Stripe Radar vs Sift

Agent detection

1. ai-native userThe product distinguishes malicious bots from legitimate AI buying agents, so agent-driven purchases aren't blanket-blocked as fraud

   weight 2 · round drawn
   Stripe Radar—none0/10

   The evidence describes Radar's general fraud rules, risk scoring, reviews, and lists, but nothing addresses distinguishing legitimate AI purchasing agents from malicious bots — an axis specific to agentic commerce that is plausible for a fraud-prevention product but unevidenced here. Community feedback even shows false positives blocking legitimate low-risk customers, with no mention of agent-specific allowlisting or detection.

   • [claimed-docs] “Radar Standard: Out-of-the-box fraud protection for all payment methods to detect and prevent transaction fraud, and identify fraudulent acc…”
   • [claimed-docs] “Risk settings let you balance authorization and fraud on your account by using risk controls.”
   • [community] “We are seeing an opposite side: customers using Stripe that had very low fraud rates previously are now getting more false positives causing…”
   • [community] “We got a couple cases of false positives ourselves, and the Stripe UI wasn't very clear that we couldn't override the 'block' (the button wa…”

   proof ↗⚑ flag

   Sift—none0/10

   Evidence covers general fraud scoring, workflows, and decisions but nothing addresses distinguishing AI shopping agents from malicious bots or agentic-commerce specific handling; missing for 10: any mention of AI agent traffic detection, agent-vs-bot classification, or agentic-commerce policy/whitelisting.

   ⚑ flag

Agent identity

1. developerPass verified agent identity — agentic-payment protocols, signed agent tokens, delegated spending scopes — into the risk decision as a first-class signal

   weight 2 · round drawn
   Stripe Radar—none0/10

   No evidence that Radar accepts verified agent identity, agentic-payment protocol tokens, signed agent tokens, or delegated spending scopes as first-class risk inputs; Radar's documented signals are card, customer, IP, and rule/list based, not agent-identity based.

   • [claimed-docs] “you can also set up rules that are unique to your business using the supported attributes”
   • [claimed-docs] “Value lists allow you to group values together which can then be referenced in rules.”
   • [claimed-docs] “Risk settings let you balance authorization and fraud on your account by using risk controls.”

   proof ↗⚑ flag

   Sift—none0/10

   No evidence Sift supports ingesting verified agent identity, agentic-payment protocol tokens, signed agent tokens, or delegated spending scopes as risk signals; docs only describe generic user/business event and custom-field ingestion. missing for 10: any mention of agentic-commerce protocols, signed agent tokens, or delegated spending scope support as a risk signal.

   ⚑ flag

Agent access

1. ai-native userPoint an agent at llms.txt or agent-oriented docs

   weight 2 · round to Stripe Radar
   Stripe Radar✓fullprobed9/10

   Probes confirm a live llms.txt at docs.stripe.com/llms.txt (HTTP 200) and markdown-rendered agent-friendly docs pages (e.g. radar.md), directly enabling an agent to be pointed at agent-oriented docs. Missing for 10: no independent/community confirmation of agents actually consuming these docs successfully in practice.

   • [probe] “PROBE llms.txt: HTTP 200 at https://docs.stripe.com/llms.txt # Stripe Documentation When installing Stripe packages, always check the npm r…”
   • [probe] “PROBE docs-md: HTTP 200 at https://docs.stripe.com/radar.md # Radar Use Stripe Radar to protect your business against fraud. ## Get starte…”

   proof ↗⚑ flag

   Sift—none0/10

   No evidence of an llms.txt file or agent-oriented documentation format; the docs are standard developer API references with no mention of AI-agent consumption. missing for 10: llms.txt file, agent-readable docs format, any mention of AI-agent-targeted documentation.

   ⚑ flag

2. ai-native userRun the product headlessly / in CI for automation

   weight 2 · round to Stripe Radar
   Stripe Radar~partialprobed5/10

   Radar exposes REST API endpoints (rules, value lists, reviews approve/decline) and has dedicated testing docs with test card numbers for automated fraud-rule verification, and Stripe ships an official CLI — all of which support headless/CI use. However, there's no explicit CI/automation guide, and a community report explicitly flags 'lack of easy programmatic control' as a pain point for adjusting Radar decisions. missing for 10: dedicated CI/headless workflow documentation, explicit automation examples, and resolution of the programmatic-control complaint.

   • [claimed-docs] “POST /v1/reviews/:id/approve”
   • [claimed-docs] “4000000000004954 | Results in a charge with a risk level of `highest`”
   • [probe] “official CLI documented at https://docs.stripe.com/stripe-cli”
   • [community] “We are seeing an opposite side: customers using Stripe that had very low fraud rates previously are now getting more false positives causing…”

   proof ↗⚑ flag

   Sift~partialclaimed4/10

   Sift is fundamentally API-driven (REST Events/Decisions APIs, SDKs) which is inherently 'headless' and can be called from backend or automated systems, but there is no explicit evidence of a CLI, CI pipeline integration, or documented headless automation workflow. Missing for 10: explicit CI/CD examples, CLI tooling, or automation-specific documentation beyond generic API integration.

   • [claimed-docs] “Sift easily integrates into your existing systems using modern REST APIs, Javascript snippet, and SDKs for iOS and Android.”
   • [claimed-docs] “What actions your users are taking, usually key user lifecycle events (e.g., creating an account, placing an order, posting content to other…”
   • [claimed-docs] “What actions your business is taking in response to users (e.g., approve an order, block and event due to fraud, cancel order due to chargeb…”
   • [claimed-docs] “Because Sift gets smarter the more data it has about your business, you can jump start your integration by backfilling a few months worth of…”

   proof ↗⚑ flag

3. ai-native userConnect an agent via an official MCP server

   weight 3 · round to Stripe Radar
   Stripe Radar✓fullprobed8/10

   Stripe documents an official MCP server (docs.stripe.com/mcp) that lets AI agents connect to Stripe, which as a platform encompasses Radar functionality via its API. Missing for 10: Radar-specific MCP tool examples/independent hands-on corroboration of agent use.

   • [probe] “official MCP server documented at https://docs.stripe.com/mcp”

   proof ↗⚑ flag

   Sift—none0/10

   Sift is a fraud-detection SaaS with REST/Decisions APIs and SDKs, but no evidence of an official MCP server for agent connectivity. missing for 10: any mention of MCP, agent integration protocol, or official MCP server endpoint.

   • [claimed-docs] “Sift easily integrates into your existing systems using modern REST APIs, Javascript snippet, and SDKs for iOS and Android.”

   proof ↗⚑ flag

4. ai-native userUse an official CLI

   weight 2 · round to Stripe Radar
   Stripe Radar✓fullprobed8/10

   Stripe ships an official Stripe CLI (docs.stripe.com/stripe-cli) that covers Radar-related API/webhook workflows, giving AI-native users a scriptable interface. Missing for 10: no direct evidence the CLI has Radar-specific commands or independent hands-on confirmation of its use in agentic workflows.

   • [probe] “official CLI documented at https://docs.stripe.com/stripe-cli”

   proof ↗⚑ flag

   Sift—none0/10

   Sift's evidence pack covers REST APIs, SDKs, JS snippet, and Decisions/Workflows APIs, but no official CLI tool is mentioned anywhere in docs or community sources.

   ⚑ flag

5. ai-native userDrive the product through a documented public API

   weight 3 · round to Stripe Radar
   Stripe Radar✓fullprobed8/10

   Stripe Radar exposes documented REST API endpoints (e.g. reviews, early_fraud_warnings, value_lists) and is part of Stripe's broader public API with an official CLI and MCP server, confirming programmatic/agentic access. missing for 10: no discoverable OpenAPI/swagger spec file was found (probe 404s) and no independent hands-on report of an AI agent driving Radar specifically via the API.

