Console
Holders, insiders, activism and filings for any company, with every number traced to the sentence it came from.
How Console works →Hedge Funds, Quants & Data Teams
Alpha you can trace.
Named signals from SEC filings, frozen at EDGAR acceptance and linked to the sentence that triggered them. Backtest without look-ahead. Defend in a DDQ.
- 529named classifiers
- 245Mfiling sentences scored
- 41KSEC filers scored
- 0.956median AUC
- IC 0.095145-wk OOS
- 1994–point-in-time
The problem
A signal you can't explain doesn't survive the next audit
Sentiment scores with no source text. AI summaries with no citation. Vendor datasets quietly restated after the fact. Every one of them fails the same question: where did this number come from?
What you get instead
- Full auditabilitysignal → sentence → accession number
- No look-aheadscores frozen at EDGAR acceptance
- Explainable factorsnamed signals, not opaque outputs
- Defensible processDDQ and IC-ready paper trail
What solves it
Data your models can trust
APIs & Feeds
The same data as it was known at the time, by API, bulk file, or the MCP server for your AI tools.
How APIs & Feeds works →
See it on your desk
The same filings data three ways — on screen, in your AI tools, and as JSON.
Console
On screenStart from a screen, open the company, then the fund. Who is buying, how much, and how that fund's disclosed book has done against the S&P 500.
- Screens for smart money, insider clusters, crowding and short squeezes
- Every holder of a company in one list, with position changes
- Each fund's track record, turnover and top positions
- Every number links back to the filing it came from
Explore Console →
13F · 13D/G · Form 4one holder list
MCP Server
In your AI toolsAsk in plain English from Claude, ChatGPT or Cursor. The answer comes back with the signal, the filing sentences behind it and the form it came from.
- Live signals with z-scores, from the classifiers on this page
- The exact filing sentences, not a summary of a summary
- Read-only, so your tools can call it without a human approving each step
Explore MCP Server →
13 familiesof data
Ask about fund flows
Same MCP server"Which stocks saw the biggest cluster of notable funds entering last quarter?" The answer is ranked from the 13F screener, and it flags the index-driven names that only look like conviction.
- Notable new buyers and net flow per stock
- Index inclusions and leveraged ETFs called out, not hidden
Notable funds entering a position in Q1 2026, from the 13F screener
API
As dataOne key, one request, clean JSON back. Insider buys graded by each buyer's past record, holders, filings and fundamentals, ready for your models.
- Licensed by data set, so you pay only for what you use
- Live examples for every endpoint in the docs
- Point-in-time history for backtests
Explore API →
75endpoints · JSON
The claim we'll defend
In a market where every vendor claims the best signals, ours are the only ones you can actually verify.
3 quality gates, not just AUC62 classifiers cut → IC +48%4 of top 5 machine-discovered
Signals you can trace
Talk to our team about how you work today, or start a trial and explore the platform yourself.
The details
One signal, end to end
the artifact this whole page exists to show
-
01 · The signal
Gross Margin Compression
NKE / Nike · confidence 0.91 · echo 4.74
-
02 · The sentence
Gross margin for the third quarter decreased 130 basis points to 40.2% primarily due to higher tariffs in North America.
-
03 · The filing
10-Q · filed 2026-04-01
Accession 0000320187-26-000032
links straight to EDGAR
-
04 · The history
Fired 19 consecutive quarters
streak · echo · stickiness · z-score
What's in the solution
Four layers. Each independent, each useful alone.
01 · Foundation
Provenance Core
The classifier engine. 529 trained NLP models that read filings the way an analyst would — semantic, not keyword.
≈245M sentences · 41K filers
02 · Temporal
Provenance Stream
Signal memory across quarters, converged with XBRL, insider activity, news and credit metrics.
7,648 companies with live signal profiles
03 · Analyst output
Provenance Reports
Source-linked company analysis in HTML, MD and JSON — for analysts, pipelines and LLM training.
7,648 companies with reportable profiles
04 · Integration
MCP Server
A native MCP server — your AI client queries Core and Stream directly. No pipeline to maintain.
13 data families · 152 tools · read-only
Raw data is a commodity. Signal memory isn't.
A classifier firing once is an event. The pattern across quarters is the signal.
- Echoprev × 0.7 + fired. 5.0+ = entrenched
- Echo Deltagaining or losing momentum
- Streakconsecutive fires; breaking = change
- StickinessP(fire again | was firing)
- Z-Scorerarity vs. peers and vs. own history
How it lands on your desk
Same data product — different leverage point depending on the seat.
