete quarter by default, and says so. Every manager's several filing entities are rolled into one holder, Vanguard's among them. An unknown change is null, never zero, and new positions, adds and exits are tracked quarter by quarter.
| Manager | Shares | QoQ |
|---|---|---|
| Vanguard9 CIKs | 1,446.6M | +0.7% |
| BlackRock | 1,163.0M | +1.6% |
| State Street | 615.1M | +2.1% |
| Geode Capital | 371.3M | +0.7% |
| FMR | 325.3M | +5.8% |
Find out whether the number in your model is the one the company originally reported, both filings linked. Most flagged changes are not restatements, and the tool says so.
Follow one number across every filing that stated it, and cite the filing you actually used.
Measure a company against its own past 20 quarters: margins, leverage, liquidity, coverage, returns.
See which sectors are building capacity and which are starving it, from filed capital spend.
See what officers and directors did with their own shares, parsed and attributed from the daily Form 4 stream.
See who bought and who exited each quarter, with a manager’s many filing entities rolled into one holder.
Catch a filing the day it lands, timestamped to the SEC accept.
Point your agent at Distill and it reasons over point-in-time fundamentals, forensic scores, institutional ownership, and the universe screen directly. Fact history, as-of fundamentals and restatement detail return every value with the accession number of the filing it was read from, so an agent built on Distill can show its work. The tools also name the places this data misleads: a 13F quarter that has not finished filing, a share count that spans a split, a figure the filer tagged under a concept that means something narrower. An agent that reads them reports the artefact as an artefact.
# MCP config: Claude Code, Codex, Cursor, # or any client. Hosted, nothing to install. { "mcpServers": { "distill": { "url": "https://mcp.distillmarkets.com", "headers": {"X-Api-Key": "dmk_…"} } } }
In claude.ai or Claude Desktop, add a connector at mcp.distillmarkets.com and sign in. Nothing to paste. Connectors →
A cross-sectional join is only as good as the facts underneath it: one wrong value and the whole quarter’s answer is wrong. Independent readers re-derived each sampled value from the filing itself, with our parser output withheld. Each result below carries the sample it was measured on. The 2009-onward audits found zero value errors; the pre-XBRL layer states its measured rate, with the error catalogue published.
FY2010–2025 · 72 facts against the rendered statements
72 / 72 exact30 transactions re-read from the filed XML, field by field; the one difference is ambiguous by design
29 / 30 exact, 0 errorsFY1995–2009 · 196 verified-tier facts re-derived from the filings
98.0% · 95% CI 94.9–99.2Protocols and the error catalogue are on the methodology page. The per-cell comparison against EDGAR reruns at least weekly on the benchmark.
Free reaches most datasets at limited depth, including as-of fundamentals for one name up to 10 lookups a day. Analyst adds per-number verification and removes that daily cap. Pro adds the universe as of any quarter-end since 2009, the panel export, and deep history to FY1995, for $69 a month.
150 requests a day
The screener, verified fundamentals, insider filings, and the five largest 13F holders per ticker.
Data
Limits
2,000 requests a day
$19/mo for your first 6 months with code WELCOME10, through September 30.
Per-company depth: fiscal-year history, fact versions and restatements, 13F to the top 20, and as-of fundamentals for one name.
Data
Depth
Limits
15,000 requests a day
The universe as of any quarter-end since 2009, at full breadth from 2012, the panel export, and text-parsed history to FY1995.
Data
Depth
Limits
Filings we ingest from SEC EDGAR ourselves: XBRL financial facts, Form 13F holdings, and Form 4 insider transactions. Fact history, as-of fundamentals and restatement detail carry the accession number of the filing each value was read from.
Two checks, both published. A benchmark run compares the fundamentals we report against EDGAR’s companyfacts API cell by cell, matching on the same XBRL concept we picked; it refreshes at least weekly and publishes its methodology version and every cell that differs. Separately, independent readers re-derive sampled values from the filings themselves with our output withheld: 72 of 72 XBRL facts exact (n=72), 29 of 30 insider transactions exact (n=30). See the benchmark →
The universe screener, EDGAR-verified SEC fundamentals, Form 4 insider transactions, the five largest 13F holders and the five largest 5%+ blockholders per ticker, manager portfolios with whole-book totals and their five largest positions, valuation from your own price, restatement counts per ticker, 16 government-sourced macro indicators, material events from SEC filings, and the accuracy benchmark. All of it over MCP as well as HTTP. 150 requests a day, no card.
Check one company at a time: its SEC fundamentals as they stood on a past date, set beside what is on file today, so you can see what later got restated and what the market actually saw. Any date since 2009, annual or quarterly basis, ten lookups a day per account. Analyst removes the daily cap but it is still one name at a time. What it does not do is a study across many companies, or a screen of the whole universe as of a past date. That is the Pro point-in-time grid: the full equity universe at every quarter-end since 2009, survivorship-free, exportable as one CSV. Try a lookup →
Every capability is an MCP tool on the hosted server at mcp.distillmarkets.com. Point an agent at the URL with your key and it calls screen_universe, get_revisions and the rest by name. In claude.ai or Claude Desktop, add it as a connector and sign in; no key to paste. Read the docs →
You can, and every value here is derived from those same filings. EDGAR gives a lot away for free: its companyconcept endpoint returns every filed vintage of one company’s facts, keyed to its accession. What it does not give you is the cross-sectional version of that question. Its frames endpoint, the only free way to ask something across the whole market for a period, returns one value per company with no filed date attached: asked for CY2017 net income, it hands back GE’s number from a filing made in 2020, and nothing in the response says so. screen_universe answers the same kind of question for US domestic filers, as of a quarter-end, with the filed date on every row. Concept mapping across a filer’s own several tags for one line item, identity across a rename or delisting, and 13F and Form 4 parsed and attributed, are also work EDGAR leaves to you. Our benchmark compares our output back against EDGAR’s own companyfacts API cell by cell, so you can inspect that reconciliation yourself.
A 429 with a Retry-After header and a code naming which limit you reached: rate_limit_per_minute, rate_limit_daily, or subquota_exceeded for a per-endpoint budget (the briefing composite is 15/day on Free, the delta feed is 20/day). Authenticated responses carry X-RateLimit-Limit, Remaining and Reset for the request-level budgets, so an agent can pace itself against those before it gets there. The per-endpoint budgets carry no pacing header: the 429 itself is the first signal, and its Retry-After can run to the next UTC midnight. Quotas count per account rather than per key: holding more keys does not add budget.
The personal tiers cover your own research. Managing client capital, or putting outputs into a product you ship to your own users, runs under a commercial grant at $249 a month: one legal entity and its wholly-owned subsidiaries, up to five named users. Redistributing the data itself is a negotiated agreement rather than a plan. See the grants →
A free key gets you SEC fundamentals, the full-universe screener, the largest 13F holders, and the MCP server, every value keyed to the filing it came from. No card.