HECVAT Automation · Deep Dive
From 7–20 business days to a 20-minute draft risk matrix.
A HECVAT 4.1.6 review is 332 questions across 35 categories and 8 evaluation tabs, with a conditional 32-question AI subsection that no institutional team has muscle memory for yet. AegisGrid ingests the vendor artifact, extracts the controls, maps each to your published rubric (e.g., Rutgers' High / Medium-High / Medium-Low / Low taxonomy and Policy 70.1.2 classification), and produces a draft risk matrix for analyst sign-off. The analyst remains the decision-maker. AegisGrid makes the analyst faster.
§ 01 — The HECVAT 4 Reality
The math is unforgiving.
HECVAT 4.1.6 — the Higher Education Community Vendor Assessment Toolkit published by EDUCAUSE — is 332 questions organized into 35 categories across 8 evaluation tabs. For a Rutgers analyst working a third-party risk review under the ISO's published SLA of 7–20 business days, the math is unforgiving. Even at the optimistic end of the range, that is roughly 16–46 questions per business day that must be parsed, cross-referenced against internal policy, and risk-rated in the published taxonomy. The HECVAT 4 release (4.1.5 was published February 10, 2025; 4.1.6 is current) added a conditional AI subsection, which makes the workload larger — not smaller.
The HECVAT 4 AI subsection is 32 questions. It is conditional — meaning it activates when the vendor's product includes an AI / ML component — and it explicitly aligns to NIST AI RMF AIPL-05 controls. The institutional muscle memory for answering those 32 questions does not yet exist at most universities; the controls are newer than the analyst training programs that cover them. AegisGrid treats this subsection as a first-class workflow: each question is mapped to its NIST AI RMF lineage, and the extraction includes the citation chain that an auditor will demand.
332
HECVAT 4.1.6 questions across 35 categories and 8 tabs
32
Conditional AI subsection (NIST AI RMF AIPL-05)
7–20
Rutgers ISO SLA, business days per review
Sources: Rutgers ISO SLA — it.rutgers.edu/cybersecurity-risk-management-program/information-security-risk-management-frequently-asked-questions-faq/; EDUCAUSE HECVAT program page; HECVAT 4.1.6 release notes (4.1.5 published Feb 10, 2025; 4.1.6 current).
§ 02 — How AegisGrid Handles It
Five steps. The analyst stays in the driver's seat.
Each step is a discrete agent with explicit tool scope. Every action is written to the hash-chained audit log before the next step starts.
Ingest
AegisGrid accepts the vendor artifact in the formats security teams already have — SOC 2 Type II PDF, HECVAT XLSX, vendor security policy — via a ServiceNow webhook, an email drop, or a manual upload to the on-prem AegisGrid intake. Documents never leave the host. For the /proof evaluation flow, uploads are TLS 1.3 in transit, AES-256 at rest, and auto-deleted after 30 days unless you extend the retention window.
POST /api/intake { src: 'servicenow' | 'email' | 'upload', ttl: '30d' } → on-prem artifact store (AES-256)Extract
A schema-bound extraction agent parses the document against the published HECVAT 4.1.6 question schema. The agent runs against a local LLM (Llama 3.x or Mistral via Ollama or vLLM), so no inference request is sent to any commercial provider. Each extracted control statement is paired with its source location (page, section, paragraph) for lineage.
extractor.extract(doc, schema=HECVAT_4_1_6) → control[] { id, text, src: { page, section, paragraph } }Map
The mapping agent cross-references each extracted control against the relevant institutional policy clauses — Rutgers Policy 70.1.2 Restricted / Internal / Public data classification, GLBA Safeguards Rule controls, NIST 800-171 where applicable, and the local risk-rating taxonomy. The mapping output is a structured record per question, ready for analyst review.
mapper.link(control) → policy[] { '70.1.2 Restricted', 'GLBA 314.4(c)(4)', 'NIST 800-171 3.13.1' }Score
The scoring agent produces a draft risk rating in your published taxonomy (e.g., Rutgers' High / Medium-High / Medium-Low / Low scale) with a justification string for each rating. The justification cites the source document, the extracted control text, and the policy clause that drove the rating. Nothing here is a black-box score — every rating is replayable.
scorer.rate(control, mappings) → { rating: 'Medium-High', justification: string, evidence: ref[] }Human-in-Loop
The draft risk matrix lands in the analyst's existing ServiceNow queue. The analyst reviews, edits, accepts, or rejects each line. Final go / no-go authority stays with the human and the institution. AegisGrid never makes a vendor approval decision.
servicenow.create_ticket({ queue: 'TPRM', payload: draft_matrix, status: 'awaiting_analyst' })Human authority gate — agent cannot bypass
§ 03 — Compliance Lineage
Every control ships with a four-part lineage record.
Every extracted control ships with a four-part lineage record — source citation, policy mapping, risk-rating justification, and the audit-log hash that ties the entry to the prior chain. This lineage is what makes the output defensible in an internal review or an external audit. It is also what an analyst would have produced by hand — except in 20 minutes instead of 20 days.
- 01
Source document
Vendor artifact — SOC 2 Type II, HECVAT XLSX, security policy.
doc.pdf · sha256:ab12… - 02
Extracted control
Schema-bound extract, paired with page / section / paragraph.
control[Q-217] { … } - 03
Mapped policy
Rutgers Policy 70.1.2 · GLBA 314.4(c)(4) · NIST 800-171.
policy → '70.1.2 Restricted' - 04
Risk rating
Draft score in the published taxonomy with replayable justification.
rating: 'Medium-High'
§ 04 — Explicit Non-Claims
What AegisGrid doesn't do. Every one of these is binding.
No autonomous decisions
AegisGrid produces a draft risk matrix. A human analyst reviews, edits, and signs off. The system never makes a vendor go / no-go decision.
No commercial API egress
Inference runs locally on hardware you operate. No document, alert, or extracted control is sent to OpenAI, Anthropic, Google, or any other third-party API.
No replacement of the human analyst
AegisGrid reduces the time-to-draft; it does not replace the institutional knowledge the analyst brings to risk judgment.
No retention beyond the assessment record
Documents are retained for the active assessment window plus a configurable tail, then deleted. There is no shadow training corpus built from your vendor data.
§ 05 — Licensing
Three ways to license AegisGrid. Pricing on request for each.
Per-Assessment
Pay per completed HECVAT or SOC 2 review. Best fit for low-volume programs or for institutions that want to evaluate AegisGrid on a single real workload before committing. Contact for pricing.
Contact for pricing
Department Annual License
A single department or unit licenses AegisGrid for a 12-month period, with a defined annual assessment volume and full access to the on-prem or private-cloud deployment. Contact for pricing.
Contact for pricing
Institution-Wide Enterprise
Site-wide deployment, multi-department, on-prem or air-gapped topology, contractual SLA on inference availability and audit-log retention.
Includes the BAA template, NDA template, and dedicated evaluation engineer for the first 60 days. Contact for pricing.
Contact for pricing
Specific dollar figures are intentionally omitted. Per the brand honesty constraint, pricing is delivered one-to-one in a working session, not on a public page.
Ready to see it run on one of your real vendor artifacts?
The second link takes you to /proof, where you'll find the document-handling safety disclosures and the calendar embed for the working session.