   • [claimed-docs] “POST /v1/reviews/:id/approve”
   • [claimed-docs] “An early fraud warning indicates that the card issuer has notified us that a charge may be fraudulent.”
   • [claimed-docs] “Value lists allow you to group values together which can then be referenced in rules.”
   • [probe] “PROBE docs-md: HTTP 200 at https://docs.stripe.com/radar.md # Radar Use Stripe Radar to protect your business against fraud. ## Get starte…”
   • [probe] “official MCP server documented at https://docs.stripe.com/mcp”
   • [probe] “official CLI documented at https://docs.stripe.com/stripe-cli”
   • [probe] “PROBE openapi: all candidate paths 404 (https://docs.stripe.com/openapi.json, https://docs.stripe.com/swagger.json, https://docs.stripe.com/…”

   proof ↗⚑ flag

   Sift~partialcommunity6/10

   Sift documents REST APIs (Events, Decisions, Score, Workflows) that let developers programmatically drive fraud scoring and decisioning, which supports an AI-native integration story. However, this is a traditional fraud-ops API, not one purpose-built for AI agent orchestration, and there's no evidence of agent-specific tooling like an MCP server, function-calling schemas, or SDKs for LLM agents. missing for 10: agent-oriented API framing (e.g., MCP server, function-calling schema), independent hands-on validation of API usability beyond one user noting confusing docs.

   • [claimed-docs] “Sift easily integrates into your existing systems using modern REST APIs, Javascript snippet, and SDKs for iOS and Android.”
   • [claimed-docs] “What actions your users are taking, usually key user lifecycle events (e.g., creating an account, placing an order, posting content to other…”
   • [claimed-docs] “What actions your business is taking in response to users (e.g., approve an order, block and event due to fraud, cancel order due to chargeb…”
   • [claimed-docs] “Sift represents this risk with a score between 0 and 100, where risky events have higher scores.”
   • [claimed-docs] “Workflows enable you to make risk-based decisions in real time during key events. Follow this guide to create your own custom Workflow.”
   • [community] “Early user feedback: 'I have been integrating it for a day or so. The documentation is slightly confusing and they've had a few minor bugs i…”

   proof ↗⚑ flag

6. ai-native userBuild against official SDKs

   weight 2 · round to Stripe Radar
   Stripe Radar~partialprobed5/10

   The evidence shows official API endpoints (e.g., POST /v1/reviews/:id/approve, early_fraud_warnings API) and an official CLI, implying SDK-compatible API access, but there is no explicit documentation pack entry naming or linking official language SDKs (e.g., stripe-node, stripe-python) for Radar-specific features. missing for 10: explicit SDK documentation/references, code samples showing SDK usage for Radar rules/reviews, independent developer confirmation of SDK coverage.

   • [claimed-docs] “POST /v1/reviews/:id/approve”
   • [claimed-docs] “An early fraud warning indicates that the card issuer has notified us that a charge may be fraudulent.”
   • [probe] “official CLI documented at https://docs.stripe.com/stripe-cli”
   • [probe] “PROBE docs-md: HTTP 200 at https://docs.stripe.com/radar.md # Radar Use Stripe Radar to protect your business against fraud. ## Get starte…”

   proof ↗⚑ flag

   Sift—none0/10

   Evidence shows Sift has REST APIs, JS snippet, and mobile SDKs for iOS/Android, but there's no mention of SDKs oriented toward AI-native development (e.g., LLM/agent SDKs, official Python/Node/AI framework SDKs). This story concerns AI-native developer tooling, which is not addressed anywhere in the pack. missing for 10: any evidence of AI-agent-oriented SDKs, LLM integration libraries, or agent framework support.

   ⚑ flag

7. ai-native userSubscribe to events via webhooks

   weight 2 · round drawn
   Stripe Radar—none0/10

   The evidence pack documents Radar's REST APIs (reviews, early fraud warnings, value lists) but never mentions webhook event subscriptions for these Radar events, so there's no evidence of an AI-native webhook subscription capability despite this being a plausible axis for an API-driven fraud product.

   ⚑ flag

   Sift—none0/10

   Sift's documented integration model is inbound REST APIs (Events, Decisions, Score, Workflows) for sending data to Sift, not outbound webhooks for subscribing to events from Sift. No evidence pack item mentions webhook subscriptions or event push notifications to external consumers.

   ⚑ flag

Agentic features

1. ai-native userGet AI-generated insights and suggestions from my data inside the product

   weight 2 · round drawn
   Stripe Radar—none0/10

   Radar's docs describe rule-based fraud controls, risk scoring, and dashboard analytics/visualizations (docs-4, docs-12), but there is no evidence of AI-generated natural-language insights or suggestions (e.g., an assistant summarizing fraud trends or recommending rule changes) surfaced inside the product.

   ⚑ flag

   Sift—none0/10

   Sift provides risk scores and rule-based Workflows/Decisions derived from ML models, but there is no evidence of AI-generated natural-language insights or suggestions surfaced to users inside a product UI—everything is API-driven scoring and automation rather than generative/agentic insight delivery. Missing for 10: any documented AI-generated narrative insights, recommendations, or conversational/agentic assistant surfaced in-product, and any independent confirmation of such a feature.

   • [claimed-docs] “Sift represents this risk with a score between 0 and 100, where risky events have higher scores.”
   • [claimed-docs] “Workflows enable you to make risk-based decisions in real time during key events. Follow this guide to create your own custom Workflow.”
   • [claimed-docs] “Sift offers Workflows, a powerful rules automation platform, and Review Queues.”

   proof ↗⚑ flag

2. ai-native userSet up automations that run autonomously in the background

   weight 2 · round to Stripe Radar
   Stripe Radar~partialcommunity6/10

   Radar rules engine runs autonomously in the background on every transaction (auto 3DS requests, auto-pause payouts, custom rules, auto-allow trusted customers, risk settings) per docs-1/2/3/13/14, which is genuine unattended automation. However community feedback notes limited programmatic control over these automations and friction when trying to override or fine-tune them (stripe-radar-comm-2, stripe-radar-comm-3), suggesting the autonomy is somewhat constrained/dashboard-centric rather than fully agent-friendly. missing for 10: evidence of API-driven/programmatic rule creation or agent-triggered automation workflows, and resolution of the 'lack of easy programmatic control' complaint.

   • [claimed-docs] “Request 3D Secure (3DS) for all payments that support it and are made by a new customer”
   • [claimed-docs] “Review and automatically pause payouts on accounts that have a high dispute rate”
   • [claimed-docs] “you can also set up rules that are unique to your business using the supported attributes”
   • [claimed-docs] “Risk settings let you balance authorization and fraud on your account by using risk controls.”
   • [claimed-docs] “Customer IDs for trusted customers: Use this list to automatically allow payments by these customers.”
   • [community] “We are seeing an opposite side: customers using Stripe that had very low fraud rates previously are now getting more false positives causing…”
   • [community] “We got a couple cases of false positives ourselves, and the Stripe UI wasn't very clear that we couldn't override the 'block' (the button wa…”

   proof ↗⚑ flag

   Sift~partialclaimed4/10

   Sift's Workflows feature is described as a 'rules automation platform' that runs risk-based decisions autonomously in real time on incoming events, which fits the general notion of background automations. However, this is traditional rules/ML-based fraud automation, not an AI-native/agentic automation-building experience tailored to an 'AI-native user.' Missing for 10: evidence of AI-native automation authoring (e.g., natural-language or agent-driven workflow creation), broader use-case automations beyond fraud decisioning, and independent confirmation of autonomous operation quality.

   • [claimed-docs] “Workflows enable you to make risk-based decisions in real time during key events. Follow this guide to create your own custom Workflow.”
   • [claimed-docs] “Sift offers Workflows, a powerful rules automation platform, and Review Queues.”
   • [claimed-docs] “Learn to set up custom Decisions, like Ban Account or Cancel Order, that are connected to real business actions in your backend.”

   proof ↗⚑ flag

3. ai-native userDelegate tasks to a built-in AI assistant inside the product

   weight 3 · round drawn
   Stripe Radar—none0/10

   No evidence of a built-in AI assistant within Stripe Radar to which users can delegate tasks; the product offers rules, lists, reviews, and analytics but no conversational/agentic assistant feature is documented.

