AAgentProof
Methodology

Methodology changes

AgentProof maintains a deliberate Intelligence Radar cadence and a Methodology Change Log. The methodology is versioned, maintained, and updated as the AI-agent landscape evolves. This is not legal advice, not a legal approval, and not a regulatory decision.

Export runs entirely in your browser. No external service is called.

Current methodology

AgentProof scores how clearly an AI agent is described — purpose, scope, autonomy, controls, and known risks — and produces a deterministic scorecard with a softened readiness rating, AI Act-aware indicators, red flags, missing controls, prioritised recommendations, an executive decision layer, a context-pack guidance block, a reproducibility receipt, and (after Verify now) a downloadable Verification Report. Methodology changes are recorded in this Change Log against the Intelligence Radar cadence.

Methodology version stamp
0.329.90
Engine version
0.10.0
Context packs
1.2.2

AgentProof does not call any live AI provider for scoring; scoring runs on a deterministic, offline engine.

Methodology updates that draw on external research (e.g. regulator publications, model-provider changes) are recorded explicitly in this Change Log with a documented source attribution block carrying source type, source label, source reference, source published at (when known), source reviewed at, source url (only when a real reference exists), reviewed by role, impact assessment, and limitations. AgentProof does not run autonomous or continuous scanning. Approved public sources are checked on demand by a named reviewer; a validation read never changes anything, and nothing reaches the methodology without human approval.

Intelligence Radar cadence

Describes the cadence and product-impact framework AgentProof uses to keep its deterministic methodology current as the AI-agent landscape evolves. This content is a methodology process description; it is not a live scan, not a regulatory determination, and does not by itself perform any external fetch.

This static content does not perform any web call, database query, or external research. External research is recorded explicitly in the Methodology Change Log with a documented `source_type` and a real source reference; source checks are on-demand and human-reviewed, never autonomous.

  • Monthly curated review

    monthly (target)

    The target review cadence for most source categories, including AI vendor and platform changes, agent security and misuse patterns, Microsoft / Power Platform agent guidance, and cross-provider agent frameworks. Reviews are deliberate, on-demand and human-led; each item is classified by impact category and may produce a Methodology Change Log entry.

  • Quarterly curated review

    quarterly (target)

    The target review cadence for AI regulation and standards and enterprise governance sources, alongside the scheduled methodology review that revisits the deterministic engine outputs (rules, indicators, context packs, decision-layer wording). Eligible to ship a new minor methodology version.

  • Emergency review for material change

    ad-hoc

    Triggered when a monitored source category produces a material change (new regulation, major model-provider behaviour change, recurring incident pattern) before the next scheduled review. Out-of-cycle methodology updates land here.

Monitored domains
  • AI regulation and standardsEU AI Act consolidated text and delegated acts, NIST AI Risk Management Framework, ISO/IEC 42001 (AI management systems), sector-specific AI guidance (financial, health, public sector)
  • AI vendor / platform changesofficial model release notes, deprecation calendars, safety-mode documentation
  • Agent security + misuse patternspeer-reviewed agent-security research, vendor security advisories, documented incident write-ups
  • Enterprise governance trendsindustry analyst frameworks, peer enterprise governance write-ups, audit and assurance professional bodies
  • Microsoft / Power Platform agent guidanceMicrosoft Learn Power Platform guidance, Microsoft Copilot Studio documentation, Microsoft Trust Center publications
  • Cross-provider agent frameworksframework release notes, reference architecture changes, notable third-party integrations
Product-impact categories
  • No action
  • Watch item
  • Context-pack update
  • Evidence expectation update
  • Scenario-test update
  • Red-flag rule update
  • AI Act-aware indicator update
  • Decision-layer update
  • Report wording update
  • New context pack required
  • Training / documentation update
  • Re-score recommended

Recent improvements

A curated selection of the product and methodology improvements that matter most to the people who rely on AgentProof reports. Newest first. The full methodology record is versioned and maintained behind every assessment.

  1. Scoring calibration for Microsoft and common agents

    5 Jun 2026

    The scoring methodology was recalibrated for Microsoft and common out-of-the-box agents so readiness ratings track real-world readiness more closely.

    Re-scoring recommended to pick up this improvement.

  2. Agent estate dashboard

    9 May 2026

    A unified estate dashboard shows every discovered agent in one place, with portfolio-level risk and remediation in a single view.

  3. Premium readiness report experience

    8 May 2026

    The readiness report was rebuilt into a clearer, decision-grade experience with stronger evidence, risk and recommendation sections.

  4. Connector-first discovery

    3 May 2026

    AgentProof discovers your agents through a guided connector first, so you do not have to describe each one by hand.

  5. Read-only Microsoft and Copilot Studio discovery

    3 May 2026

    AgentProof can connect read-only to Microsoft Power Platform and Copilot Studio to discover the agents you already run.

  6. Downloadable verification report

    26 Apr 2026

    Verification can now be exported as a clean, shareable report to attach to a governance review or audit file.

  7. Intelligence Radar and Methodology Change Log

    26 Apr 2026

    AgentProof now maintains a deliberate watch on the AI-agent landscape and records every methodology change in a versioned, public change log.

  8. Per-scorecard methodology lineage

    26 Apr 2026

    Every saved scorecard shows which methodology version it was produced under and whether re-scoring is recommended.

  9. Source attribution for methodology changes

    26 Apr 2026

    Methodology updates that draw on external research now carry a clear, friendly record of the source, who reviewed it, and the assessed impact.

  10. Export the Methodology Change Log

    26 Apr 2026

    The full Methodology Change Log can be copied or downloaded as a reviewer-ready document for governance and audit reviews.

  11. Verify a saved scorecard

    25 Apr 2026

    A one-click verification step and integrity badge let you confirm a saved scorecard has not been altered.

  12. Reproducibility receipt

    20 Apr 2026

    Each scorecard carries a reproducibility receipt so you can prove which methodology version produced a given result.

    Re-scoring recommended to pick up this improvement.

  13. Improvement timeline

    10 Apr 2026

    An improvement timeline tracks how an agent's readiness has changed over successive assessments.

  14. Compare two scorecards

    15 Mar 2026

    You can now compare any two scorecards side by side to see exactly what improved between assessments.

  15. Executive decision layer

    20 Feb 2026

    Every report now opens with a clear executive read-out so a decision-maker can see the readiness picture at a glance before the detail.

  16. Deeper coverage for finance, support, HR, public-facing and Microsoft Copilot agents

    1 Feb 2026

    Five common agent domains were expanded into full, production-ready evidence and scenario sets so reports speak directly to those use cases.

    Re-scoring recommended to pick up this improvement.

  17. Context-aware scoring

    15 Jan 2026

    Reports now detect the kind of agent under review and tailor the evidence and test scenarios to that context, instead of giving generic advice.

When to re-score

Saved documentation packs remain valid against the methodology version they were generated under. Re-score an agent when a Change Log entry above is marked Re-score recommended, when you have made meaningful edits to the agent description, or when you want a fresh reproducibility receipt under the current engine and content versions.

Each saved documentation pack also shows a Methodology lineage panel on its results page. The lineage panel displays which changes shipped at or before that pack's engine + context-pack versions, how many newer changes have shipped since, and whether re-scoring is recommended.

This is not legal advice, not a legal approval, and not a regulatory decision.