Section 01 · Cover
The page your auditor sees first.
Organization name, framework + version, period covered, generated-at timestamp, SHA-256 of the underlying event log, and the signing identity. No marketing. No logo larger than the framework name.
Seven sections. One PDF. Signed and hash-verifiable. Built for the auditor's review, not the vendor's pitch. Sample is the real format, with realistic synthetic data.
Telemetry surfaces included with every Evidence purchase · observation-only by default
01 / Anatomy
Every section earns its place. Auditors asked for it during interviews. Nothing is here to look impressive on a deck.
Section 01 · Cover
Organization name, framework + version, period covered, generated-at timestamp, SHA-256 of the underlying event log, and the signing identity. No marketing. No logo larger than the framework name.
Section 02 · Scope
Surfaces observed (browser, desktop, CLI, MCP). Agents and users covered. Explicit out-of-scope: services accessed via API tokens that bypass the browser and desktop, on-prem inference that never leaves the network, and any usage not captured by an installed Northbeams client. Honesty up front. Auditors trust scoping more than blanket claims.
Section 03 · Control mapping
For each control in the framework: AUTO-evidenced (Northbeams telemetry alone proves it) / ATTESTed (someone signed off, with timestamp + actor) / scoped-out (with reason). Each row links to the query and the sampled events in the appendix.
Section 04 · Evidence appendix
For each AUTO control, the top-N high-risk tool calls, blocked actions, and attested controls with timestamps and actors. Argument hashes, never argument values (the values stay on the user's device by design). Auditors get enough to verify the AUTO claim without needing the raw data.
Section 05 · Exceptions & gaps
Controls not satisfied. Each one carries a reason ("compensating control X covers this," "remediation in flight," "explicitly out of scope") and a remediation owner with a deadline. This is the section auditors flip to first to test honesty.
Section 06 · Methodology
Collection model (browser extension, desktop app, MCP Gateway in-path). Retention defaults per tier. Integrity claim: append-only event log, hash-chained, every batch's hash committed to the next batch's header. The pack carries the chain's tail hash so an auditor can verify integrity end-to-end.
Section 07 · Signatures
An owner attestation block (the customer's compliance officer or vCISO signs). A cryptographic signature over the document from Northbeams. HMAC v1 today; X.509 v2 within 90 days; optional customer-side key escrow on Enterprise. Verifiable at /trust/verify.
02 / Integrity
Auditors do not trust vendor PDFs. They trust verifiable claims. The pack is engineered to be verifiable, even if Northbeams is offline.
01 · Signature
Every pack carries a hash of its underlying event log and a detached cryptographic signature. Re-hash the file; compare to the cover. Paste the hash at /trust/verify.
02 · Hash chain
Each daily event batch carries the hash of the previous batch in its header. Tampering with history breaks the chain. The pack carries the chain's tail hash for end-to-end verification.
03 · Scope honesty
The scope statement lists what we don't see (API-token usage, on-prem inference, off-network devices). Auditors trust packs that admit limits over packs that claim everything.
03 / Cadence
Generate one before a vendor questionnaire. Schedule one a month. Wire your GRC platform to pull a fresh pack on every control-status change.
⌁
Click "Generate Evidence Pack" in the dashboard. Pick framework + period. Pack lands in under a minute. Stored under /orgs/{org}/evidencePacks/, downloadable forever.
◷
A scheduled Firebase function runs at month-end, generates the pack, emails it to nominated recipients (compliance officer, vCISO, your auditor's shared mailbox), and pushes evidence to the linked GRC integrations.
⇲
Register a webhook on the Evidence API. Whenever a control flips status (AUTO → gap, ATTEST expired, sampled exception found), the webhook fires. Your GRC platform pulls a fresh delta pack on event.
04 / Variants
Same anatomy. Same integrity model. The Control Mapping section differs per framework, because the controls do. Pick a framework to see how the section looks.
14 AUTO, 19 ATTEST, 5 scoped-out by default. Customer can re-classify any row with a written reason.
| Control | Title | Status | Evidence source |
|---|---|---|---|
| A.6.2.6 | AI system monitoring | AUTO | Continuous event log across 4 surfaces |
| A.6.2.4 | Verification and validation | ATTEST | Quarterly attestation by Head of AI |
| A.7.4 | Quality of data | ATTEST | Data steward attestation |
| A.8.4 | Reporting concerns | AUTO | In-product report channel + log |
| A.9.3 | Objectives of responsible AI use | ATTEST | Annual policy attestation |
| A.10.2 | Allocation of responsibilities | ATTEST | RACI document signed off |
| A.10.3 | Supplier relationships | SCOPED | Procurement system, not Northbeams |
Logging, human oversight, monitoring. The deployer-specific clauses are the load-bearing ones for non-EU companies whose tools land inside the EU.
| Article | Obligation | Status | Evidence source |
|---|---|---|---|
| 26(1) | Use in accordance with instructions | ATTEST | Policy attestation |
| 26(2) | Human oversight assigned | ATTEST | RACI |
| 26(5) | Operate and monitor in line with instructions | AUTO | Continuous monitoring + per-tool policy |
| 26(6) | Keep logs at least 6 months | AUTO | Retention 90d / 1y / 7y per tier |
| 26(7) | Inform workers' representatives | ATTEST | HR notification record |
Northbeams covers the MEASURE function most fully (continuous monitoring). GOVERN and MANAGE rely on attestation; MAP is scoped to the AI inventory side that the four-surface telemetry produces.
| Function | Outcome | Status | Evidence source |
|---|---|---|---|
| MEASURE 2.7 | AI system performance monitored | AUTO | Per-tool, per-user telemetry |
| MEASURE 2.6 | AI risks and benefits documented | ATTEST | Quarterly risk register |
| MAP 4.1 | AI system inventory maintained | AUTO | Discovery refreshed continuously |
| MANAGE 3.1 | AI risks managed | ATTEST | Policy + change-management record |
Annexed to your existing Type II. AICPA's 2017 criteria don't name AI; we map to the controls auditors actually invoke when AI is in scope (CC6.1, CC7.2) plus selected CSA matrix entries.
| Control | Subject | Status | Evidence source |
|---|---|---|---|
| CC6.1 | Logical access (AI tools) | AUTO | Per-tool allow/warn/block + audit log |
| CC7.2 | Monitoring of system activity | AUTO | Continuous event log |
| CSA AI-04 | Data classification at capture | AUTO | On-device classifier |
| CSA AI-09 | Third-party AI risk | ATTEST | Vendor security review record |
05 / Pricing
The pack is included on the Compliance tier. Annual fee, flat, no per-seat math. Three options sized for what you actually need.
Starter
$12,000 / yr1 framework. Monthly pack. 1 GRC integration. 90-day retention.
Pro
$36,000 / yrAll frameworks. On-demand + scheduled. All integrations. 1-year retention.
Enterprise
$72,000+ / yrSSO, custom controls, 7-year retention, dedicated CSM, SLAs.
The sample is the real format with synthetic data. Reply with anything the auditor flags and we'll iterate together. Build partners are how we got the format right; we want more.