   ⚑ flag

   Sift—none0/10

   The axis applies to this product kind (peer products hold positive or none verdicts on this story), so lack of evidence for an applicable capability is "none", never "na". (na/none harmonized at arena bring-up — see pipeline/scripts/na-harmonize.ts.)

   ⚑ flag

4. ai-native userOperate the product with natural-language commands

   weight 2 · round drawn
   Stripe Radar—none0/10

   Radar's rule configuration is a structured DSL (attributes/expressions) rather than natural-language commands, and while Stripe has a generic MCP server (stripe-radar-probe-4), there is no evidence it exposes Radar-specific fraud rule management or that Radar can be operated via free-form NL instructions.

   • [claimed-docs] “you can also set up rules that are unique to your business using the supported attributes”
   • [claimed-docs] “Value lists allow you to group values together which can then be referenced in rules.”
   • [probe] “official MCP server documented at https://docs.stripe.com/mcp”

   proof ↗⚑ flag

   Sift—none0/10

   The axis applies to this product kind (peer products hold positive or none verdicts on this story), so lack of evidence for an applicable capability is "none", never "na". (na/none harmonized at arena bring-up — see pipeline/scripts/na-harmonize.ts.)

   ⚑ flag

Api quality

1. ai-native userExplore an interactive API reference with runnable examples

   weight 2 · round drawn
   Stripe Radar—none0/10

   No evidence of an interactive API reference with runnable examples; the OpenAPI probe returned 404s and docs listed are static markdown pages describing endpoints without runnable/interactive playground features. missing for 10: interactive API explorer/playground, runnable code examples, OpenAPI spec availability.

   • [probe] “PROBE openapi: all candidate paths 404 (https://docs.stripe.com/openapi.json, https://docs.stripe.com/swagger.json, https://docs.stripe.com/…”

   proof ↗⚑ flag

   Sift—none0/10

   Evidence shows REST/Decisions/Events API docs exist but nothing describes an interactive API reference with runnable examples (e.g., embedded sandbox, try-it console). Missing for 10: interactive API explorer, runnable code examples, sandbox/try-it functionality.

   ⚑ flag

2. ai-native userDownload a machine-readable API spec (OpenAPI or equivalent)

   weight 2 · round drawn
   Stripe Radar—none0/10

   The evidence pack includes a direct probe for OpenAPI/swagger spec files at common Stripe docs paths, all returning 404, and no other citation shows a downloadable machine-readable spec for Radar's API. Only docs pages and llms.txt-style markdown are confirmed, not an OpenAPI/Swagger file.

   • [probe] “PROBE openapi: all candidate paths 404 (https://docs.stripe.com/openapi.json, https://docs.stripe.com/swagger.json, https://docs.stripe.com/…”

   proof ↗⚑ flag

   Sift—none0/10

   Evidence shows Sift has REST APIs (Events, Decisions, Score, Workflows) with docs, but no mention of a downloadable OpenAPI spec or other machine-readable API schema anywhere in the pack.

   • [claimed-docs] “Sift easily integrates into your existing systems using modern REST APIs, Javascript snippet, and SDKs for iOS and Android.”
   • [claimed-docs] “What actions your users are taking, usually key user lifecycle events (e.g., creating an account, placing an order, posting content to other…”
   • [claimed-docs] “What actions your business is taking in response to users (e.g., approve an order, block and event due to fraud, cancel order due to chargeb…”
   • [claimed-docs] “Workflows enable you to make risk-based decisions in real time during key events. Follow this guide to create your own custom Workflow.”

   proof ↗⚑ flag

3. ai-native userTest against a sandbox environment without touching production data

   weight 1 · round to Stripe Radar
   Stripe Radar~partialprobed7/10

   Stripe Radar's testing docs explicitly provide test card numbers (e.g., 4000000000004954) that simulate specific risk levels, enabling developers to validate fraud rules and review logic without using real transactions or production data. However, the evidence doesn't detail a full sandbox environment (e.g., test-mode API key isolation, sandbox dashboards) beyond these test cards, and there's no independent/hands-on confirmation of sandbox fidelity for AI-native workflows. Missing for 10: explicit documentation of test-mode/live-mode key separation for Radar, broader sandbox environment description, and independent corroboration of safe non-production testing.

   • [claimed-docs] “4000000000004954 | Results in a charge with a risk level of `highest`”
   • [probe] “PROBE docs-md: HTTP 200 at https://docs.stripe.com/radar.md # Radar Use Stripe Radar to protect your business against fraud. ## Get starte…”

   proof ↗⚑ flag

   Sift—none0/10

   No evidence pack item mentions a sandbox environment, test mode, or any way to test Sift integrations without touching production data; all docs reference live Events/Decisions/Score APIs. Missing for 10: sandbox/test environment docs, test API keys or staging mode, any mention of separating test vs production data.

   ⚑ flag

4. ai-native userRely on versioned APIs with a documented deprecation policy

   weight 2 · round drawn
   Stripe Radar—none0/10

   The evidence pack contains no mention of Stripe API versioning scheme, version pinning, or a documented deprecation policy for Radar's API endpoints; only generic docs and community pricing/false-positive discussions are present.

   ⚑ flag

   Sift—none0/10

   Evidence shows Sift has REST APIs, SDKs, and docs but no mention of API versioning scheme or a documented deprecation policy anywhere in the pack. Missing for 10: explicit API versioning documentation, deprecation policy or changelog, migration timelines for breaking changes.

   ⚑ flag

1. ai-native userPerform bulk operations across many items at once

   weight 2 · round to Stripe Radar
   Stripe Radar~partialclaimed5/10

   Radar rules, value lists, and targeted transaction-review lists let a rule or list apply automatically across many transactions/customers at once (stripe-radar-docs-1,2,3,10,11,14), which is a form of bulk automation, but the documented Review API only exposes per-item actions (POST /v1/reviews/:id/approve) with no bulk/batch endpoint or explicit multi-item API call shown. missing for 10: evidence of a bulk API endpoint for approving/declining multiple reviews or disputes at once, and any AI-native tooling for programmatic multi-item operations.

   • [claimed-docs] “Request 3D Secure (3DS) for all payments that support it and are made by a new customer”
   • [claimed-docs] “you can also set up rules that are unique to your business using the supported attributes”
   • [claimed-docs] “Value lists allow you to group values together which can then be referenced in rules.”
   • [claimed-docs] “You can create a targeted list of payments to review with criteria that you specify, and review them in the Dashboard.”
   • [claimed-docs] “Customer IDs for trusted customers: Use this list to automatically allow payments by these customers.”
   • [claimed-docs] “POST /v1/reviews/:id/approve”

   proof ↗⚑ flag

   Sift~partialclaimed3/10

   Sift documents backfilling months of historical data as a bulk-load capability, but there is no evidence of a dedicated bulk API for batch-scoring, batch-deciding, or bulk-updating many items at once (e.g., no batch endpoint documentation). Missing for 10: explicit bulk/batch API documentation, batch decision or batch scoring examples, and any hands-on evidence of bulk workflows being used successfully.

   • [claimed-docs] “Because Sift gets smarter the more data it has about your business, you can jump start your integration by backfilling a few months worth of…”
   • [claimed-docs] “What actions your users are taking, usually key user lifecycle events (e.g., creating an account, placing an order, posting content to other…”
   • [claimed-docs] “What actions your business is taking in response to users (e.g., approve an order, block and event due to fraud, cancel order due to chargeb…”

   proof ↗⚑ flag

2. ai-native userDefine rules that trigger actions automatically on events

   weight 3 · round to Stripe Radar
   Stripe Radar~partialcommunity6/10

   Radar's rule engine clearly supports defining conditional rules that trigger automatic actions (3DS challenge, block, review, allow, pause payouts) based on transaction attributes, and value lists let rules be parameterized (docs-1, docs-2, docs-3, docs-10, docs-13, docs-14). However, rule authoring is Dashboard-centric with no documented rules-creation API, and community feedback explicitly flags a 'lack of easy programmatic control' plus restricted access to Allow Rules for newer accounts (comm-2, comm-7), limiting fit for an AI-native/automated workflow. Missing for 10: a documented API/SDK for programmatically creating or updating rules, and evidence of unrestricted automation access for all account types.

