Evidence-Ledger & Red-Team Audit Protocol
Rating methodology
Every profile on MarTechRadar.ai combines timestamped captures of the vendor’s own website, automated Sentinel consistency checks (S2–S7), and an editor’s check of key facts against those captures. Each rated field is labelled with its provenance, so you can see whether a data point is a machine estimate, founder-submitted, or editor-checked against vendor sources. Most reviews first published in October 2026 were drafted with AI from the captures and checked by an editor; since November 2026 we have been re-researching and rewriting them one at a time. We do not run hands-on product trials unless a review says so.
What do Moat Score (1–5) and L1–L5 Autonomy mean?
Moat Score (1.0–5.0) measures how hard a software product would be to replace with a general-purpose AI model such as ChatGPT or Claude: 1.0–1.9 (Thin wrapper) means a lightweight prompt interface over third-party APIs; 2.0–3.4 (Workflow-embedded) means a tool wired into team approvals, CRM records, or publishing workflows; and 3.5–5.0 (Proprietary moat) means first-party crawlers, proprietary intent graphs, or deep system-of-record infrastructure.
AI Autonomy Levels (L1–L5) measure how much work the software executes without a human operator: L1 returns single-shot prompt outputs; L2 acts as a human-reviewed workflow copilot, editor, or analytics workbench; L3 runs rule-, schedule-, or webhook-triggered automation within fixed guardrails; L4 executes multi-step goals autonomously across external tools; and L5 coordinates self-optimising multi-agent systems.
1. First-party data collection & tiered evidence ledger
- Vendor page capture: Each profile is built from captures of the vendor’s homepage and pricing page, stored with SHA-256 content hashes. Re-researched reviews add captures of documentation, integration and other pages that support specific claims. Pricing pages are re-captured on a rolling schedule to detect price changes.
- Pricing normalisation (705 public URLs / 642 custom or enterprise): Commercial terms are normalised into billing model (subscription, per-seat, usage/credit, revenue share, custom enterprise quote, or open source), currency, entry monthly floor, and free-tier availability. Per-credit or metered micro-rates are strictly separated from monthly SaaS floors, and a numeric price is published only when anchored to a vendor URL on the product’s own root domain.
- Provenance badges (
Evidence-Ledger Verifiedvs.First-Party Snapshot Audited): Profiles with a LinkedIn or Crunchbase company link (sameAs) matched to the vendor’s domain carry the Evidence-Ledger Verifiedbadge; the rest carry First-Party Snapshot Audited. The badge describes how the company was identified, not that its company figures were confirmed: founding year, headquarters, funding, investors and acquirer were first collected from such public profiles and are being re-checked against primary sources (the vendor’s own site, press releases and filings). - Lifecycle tracking: Products whose domains fail repeated availability checks or merge into parent platforms are marked as unreachable or acquired; their profiles remain live with documented migration alternatives.
2. The AI Marketing Autonomy Standard (L1–L5)
To cut through inflated “autonomous agent” claims, editors classify each product’s runtime control plane on a five-level operational scale:
- L1 Prompt-assisted: The user supplies a manual prompt and the tool returns a single-shot output (copy, image, or query) from a foundation model with no state persistence or workflow orchestration.
- L2 Workflow copilot: Multi-step templates, canvas builders, and guided sequences where AI drafts intermediate assets, but a human operator must review, approve, and advance every stage.
- L3 Conditional automation: Event-, webhook-, or rule-triggered pipelines (such as waterfall lead enrichment, abandoned-cart triggers, or dynamic bid rules) that execute autonomously within deterministic guardrails while routing exceptions to humans.
- L4 Autonomous agent: End-to-end goal decomposition, external API/tool invocation, and closed-loop self-correction within a bounded GTM scenario (for example, autonomous outbound SDR prospecting or automated landing-page CRO testing).
- L5 Self-evolving swarm: Multi-agent negotiation, cross-channel budget reallocation, and continuous reinforcement learning across attribution and revenue outcomes with policy-level human governance.
Levels are assigned by an editor from the vendor’s own documentation. They are judgements, not hands-on tests. Unrated products display “Not yet rated”.
3. Moat tiers & quantitative score (1.0–5.0)
Each reviewed tool receives an editor-assigned moat tier and a score from 1.0 to 5.0 covering switching costs, data defensibility, and resilience against foundation model updates. When we have no product-specific reasoning for a score, the profile shows the score without a written rationale; treat such scores as provisional:
- Thin wrapper (1.0–1.9): Relies primarily on third-party LLM/media APIs or single-purpose browser overlays with minimal workflow state or proprietary data; highly exposed to upstream model commoditisation.
- Workflow-embedded (2.0–3.4): Built on general or hybrid models and integrated into customer operational workflows (CRM, CMS, commerce, analytics, or team approval pipelines), creating practical operational switching costs (
2.0–2.9standard,3.0–3.4strong ecosystem integration). - Proprietary moat (3.5–5.0): Combines proprietary crawlers, first-party identity or intent graphs, deep bidirectional system-of-record writeback, enterprise governance attestations, or specialised execution infrastructure (
3.5–4.2proprietary moat,4.3–5.0category-defining infrastructure).
4. Seat-based cost estimates & operational limitations
Headline starting prices rarely reflect what a 5- or 15-person GTM team actually pays. Each review lists the operational limitations we found in the vendor’s material (such as credit exhaustion under bulk enrichment, API rate limits, or annual lock-in clauses), and our Procurement Stress Tester applies published prices to different seat counts. Its figures are arithmetic on published prices, not quotes; where a vendor publishes no price, it shows none.
5. Strict editorial guardrails (what we never do)
- No fabricated star ratings: We never publish synthetic star ratings or
aggregateRatingschema markup without a statistically valid volume of moderated, verified buyer reviews. - Zero pay-to-play ranking manipulation: Affiliate commissions, badge embeds, and vendor partnerships never alter moat scores, autonomy ratings, or comparison table rankings. See our affiliate & independence disclosure.
- Sentinel gates: Every published review passes automated Sentinel checks (S2–S7) for agreement between the data card and the prose, pricing links that point at another vendor’s site, and banned promotional language.
6. Corrections, freshness & founder submissions
Pricing pages are re-captured on a rolling schedule, and a profile is re-checked when a price change is detected. Each profile shows the date its review was last signed off. Vendors and buyers can submit a new tool, claim an existing profile, or submit primary-source corrections, or email the Research Desk directly at hello@martechradar.ai (48-hour correction triage SLA). Verified updates are logged in our editorial audit trail and shipped in the next release.