Trust & security
Trust every number.
Private capital runs on trust. GoodStream is built so every data point in your portfolio can be traced to the document it came from, the models that extracted it, and the validation that confirmed it. No black boxes. No "trust us." Verify everything.
SOC 2 Type 2 examined, with an unqualified opinion. Encrypted in transit and at rest. LLMs will never train on your data. The report is available under NDA.
INDEPENDENTLY EXAMINED
SOC 2
TYPE 2
Security criteria · Unqualified opinion
Examination period Mar 25 to Jun 25, 2026 · Report Sep 1, 2026
Ezzy & Associates LLC, independent CPA firm
The basics
The basics.
SOC 2 Type 2
An independent CPA firm tested the design and operating effectiveness of our security controls from March 25 to June 25, 2026 and issued an unqualified opinion. The full report is available under NDA.
Encrypted and isolated
TLS 1.2 or higher in transit, AES-256 at rest, row-level isolation between organizations, and access logging on every call.
Your data stays yours
LLMs will never train on your data. Our agreements with every model provider prohibit it, and providers retain nothing beyond the processing session.
Provenance
Every number has a source. Every source has a trail.
Click any value and see the page it came from: the SPA, the charter section, the board deck slide. We call it full provenance; your auditor will call it a complete audit trail. Most portfolio tools show you a number. We show you where it came from.
Acme Robotics - Post-money valuation
SPA p.14$420.0M
Source
Series C Preferred Stock Purchase Agreement - p.14
...a post-money valuation of $420,000,000...
Validated
Multiple models extracted this value independently. They agree.
Reviewed
Values below the confidence bar route to our review team before they enter the record. Verified is reserved for values tied to your source of record.
Logged
Append-only audit log - who, what, when. Nothing silently overwritten.
2026-08-14 - extraction - source SPA p.14
Validation
Accuracy by skepticism, not optimism.
Multi-LLM consensus
Every extraction passes through multiple independent models from different providers. Values that don't agree don't pass.
Confidence scoring
Every value carries a confidence score. High confidence flows; low confidence routes to human review automatically.
Our review team closes the gap
One AI model reading a charter on its own gets about 63% of the terms right. Our extraction pipeline gets to 96%, and our trained review team checks every value the agents flag, which takes it to 100%. The service is ours to run; the accuracy is yours to keep.
History
The record remembers everything.
Append-only audit log
Every change is recorded; nothing is silently overwritten. Full history on every field.
Attribution on every value
Human or AI, extraction or correction: every value shows who touched it, when, and why.
Correction
When a number is wrong.
No extraction system is perfect, so the record is built for correction. Flag any value and it routes to review; the fix lands with full attribution; the append-only history keeps what changed, when, and why; and every surface that used the old value updates from the record. An error in GoodStream is a visible, traceable event - not a silent cell in someone's spreadsheet.