   • [claimed-docs] “Request 3D Secure (3DS) for all payments that support it and are made by a new customer”
   • [claimed-docs] “Review and automatically pause payouts on accounts that have a high dispute rate”
   • [claimed-docs] “you can also set up rules that are unique to your business using the supported attributes”
   • [claimed-docs] “Value lists allow you to group values together which can then be referenced in rules.”
   • [claimed-docs] “Risk settings let you balance authorization and fraud on your account by using risk controls.”
   • [claimed-docs] “Customer IDs for trusted customers: Use this list to automatically allow payments by these customers.”
   • [community] “We are seeing an opposite side: customers using Stripe that had very low fraud rates previously are now getting more false positives causing…”
   • [community] “Allow Rules, if not implemented properly, could open a vector for fraud. That's why it's not enabled for newer businesses on Stripe—we ask b…”

   proof ↗⚑ flag

   Sift!disputedcontradicted5/10

   Sift's docs clearly describe a 'Workflows' rules automation platform that triggers real-time Decisions (Ban Account, Cancel Order) based on event risk scores, directly matching the story. However, a Sift co-founder is quoted saying 'we do not have any rules in our product... rules are rather easy for fraudsters to circumvent,' directly contradicting the vendor's later rules-automation framing. missing for 10: reconciliation of this contradiction and independent hands-on confirmation that customer-defined Workflow rules reliably trigger automated actions in production.

   • [claimed-docs] “Workflows enable you to make risk-based decisions in real time during key events. Follow this guide to create your own custom Workflow.”
   • [claimed-docs] “Learn to set up custom Decisions, like Ban Account or Cancel Order, that are connected to real business actions in your backend.”
   • [claimed-docs] “Sift offers Workflows, a powerful rules automation platform, and Review Queues.”
   • [community] “CEO explains no-rules ML approach: 'we do not have any rules in our product... rules are rather easy for fraudsters to circumvent, and they …”

   proof ↗⚑ flag

Guarantee

1. finance leadShift fraud liability to the vendor — a chargeback guarantee that reimburses approved-then-disputed orders, with clear coverage terms

   weight 2 · round drawn
   Stripe Radar—none0/10

   The evidence pack covers Radar's fraud-scoring, rules, reviews, and dispute-rate monitoring features, but contains no mention of a chargeback guarantee, reimbursement for approved-then-disputed orders, or liability shift terms — that is a distinct Stripe product (Chargeback Protection), not documented here as part of Radar.

   ⚑ flag

   Sift—none0/10

   No evidence anywhere in the pack mentions a chargeback guarantee, liability shift, reimbursement for approved-then-disputed orders, or coverage terms — Sift is documented only as a fraud-scoring/risk-decision API, not an insurance-backed guarantee product.

   ⚑ flag

Outcome reporting

1. finance leadSee the numbers that matter — dispute rate, false-positive rate, approval-rate lift, review workload — and export them for the board

   weight 2 · round to Stripe Radar
   Stripe Radar~partialclaimed4/10

   Docs show dispute-rate calculation on the Radar dashboard (stripe-radar-docs-7) and fraud-rate/volume trend visualizations (stripe-radar-docs-4), plus reviewable queues (stripe-radar-docs-11, docs-8) that imply review workload tracking. However there is no evidence of a false-positive-rate metric, approval-rate lift measurement, or any export/reporting feature for board consumption. missing for 10: false-positive rate metric, approval-rate lift metric, CSV/board export capability, independent confirmation of dashboard completeness.

   • [claimed-docs] “We show this calculation on the Radar page in the Dashboard.”
   • [claimed-docs] “Visualize trends in transaction volume and fraud rates over time.”
   • [claimed-docs] “You can create a targeted list of payments to review with criteria that you specify, and review them in the Dashboard.”
   • [claimed-docs] “POST /v1/reviews/:id/approve”

   proof ↗⚑ flag

   Sift—none0/10

   Evidence covers APIs, scoring, workflows, and general product sentiment, but there is no mention of finance-lead dashboards, dispute-rate/false-positive/approval-lift metrics, review workload reporting, or board-ready export capability. missing for 10: dispute-rate and false-positive-rate metrics reporting, approval-rate lift analytics, review workload/queue metrics, export or reporting functionality for finance/board consumption.

   ⚑ flag

Representment

1. ops userChargeback responses are automated — evidence compiled from order, delivery, and session data and submitted to the issuer without manual copy-paste

   weight 3 · round drawn
   Stripe Radar—none0/10

   Radar's documented capabilities are fraud scoring, rules, reviews, and dispute-rate monitoring/analytics (docs-2, docs-4, docs-7, docs-9) — none of the evidence describes compiling order/delivery/session evidence and auto-submitting it to card issuers for chargeback responses. This is a distinct dispute-evidence-automation capability that the evidence pack simply does not show Radar performing.

   • [claimed-docs] “We show this calculation on the Radar page in the Dashboard.”
   • [claimed-docs] “An early fraud warning indicates that the card issuer has notified us that a charge may be fraudulent.”
   • [claimed-docs] “Review and automatically pause payouts on accounts that have a high dispute rate”

   proof ↗⚑ flag

   Sift—none0/10

   The axis applies to this product kind (peer products hold positive or none verdicts on this story), so lack of evidence for an applicable capability is "none", never "na". (na/none harmonized at arena bring-up — see pipeline/scripts/na-harmonize.ts.)

   ⚑ flag

Agent operations

1. ai-native userAn agent can read my fraud posture and manage rules and lists programmatically — propose a velocity rule, update a blocklist — with human approval gates

   weight 3 · round to Stripe Radar
   Stripe Radar!disputedcontradicted4/10

   Stripe exposes APIs for value lists (blocklists/allowlists) and review approval (a human-approval gate: POST /v1/reviews/:id/approve), plus a documented MCP server, which together could let an agent read fraud data and manage lists with approval steps. However, rule creation/velocity-rule authoring is Dashboard-centric with no documented rules-create API, and community evidence directly contradicts the 'agent manages rules programmatically' premise: users report 'the lack of easy programmatic control is an issue for us' and that Allow Rules are disabled by default for newer accounts requiring a manual support request to enable. Missing for 10: a documented API/MCP tool to create or propose new velocity/fraud rules, and independent confirmation that programmatic rule/list management works smoothly without support intervention.

   • [claimed-docs] “POST /v1/reviews/:id/approve”
   • [claimed-docs] “Value lists allow you to group values together which can then be referenced in rules.”
   • [claimed-docs] “Customer IDs for trusted customers: Use this list to automatically allow payments by these customers.”
   • [probe] “official MCP server documented at https://docs.stripe.com/mcp”
   • [community] “We are seeing an opposite side: customers using Stripe that had very low fraud rates previously are now getting more false positives causing…”
   • [community] “Allow Rules, if not implemented properly, could open a vector for fraud. That's why it's not enabled for newer businesses on Stripe—we ask b…”

   proof ↗⚑ flag

   Sift—none0/10

   Evidence shows REST/Decisions/Workflows APIs for events, scoring, and rules automation, but nothing describes an AI agent reading fraud posture or proposing/updating rules and lists with a human-approval workflow — this is a generic developer API story, not an agent-access pattern. missing for 10: any mention of AI agent integration, agentic rule-proposal workflow, or human-approval gate mechanism tied to programmatic rule/list changes.

   • [claimed-docs] “Sift easily integrates into your existing systems using modern REST APIs, Javascript snippet, and SDKs for iOS and Android.”
   • [claimed-docs] “What actions your business is taking in response to users (e.g., approve an order, block and event due to fraud, cancel order due to chargeb…”
   • [claimed-docs] “Sift offers Workflows, a powerful rules automation platform, and Review Queues.”

   proof ↗⚑ flag

Agent triage

1. ai-native userAn agent can work the review queue — pull flagged cases with their context, summarize the evidence, and recommend a decision for a human to confirm

   weight 2 · round to Stripe Radar
   Stripe Radar~partialprobed6/10

   Radar exposes a Reviews API (list/retrieve/approve reviews), risk-insights showing related-payment networks, and early-fraud-warning data that an agent could pull as 'flagged case context,' and Stripe documents an official MCP server that could expose these APIs to an agent. However there's no first-party feature for automated evidence summarization or a recommend-then-human-confirm workflow — that logic would have to be built by the integrator, and community reports note UI/override friction around review decisions. Missing for 10: a documented agent/summarization workflow for reviews, evidence the MCP server actually exposes the reviews/early-fraud-warning endpoints, and independent confirmation of an agent successfully working the queue end-to-end.

