VAJRAKSHSENTINEL Technical briefing ↗

Software assessment update assurance for defence evaluation

Evidence, held to account.

Understand the software being delivered, observe its tested behaviour, and retain the evidence behind its security assessment.

Review what suppliers deliver and what each update changes inside your own enclave, with a human making the call.

Fig. 1 — Access, weakness, asset. Every node carries its evidence label.Scroll to assess ↓

Assessment posture

The assessment system is designed around the conditions in which defence evaluators work.

01

Offline by design

Assessment, investigation and report review work without an external service.

02

Binary-first

Starts from the software that was delivered. Source and build records are optional extras.

03

Analyst-run validation

Analysts choose and run the tests. SENTINEL never launches attacks on its own.

04

Human decision

An authorized person keeps the deployment decision. There is no autopilot.

The evaluator's problem

Six connected questions every release review has to answer.

Dependency lists, scanner output, behaviour records and manual notes rarely answer them together. SENTINEL keeps them in one assessment record.

01 · Identity

Is this the exact release that was assessed?

02 · Composition

Which components can be identified, and what is still uncertain?

03 · Known risk

Which advisories may apply to those components and versions?

04 · Behaviour

What happened during the recorded installation and test scenarios?

05 · Attack paths

How could access, weaknesses and conditions combine to affect a protected asset?

06 · Change

What differs from the last assessed release, and does it matter?

Attack-path analysis

Understand the path from weakness to impact.

An advisory match says a component may be affected. It does not say whether an attacker can reach it, what access they would need, or what it would mean for a protected asset.

    No invented steps

    A vulnerability ID alone never creates an entry point, a privilege or an impact.

    Contradictions stay on screen

    A blocked route is recorded as blocked under those conditions.

    One step is not the chain

    Validating a single step never silently marks the whole path demonstrated.

    No score, no probability

    Analysts prioritise by evidence quality, access needed and consequence.

    End-to-end workflow

    From delivered release to reviewed dossier.

    An incomplete stage stays incomplete. A quiet run, missing evidence or a failed reproduction never quietly becomes a clean result.

    Evidence change

    Keep the evidence behind every conclusion.

    Reports and structured outputs come from the same assessment record, so what a reviewer reads is what the tools found.

    Assessment dossier Illustrative layout
    Input identities
    hashes · sizes · provenance
    Tools and advisory snapshot
    versions · freshness
    Environment and policy
    profile · scope
    Findings and attack paths
    step-level evidence · revision
    Validation outcomes
    captured · manually attached
    Coverage limits
    unsupported · unknown · not tested
    Reviewer decisions
    attributable · exceptions · history

    Evidence bundles can be signed so tampering is detectable. A signature protects integrity; it does not make every recorded claim true.

    Release comparison

    Candidate vs reference

    Changes in the software

    • Files and components
    • Permissions
    • Package contents
    • Observed behaviour
    • Advisory exposure
    • Path evidence prerequisites

    Changes in assessment conditions

    • Advisory snapshot
    • Tool versions
    • Policy
    • Configuration
    • Test coverage

    A new finding may come from new knowledge about unchanged software. SENTINEL does not call that a newly introduced defect.

    Deployment posture

    Designed for the evaluation enclave.

    SENTINEL assumes restricted connectivity from the start. Assessed software, customer evidence and results do not need to leave your environment.

    01

    Offline end to end

    Assessment, investigation, report generation and evidence review run with no external service or hidden cloud dependency.

    02

    Controlled offline updates

    Advisory data, tools and policy arrive through a reviewed offline process with recorded provenance.

    03

    Isolated execution

    Submitted software runs in disposable environments with no route to controller credentials or operational networks.

    04

    Separated responsibilities

    Submission, analysis, administration and review are distinct roles, configured to approved procedure.

    05

    Evidence stays inside

    No implicit upload of assessed content. Secrets are shown masked. Exports are controlled.

    06

    Tamper-evident records

    Modified bundles fail their integrity check. Every disposition and exception is attributable.

    Trust boundaries

    Clear about what it does and what it doesn't promise.

    SENTINEL does

    • Link every conclusion to the exact artifact and its evidence.
    • Keep unsupported content, unknowns and coverage gaps visible in the report.
    • Capture analyst-run validation with procedure, conditions and outcome.
    • Separate observed behaviour from permitted behaviour.
    • Bring data, tools and updates through a controlled offline process.

    SENTINEL does not

    • Guarantee it finds every unknown vulnerability or malicious logic.
    • Treat a CVE match as proof of exploitability.
    • Generate or run attacks on its own.
    • Authorize deployment or confer certification.
    • Replace the evaluator's judgment with an opaque safety score.

    Evaluation briefing

    Start with a technical briefing.

    A briefing walks evaluators through the workflow on a synthetic demonstration target and agrees what a representative evaluation would measure.

    Details are used only to arrange the briefing.