Quantitative analysts
The scenario
You're extending an equity factor model and need interpretable signals from filings — clean point-in-time history, and a paper trail that survives a DDQ.
CoreStreamConnect
How Provenance helps
- Add classifiers to existing factor libraries with a single daily table join
- Echo weighting amplifies signals that persist across consecutive filings
- Streak metadata separates first occurrences from entrenched patterns — critical for alpha decay
- Frozen point-in-time scores guarantee no look-ahead in walk-forward backtests
- Every factor value traceable to a source sentence and EDGAR accession number
Investment analysts
The scenario
You cover 25–40 names. Every quarter dozens of 10-Ks and 10-Qs land within days. You need to know which filings matter before reading a page.
CoreStreamReports
How Provenance helps
- Screen the whole coverage universe in minutes — which filings had unexpected signal changes
- Streak tracking separates a first-mention disclosure from language repeated six quarters running
- Source sentences link straight to EDGAR for one-click verification
- Build source-linked notes where every claim carries the exact filing sentence
- Catch thesis drift early: see which bullish classifiers are fading
Traders & trading desks
The scenario
You run event-driven strategies around filings. You need named signals near real-time after acceptance — with enough context to size and justify a position.
StreamConnectCore
How Provenance helps
- Classifiers fire near real-time after EDGAR acceptance, before price fully adjusts
- Echo delta separates genuinely new signals from confirmation of known themes
- Streak-1 flags mark first occurrences — highest surprise, strongest pricing implication
- Source sentences give the risk desk a justification for position sizing
- Company archetype context: the same signal means different things for a biotech and an energy major
Every product in this solution
Licence the whole stack or any single layer. Highlighted products are the ones most desks start with.
Full data catalog →Signal platform
- Provenance Core 529 named classifiers over ≈245M filing sentences Classifier engine
- Provenance Stream Signal memory across quarters, converged channels Temporal layer
- Provenance Reports Source-linked company analysis, 7,648 companies HTML · MD · JSON
- Provenance Explorer Browse what is firing right now across the universe UI
MCP families · license any subset
- Overviews Snapshot, reference data, event calendar, people In every plan
- News, Research & Filings Filings tagged by 529 classifiers, news events In every plan
- Ownership 13F, 13D/G, Form 4 insiders, N-PORT, short interest Live
- Activism Campaigns, 5%+ stakes with stated intent, proxy votes Live
- Credit & Debt Bond cards, holders, refinancing screens, BDC books Live
- Private Capital Rounds, Form D flow, investor portfolios Live
- Advisers (ADV) Profiles, fees, assets over time, disciplinary history Live
- Capital Graph Entities, neighbors, paths and money flows Live
- Fundamentals XBRL quarters, de-cumulated TTM, executive pay Live
- Sector Intelligence Sector monitors, plus Biotech, Power and Mining packs Live
- Charts Price history and the events that moved it Rolling out
- People Profiles, boards, transactions, vesting calendars Live
- M&A Deal history, comparables, trending acquirers Live
Delivery & tooling
- MCP Server Query Core and Stream from your AI client Integration
- Bulk & Streaming APIs Backfill to 2014, then stay current API
- Alerts Convergence and threshold triggers Workflow
- SEC Pro+ The filings behind every signal UI
Query it from your AI client
Claude · ChatGPT · Gemini · Cursor- semantic_search
- get_sentences
- holders_combined
- thirteenf_smart_money_consensus
- insider_signal
- read-only · auto-approvable
Which Financial Distress signals are most persistent across my coverage this quarter, and where is credit spread widening confirming it?
→ returns ranked results with source citations, in seconds.
Or take it as data
- Daily table join — one column into your existing factor library
- Bulk history back to 2014, frozen point-in-time
- Reports in .HTML / .MD / .JSON
| How it compares | Provenance | Generic NLP sentiment | Sell-side research |
|---|---|---|---|
| Signal type | Named classifiers, human + machine-discovered | One continuous score | Subjective opinion |
| Quality validation | 3 gates: separability, fire rate, coherence | AUC only, if tested | Peer review, if any |
| Interpretability | Click a signal and read the exact sentences | "Sentiment is 0.73" | "We downgrade to hold" |
| Update speed | Near real-time after acceptance | Varies | Days to weeks |
| Backtestable | 10+ years, frozen scores | Often limited | Not structured |