   • [claimed-docs] “POST /v1/reviews/:id/approve”
   • [claimed-docs] “You can create a targeted list of payments to review with criteria that you specify, and review them in the Dashboard.”
   • [claimed-docs] “You can also view the network of related payments, which includes any other payments made to your business using the same customer ID, IP ad…”
   • [claimed-docs] “An early fraud warning indicates that the card issuer has notified us that a charge may be fraudulent.”
   • [probe] “official MCP server documented at https://docs.stripe.com/mcp”
   • [community] “We got a couple cases of false positives ourselves, and the Stripe UI wasn't very clear that we couldn't override the 'block' (the button wa…”

   proof ↗⚑ flag

   Sift—none0/10

   Sift has Review Queues and REST/Decisions APIs for humans to review flagged cases (sift-docs-12, sift-docs-3), and community evidence confirms a human-in-the-loop review model (sift-comm-4), but there is no evidence of an AI agent programmatically pulling flagged cases, summarizing evidence, or recommending decisions for human confirmation.

   • [claimed-docs] “Sift offers Workflows, a powerful rules automation platform, and Review Queues.”
   • [claimed-docs] “What actions your business is taking in response to users (e.g., approve an order, block and event due to fraud, cancel order due to chargeb…”
   • [community] “Sift Science founder on false positives: 'Most of our customers review each user we flag, and have a human make a final go/no-go decision ab…”

   proof ↗⚑ flag

Builtin ai

1. risk analystThe product ships its own AI assistant — natural-language queries over my fraud data, drafted rules, investigation summaries — built into the console

   weight 2 · round drawn
   Stripe Radar—none0/10

   No evidence of a built-in AI assistant in the Radar console for natural-language queries, drafted rules, or investigation summaries; documentation covers rules engine, reviews, analytics, and lists but nothing about an AI/NLP assistant feature.

   ⚑ flag

   Sift—none0/10

   The evidence pack covers Sift's APIs, scoring, workflows/rules automation, and review queues, but there is no mention of any built-in AI assistant, natural-language query capability, drafted rules generation, or investigation summary generation in the console. Missing for 10: any documentation or mention of a natural-language/AI assistant feature, evidence of NL-to-query capability, evidence of AI-drafted rules or summaries.

   • [claimed-docs] “Workflows enable you to make risk-based decisions in real time during key events. Follow this guide to create your own custom Workflow.”
   • [claimed-docs] “Sift offers Workflows, a powerful rules automation platform, and Review Queues.”

   proof ↗⚑ flag

Abuse coverage

1. risk analystProtection extends beyond checkout — account takeover, fake account creation, promo and policy abuse are scored and managed in the same system

   weight 2 · round to Sift
   Stripe Radar~partialclaimed5/10

   Radar Pro docs explicitly extend beyond checkout fraud to 'multi-account, free trial, and pay-as-you-go abuse,' covering fake-account and promo/policy abuse in the same platform, but there is no evidence of dedicated account-takeover detection or scoring — the docs focus on payment/charge risk, reviews, and disputes rather than login/session anomaly detection typical of ATO protection. Missing for 10: explicit account-takeover detection/scoring capability, unified dashboard evidence showing ATO alongside promo-abuse cases, and independent confirmation these abuse types are actually managed in one system rather than just marketed together.

   • [claimed-docs] “Radar Pro: Advanced protection against emerging fraud threats and customer abuse. Detect multi-account, free trial, and pay-as-you-go abuse.”
   • [claimed-docs] “Radar Standard: Out-of-the-box fraud protection for all payment methods to detect and prevent transaction fraud, and identify fraudulent acc…”
   • [claimed-docs] “An early fraud warning indicates that the card issuer has notified us that a charge may be fraudulent.”

   proof ↗⚑ flag

   Sift✓fullclaimed8/10

   Docs explicitly cover multiple fraud surfaces beyond checkout — account takeover ('Block unauthorized access in real time'), fake account creation ('Stop fraudulent signups at the door'), and abuse-specific scoring ('fight multiple types of fraud at once... add an abuse-specific risk score'), all managed within the same Workflows/Decisions/Review Queue system. Missing for 10: dedicated documentation or case study specifically on promo/policy abuse handling, and independent hands-on verification of multi-surface scoring in one unified dashboard.

   • [claimed-docs] “Block unauthorized access in real time.”
   • [claimed-docs] “Stop fraudulent signups at the door.”
   • [claimed-docs] “Want to fight multiple types of fraud at once? Learn how to add an abuse-specific risk score so that you can easily do so.”
   • [claimed-docs] “Workflows enable you to make risk-based decisions in real time during key events. Follow this guide to create your own custom Workflow.”
   • [claimed-docs] “Sift offers Workflows, a powerful rules automation platform, and Review Queues.”
   • [claimed-docs] “Learn to set up custom Decisions, like Ban Account or Cancel Order, that are connected to real business actions in your backend.”

   proof ↗⚑ flag

Integrations

1. ops userThere are maintained integrations for my commerce stack — Shopify, Salesforce Commerce, BigCommerce, and the major PSPs — not just a raw API

   weight 2 · round drawn
   Stripe Radar—none0/10

   Evidence covers Radar's rules, reviews, risk settings, and API/CLI/MCP tooling, but nothing mentions maintained integrations or plugins for commerce platforms like Shopify, Salesforce Commerce, or BigCommerce, or PSP-specific integrations beyond Stripe's own API. Missing for 10: any documentation of Shopify/BigCommerce/Salesforce Commerce app integrations, partner PSP integrations, or an integrations marketplace listing.

   • [claimed-docs] “Radar Standard: Out-of-the-box fraud protection for all payment methods to detect and prevent transaction fraud, and identify fraudulent acc…”
   • [claimed-docs] “Risk settings let you balance authorization and fraud on your account by using risk controls.”
   • [probe] “PROBE docs-md: HTTP 200 at https://docs.stripe.com/radar.md # Radar Use Stripe Radar to protect your business against fraud. ## Get starte…”

   proof ↗⚑ flag

   Sift—none0/10

   Evidence shows only generic REST API, JS snippet, and mobile SDKs — no mention of maintained Shopify, Salesforce Commerce, BigCommerce, or PSP-specific integrations. Everything points to a raw API/SDK integration model, not prebuilt commerce-platform connectors.

   • [claimed-docs] “Sift easily integrates into your existing systems using modern REST APIs, Javascript snippet, and SDKs for iOS and Android.”
   • [claimed-docs] “What actions your users are taking, usually key user lifecycle events (e.g., creating an account, placing an order, posting content to other…”
   • [claimed-docs] “What actions your business is taking in response to users (e.g., approve an order, block and event due to fraud, cancel order due to chargeb…”

   proof ↗⚑ flag

Psp coverage

1. developerUse the product across whatever payment stack I run — multiple PSPs, gateways, and platforms — rather than being locked to one processor's rails

   weight 3 · round to Sift
   Stripe Radar—none0/10

   All evidence describes Radar as a feature built directly into Stripe's own payments processing (rules, reviews, risk settings, session tokenization, testing tied to Stripe test cards) with no mention of usable integration with other PSPs, gateways, or platforms. Fraud-detection tools in general could plausibly support multi-processor use, but nothing in the pack shows Radar operating outside Stripe's own rails — it is described purely as a Stripe-native capability.

   • [claimed-docs] “Radar Standard: Out-of-the-box fraud protection for all payment methods to detect and prevent transaction fraud, and identify fraudulent acc…”
   • [claimed-docs] “By using Radar Sessions, you can capture critical fraud information without tokenizing on Stripe.”
   • [claimed-docs] “4000000000004954 | Results in a charge with a risk level of `highest`”
   • [probe] “PROBE docs-md: HTTP 200 at https://docs.stripe.com/radar.md # Radar Use Stripe Radar to protect your business against fraud. ## Get starte…”

   proof ↗⚑ flag

   Sift~partialclaimed5/10

   Sift's architecture (generic REST Events/Decisions API, JS snippet, SDKs) is inherently processor-agnostic — it ingests events from any application regardless of which PSP or gateway processes the payment, rather than being tied to one processor's rails. However, there is no explicit documentation or case evidence naming specific PSPs/gateways/platforms it integrates with or confirming multi-processor deployments in practice. Missing for 10: explicit multi-PSP/gateway integration examples, named partner processors, or customer testimony confirming cross-stack usage.

   • [claimed-docs] “Sift easily integrates into your existing systems using modern REST APIs, Javascript snippet, and SDKs for iOS and Android.”
   • [claimed-docs] “What actions your users are taking, usually key user lifecycle events (e.g., creating an account, placing an order, posting content to other…”
   • [claimed-docs] “What actions your business is taking in response to users (e.g., approve an order, block and event due to fraud, cancel order due to chargeb…”
   • [claimed-docs] “Custom events and fields describe user actions not captured by the supported events in our Events API.”

   proof ↗⚑ flag

Explainability

1. risk analystEvery score comes with its top risk factors — why this transaction looks risky — not just an opaque number

   weight 3 · round to Stripe Radar
   Stripe Radar~partialclaimed4/10

   Radar's Reviews/Risk Insights docs show a 'network of related payments' (same customer ID, IP, card) and analytics trends, suggesting some contextual signals behind a score, but the evidence never describes an explicit list of 'top risk factors' or feature-level explanation attached to each transaction's score. Missing for 10: explicit per-transaction factor breakdown/SHAP-style explanation, quantified factor weighting, independent confirmation that risk-insights actually enumerates specific risk drivers rather than just related-payment context.

   • [claimed-docs] “You can also view the network of related payments, which includes any other payments made to your business using the same customer ID, IP ad…”
   • [claimed-docs] “Visualize trends in transaction volume and fraud rates over time.”
   • [claimed-docs] “Risk settings let you balance authorization and fraud on your account by using risk controls.”

   proof ↗⚑ flag

   Sift—none0/10

   The evidence pack documents Sift's 0–100 risk score, Workflows, and Decisions APIs, but nowhere mentions score explanations, reason codes, or top contributing risk factors accompanying each score. Missing for 10: any documentation of explainability/reason-code output, feature-importance breakdowns, or examples of a score being paired with human-readable risk drivers.

   • [claimed-docs] “Sift represents this risk with a score between 0 and 100, where risky events have higher scores.”
   • [claimed-docs] “Workflows enable you to make risk-based decisions in real time during key events. Follow this guide to create your own custom Workflow.”
   • [claimed-docs] “Sift offers Workflows, a powerful rules automation platform, and Review Queues.”

   proof ↗⚑ flag

Model evaluation

1. finance leadMeasure the model itself — precision and recall on my traffic, shadow-mode trials of new models or rules before they take over decisions

   weight 2 · round drawn
   Stripe Radar—none0/10

   Docs mention fraud-rate analytics and dispute measurement (fraud insights, risk settings) but nowhere describe precision/recall metrics on the merchant's own traffic or a shadow-mode mechanism to trial new models/rules before they affect decisions. Rules can be created and reviewed, but there's no evidence of a non-blocking 'test' or 'shadow' deployment mode for models.

   • [claimed-docs] “Visualize trends in transaction volume and fraud rates over time.”
   • [claimed-docs] “We show this calculation on the Radar page in the Dashboard.”
   • [claimed-docs] “Risk settings let you balance authorization and fraud on your account by using risk controls.”

   proof ↗⚑ flag

   Sift—none0/10

   Evidence covers scoring, workflows, decisions, and events APIs, but there is no mention of precision/recall metrics on the customer's own traffic, model performance measurement, or shadow-mode trialing of new models/rules before they go live. missing for 10: precision/recall measurement tooling, shadow-mode/challenger model testing, any model transparency reporting for finance/business stakeholders.

   ⚑ flag

1. ai-native userDo everything through the API that I can do in the UI

   weight 2 · round to Sift
   Stripe Radar!disputedcontradicted4/10

   Stripe exposes API endpoints for some Radar objects (reviews approve/reject, early fraud warnings, value lists) but the core capability of authoring/editing Radar Rules and risk settings is documented only via Dashboard-oriented docs (stripe-radar-docs-3, docs-13) with no corresponding rules-API endpoint evidenced. Community feedback explicitly calls out this gap ('The lack of easy programmatic control is an issue for us', stripe-radar-comm-2) and notes Allow Rules require manual support intervention rather than self-service API access (stripe-radar-comm-7), directly contradicting full API/UI parity. missing for 10: documented API endpoints for creating/editing Radar rules, programmatic risk-settings control, independent confirmation that all dashboard actions have API equivalents.

   • [claimed-docs] “you can also set up rules that are unique to your business using the supported attributes”
   • [claimed-docs] “Risk settings let you balance authorization and fraud on your account by using risk controls.”
   • [claimed-docs] “POST /v1/reviews/:id/approve”
   • [claimed-docs] “An early fraud warning indicates that the card issuer has notified us that a charge may be fraudulent.”
   • [claimed-docs] “Value lists allow you to group values together which can then be referenced in rules.”
   • [community] “We are seeing an opposite side: customers using Stripe that had very low fraud rates previously are now getting more false positives causing…”
   • [community] “Allow Rules, if not implemented properly, could open a vector for fraud. That's why it's not enabled for newer businesses on Stripe—we ask b…”

   proof ↗⚑ flag

   Sift~partialclaimed6/10

   Sift's docs show extensive REST APIs covering events, decisions, scoring, and even a Workflows API for automation, indicating broad programmatic access mirroring core UI functions (sift-docs-1,2,3,5,6,12). However, there's no explicit evidence that Review Queues or full workflow configuration UI actions are fully API-equivalent, and no independent confirmation of complete UI/API parity. missing for 10: explicit parity claim for Review Queues and workflow UI configuration, independent verification of full CRUD API coverage matching every console action.

   • [claimed-docs] “Sift easily integrates into your existing systems using modern REST APIs, Javascript snippet, and SDKs for iOS and Android.”
   • [claimed-docs] “What actions your users are taking, usually key user lifecycle events (e.g., creating an account, placing an order, posting content to other…”
   • [claimed-docs] “What actions your business is taking in response to users (e.g., approve an order, block and event due to fraud, cancel order due to chargeb…”
   • [claimed-docs] “Sift represents this risk with a score between 0 and 100, where risky events have higher scores.”
   • [claimed-docs] “Workflows enable you to make risk-based decisions in real time during key events. Follow this guide to create your own custom Workflow.”
   • [claimed-docs] “Sift offers Workflows, a powerful rules automation platform, and Review Queues.”

   proof ↗⚑ flag

2. ai-native userExport all of my data in open formats and leave

   weight 3 · round drawn
   Stripe Radar—none0/10

   No evidence of any data export feature or open-format data portability for Radar; docs cover rules, reviews, analytics, and APIs but nothing about exporting all account/fraud data to leave the platform. missing for 10: bulk/full data export tooling, open-format (CSV/JSON) export docs, any account-closure/data-portability guarantee.

   ⚑ flag

   Sift—none0/10

   Sift is a fraud-detection API platform; evidence covers sending events/data into Sift via REST API, scoring, workflows, but there is no mention of exporting or bulk-downloading customer data out of Sift in open formats, nor any data-portability/export feature. missing for 10: any documented data export/download capability, open-format export (CSV/JSON dumps), or account data portability tooling.

   ⚑ flag

1. ai-native userChoose where my data is stored (region/residency)

   weight 2 · round drawn
   Stripe Radar—none0/10

   No evidence in the pack addresses data residency, regional storage options, or data location controls for Stripe Radar; the documentation covers fraud rules, reviews, and risk settings but not where data is stored.

   ⚑ flag

   Sift—none0/10

   No evidence in the pack mentions data residency, regional storage options, or geographic control over where Sift stores customer data; one community comment even notes hesitation about sending data to Sift without clarity on compliance handling. Missing for 10: any documentation of region selection, data residency options, or compliance certifications addressing storage location.

   • [community] “Potential customer notes hesitation: reservations about sending personal user data to an aggregator that would then require legal review for…”

   proof ↗⚑ flag

2. ai-native userControl data retention and deletion

   weight 2 · round drawn
   Stripe Radar—none0/10

   The evidence pack contains no mention of data retention policies, deletion controls, or privacy/data lifecycle management features for Radar; documentation covers fraud rules, reviews, testing, and pricing but nothing about controlling how long data is kept or deleting it.

   ⚑ flag

   Sift—none0/10

   Sift is a fraud-detection API/platform, and the evidence pack contains no documentation of data retention controls, deletion APIs, or configurable data lifecycle policies for users. One community comment (sift-comm-3) even flags unresolved privacy/legal review concerns about sending personal data to Sift, but this is not concrete evidence of a retention/deletion mechanism either way. missing for 10: any documentation of data retention settings, a deletion/right-to-be-forgotten API, or data lifecycle/export controls.

   • [community] “Potential customer notes hesitation: reservations about sending personal user data to an aggregator that would then require legal review for…”

   proof ↗⚑ flag

Residency

1. ops userControl where fraud data lives and how long it's kept — regional residency options and retention controls that survive a privacy review

   weight 2 · round drawn
   Stripe Radar—none0/10

   No evidence pack item addresses data residency options, regional storage location controls, or configurable retention periods for fraud/Radar data; the docs cover rules, reviews, lists, and testing but nothing about compliance/residency controls.

   ⚑ flag

   Sift—none0/10

   No evidence pack items discuss regional data residency, data hosting locations, retention periods, or compliance/privacy review controls; one comment even notes a prospective customer's hesitation about privacy/legal review with no resolution shown. Missing for 10: any documentation of regional hosting/data residency options, retention/configurable deletion policies, or compliance certifications addressing privacy review.

   • [community] “Potential customer notes hesitation: reservations about sending personal user data to an aggregator that would then require legal review for…”

   proof ↗⚑ flag

Sca

1. developerEuropean traffic is routed intelligently through SCA — 3DS triggered when required or risky, exemptions requested when safe — to protect both compliance and conversion

   weight 2 · round to Stripe Radar
   Stripe Radar~partialclaimed4/10

   Radar rules docs show it can trigger 3DS for specific conditions (e.g., new customers) via custom rules, but there is no evidence of intelligent SCA-wide routing that automatically requests exemptions (TRA, low-value, etc.) to preserve conversion — the exemption side of the story is unaddressed and this Radar-specific capability differs from Stripe's core SCA/Payment Intents engine. missing for 10: evidence of automated exemption requests (TRA/low-value/trusted-beneficiary), evidence of end-to-end SCA compliance logic beyond manual rule authoring, and independent confirmation of conversion impact.

   • [claimed-docs] “Request 3D Secure (3DS) for all payments that support it and are made by a new customer”
   • [claimed-docs] “you can also set up rules that are unique to your business using the supported attributes”

   proof ↗⚑ flag

   Sift—none0/10

   The axis applies to this product kind (peer products hold positive or none verdicts on this story), so lack of evidence for an applicable capability is "none", never "na". (na/none harmonized at arena bring-up — see pipeline/scripts/na-harmonize.ts.)

   ⚑ flag

Case review

1. risk analystFlagged transactions land in a review queue that shows the full context — customer history, signals, similar cases — so I can decide quickly and consistently

   weight 3 · round to Stripe Radar
   Stripe Radar~partialcommunity7/10

   Stripe's docs show a genuine review queue workflow: targeted transaction reviews (docs-11), an approve/reject API (docs-8), and a 'risk insights' view showing related payments across customer ID, IP, or card number for spotting similar cases (docs-12). However, evidence doesn't explicitly confirm a unified single screen combining full customer history + fraud signals + similar-case surfacing in one glance, and community feedback notes UI friction when analysts try to act on flagged transactions (comm-3, unclear override controls). missing for 10: explicit documentation of a consolidated 'customer history' panel within the review UI, and independent/hands-on confirmation that the queue enables fast, consistent decisions without friction.

   • [claimed-docs] “You can create a targeted list of payments to review with criteria that you specify, and review them in the Dashboard.”
   • [claimed-docs] “POST /v1/reviews/:id/approve”
   • [claimed-docs] “You can also view the network of related payments, which includes any other payments made to your business using the same customer ID, IP ad…”
   • [community] “We got a couple cases of false positives ourselves, and the Stripe UI wasn't very clear that we couldn't override the 'block' (the button wa…”

   proof ↗⚑ flag

   Sift~partialcommunity5/10

   Sift documents Review Queues as a product feature alongside Workflows, and community evidence confirms human analysts do review flagged users to make final decisions, supporting the general workflow. However, there is no evidence detailing the review queue UI showing customer history, similar cases, or signal context in one view. missing for 10: documentation/screenshots of review queue UI showing customer history and similar cases, evidence of consolidated signal context, independent hands-on confirmation of queue usability.

   • [claimed-docs] “Sift offers Workflows, a powerful rules automation platform, and Review Queues.”
   • [community] “Sift Science founder on false positives: 'Most of our customers review each user we flag, and have a human make a final go/no-go decision ab…”

   proof ↗⚑ flag

Feedback loop

1. risk analystMy review decisions and confirmed fraud outcomes feed back into the model and rules, so the system learns from every case we work

   weight 2 · round to Sift
   Stripe Radar~partialclaimed4/10

   Radar lets analysts approve/decline reviews (docs-8) and maintain allow/block value lists that then drive future rule evaluation (docs-10, docs-14), which is a manual feedback mechanism, and early fraud warnings feed dispute-rate risk signals (docs-9, docs-7). However there is no documented evidence that confirmed fraud outcomes or review decisions automatically retrain Radar's underlying ML model — the docs describe rules/lists as merchant-configured, not an automated learning loop tied to case outcomes. Missing for 10: explicit documentation of the ML model being retrained from analyst decisions/fraud confirmations, and independent/hands-on confirmation that outcomes measurably change future scoring.

   • [claimed-docs] “POST /v1/reviews/:id/approve”
   • [claimed-docs] “Value lists allow you to group values together which can then be referenced in rules.”
   • [claimed-docs] “Customer IDs for trusted customers: Use this list to automatically allow payments by these customers.”
   • [claimed-docs] “An early fraud warning indicates that the card issuer has notified us that a charge may be fraudulent.”
   • [claimed-docs] “We show this calculation on the Radar page in the Dashboard.”

   proof ↗⚑ flag

   Sift~partialcommunity6/10

   Sift's Decisions API explicitly captures business actions like 'block due to fraud' or 'cancel due to chargeback' and feeds them back to Sift, and docs state Sift 'gets smarter the more data it has,' supporting a feedback loop from confirmed outcomes into the model. Review Queues and Workflows are documented as connected to decisions, and a founder note confirms customers commonly have analysts make final go/no-go calls that presumably feed Decisions API. Missing for 10: explicit documentation that individual analyst review-queue verdicts (not just automated decisions) are looped back per-case to retrain the model, and any detail on how confirmed fraud outcomes update rules/workflows specifically.

   • [claimed-docs] “What actions your business is taking in response to users (e.g., approve an order, block and event due to fraud, cancel order due to chargeb…”
   • [claimed-docs] “Because Sift gets smarter the more data it has about your business, you can jump start your integration by backfilling a few months worth of…”
   • [claimed-docs] “Sift offers Workflows, a powerful rules automation platform, and Review Queues.”
   • [community] “Sift Science founder on false positives: 'Most of our customers review each user we flag, and have a human make a final go/no-go decision ab…”

   proof ↗⚑ flag

Team workflows

1. ops userReview work is a team workflow — assignment, escalation, SLAs, and a decision audit trail that shows who approved what and why

   weight 2 · round to Sift
   Stripe Radar~partialclaimed3/10

   Radar's reviews API supports approve/decline actions on flagged payments and lets teams build custom review queues, which implies some decision logging, but there is no documented support for assignment to specific reviewers, escalation paths, or SLA tracking, and no explicit 'who approved what and why' audit trail beyond the approve/reject call itself. missing for 10: reviewer assignment, escalation workflow, SLA tracking, structured decision-rationale audit trail.

   • [claimed-docs] “POST /v1/reviews/:id/approve”
   • [claimed-docs] “You can create a targeted list of payments to review with criteria that you specify, and review them in the Dashboard.”
   • [claimed-docs] “You can also view the network of related payments, which includes any other payments made to your business using the same customer ID, IP ad…”

   proof ↗⚑ flag

   Sift~partialcommunity5/10

   Sift documents Review Queues and Workflows for human review, plus a Decisions API that logs business actions like approve/ban/cancel tied to real events, which supports a basic audit trail; a founder comment also confirms a human-in-the-loop review pattern ('have a human make a final go/no-go decision'). However, there's no evidence of explicit team-workflow features like assignment routing, escalation paths, or SLA tracking. missing for 10: assignment/routing mechanics, escalation workflows, SLA timers/tracking, explicit 'who approved what and why' UI/reporting beyond raw decision logs.

   • [claimed-docs] “Sift offers Workflows, a powerful rules automation platform, and Review Queues.”
   • [claimed-docs] “What actions your business is taking in response to users (e.g., approve an order, block and event due to fraud, cancel order due to chargeb…”
   • [claimed-docs] “Learn to set up custom Decisions, like Ban Account or Cancel Order, that are connected to real business actions in your backend.”
   • [community] “Sift Science founder on false positives: 'Most of our customers review each user we flag, and have a human make a final go/no-go decision ab…”

   proof ↗⚑ flag

Custom signals

1. developerFeed the model my own signals — device fingerprints, behavioral data, custom metadata — so scoring reflects my business, not just network defaults

   weight 2 · round to Sift
   Stripe Radar~partialcommunity5/10

   Radar lets developers write custom rules against 'supported attributes' and value lists (docs-3, docs-10, docs-14), and Radar Session captures device/browser signals for fraud evaluation without full tokenization (docs-15) — this shows some capacity to incorporate custom signals. However, there's no evidence that arbitrary custom metadata or behavioral data actually feeds into or retrains Radar's core ML risk score itself (rather than just triggering rule-based overrides), and community commentary notes a 'lack of easy programmatic control' over scoring (comm-2). missing for 10: explicit documentation that custom metadata/behavioral inputs alter the underlying risk score model, first-party guidance on feeding proprietary signals into scoring, and independent confirmation this works as intended.

   • [claimed-docs] “you can also set up rules that are unique to your business using the supported attributes”
   • [claimed-docs] “Value lists allow you to group values together which can then be referenced in rules.”
   • [claimed-docs] “Customer IDs for trusted customers: Use this list to automatically allow payments by these customers.”
   • [claimed-docs] “By using Radar Sessions, you can capture critical fraud information without tokenizing on Stripe.”
   • [community] “We are seeing an opposite side: customers using Stripe that had very low fraud rates previously are now getting more false positives causing…”

   proof ↗⚑ flag

   Sift✓fullclaimed8/10

   Sift's docs clearly support ingesting custom events, custom fields, session/device signals, and business action data via Events/Decisions APIs, and scoring incorporates this custom data plus historical backfill to tailor risk to the business rather than generic network defaults. Missing for 10: independent hands-on verification of custom-field scoring impact and explicit device-fingerprint API documentation beyond general SDK mentions.

   • [claimed-docs] “What actions your users are taking, usually key user lifecycle events (e.g., creating an account, placing an order, posting content to other…”
   • [claimed-docs] “What actions your business is taking in response to users (e.g., approve an order, block and event due to fraud, cancel order due to chargeb…”
   • [claimed-docs] “Because Sift gets smarter the more data it has about your business, you can jump start your integration by backfilling a few months worth of…”
   • [claimed-docs] “Custom events and fields describe user actions not captured by the supported events in our Events API.”
   • [claimed-docs] “When users are anonymous for parts or all of their experience on your site, we provide a special `session_id` field in the JS Snippet and Ev…”
   • [claimed-docs] “Sift represents this risk with a score between 0 and 100, where risky events have higher scores.”

   proof ↗⚑ flag

Network effects

1. founderScoring benefits from a cross-merchant network — a card or identity seen across thousands of other businesses informs the risk decision on mine

   weight 2 · round to Stripe Radar
   Stripe Radar~partialclaimed5/10

   Docs show Radar surfaces a 'network of related payments' across customer ID, IP, or card number (docs-12) and ingests card-issuer signals via Early Fraud Warnings (docs-9), implying some cross-account risk signal, and docs-6 describes 'out-of-the-box' pre-trained fraud detection. However, none of the evidence explicitly states the model is trained on or scores against Stripe's full cross-merchant network of thousands of businesses. Missing for 10: explicit documentation of the network-wide ML training/scoring claim, and independent confirmation that cross-merchant signals (not just same-business history) drive individual risk scores.

   • [claimed-docs] “You can also view the network of related payments, which includes any other payments made to your business using the same customer ID, IP ad…”
   • [claimed-docs] “An early fraud warning indicates that the card issuer has notified us that a charge may be fraudulent.”
   • [claimed-docs] “Radar Standard: Out-of-the-box fraud protection for all payment methods to detect and prevent transaction fraud, and identify fraudulent acc…”

   proof ↗⚑ flag

   Sift—none0/10

   The evidence pack describes Sift's own ML scoring, event/decision APIs, and workflows, but nowhere claims that risk signals (e.g., a card or identity) are shared or aggregated across Sift's merchant customer base to inform another business's score. sift-docs-4 only references a business's own historical data improving its own model, not cross-merchant network effects.

   ⚑ flag

Score actions

1. ops userMap score ranges to actions — allow, review, block, step-up 3DS — and tune thresholds to my own risk appetite instead of a fixed cutoff

   weight 2 · round to Sift
   Stripe Radar~partialcommunity6/10

   Stripe Radar's docs clearly support mapping risk levels to actions (block, review, request 3DS, allow via value lists) and tuning via risk-settings/rules with custom attributes, directly matching the story. However, community evidence shows real friction: allow rules are gated behind manual support enablement for newer accounts (comm-7), an ops user hit a 'block' override button that silently did nothing requiring a support ticket (comm-3), and another reports 'lack of easy programmatic control' over false positives (comm-2), indicating the threshold-tuning experience isn't as self-service as docs imply. Missing for 10: independent verification that threshold-to-action mapping is fully self-serve without support intervention, and resolution of the UI override bug.

   • [claimed-docs] “Request 3D Secure (3DS) for all payments that support it and are made by a new customer”
   • [claimed-docs] “Risk settings let you balance authorization and fraud on your account by using risk controls.”
   • [claimed-docs] “Customer IDs for trusted customers: Use this list to automatically allow payments by these customers.”
   • [claimed-docs] “you can also set up rules that are unique to your business using the supported attributes”
   • [community] “We are seeing an opposite side: customers using Stripe that had very low fraud rates previously are now getting more false positives causing…”
   • [community] “We got a couple cases of false positives ourselves, and the Stripe UI wasn't very clear that we couldn't override the 'block' (the button wa…”
   • [community] “Allow Rules, if not implemented properly, could open a vector for fraud. That's why it's not enabled for newer businesses on Stripe—we ask b…”

   proof ↗⚑ flag

   Sift~partialcommunity7/10

   Sift's docs describe Workflows as a rules automation platform for real-time risk-based decisions, with custom Decisions (e.g., Ban Account, Cancel Order) tied to score ranges (0-100), and Review Queues for human review — directly supporting configurable score-to-action mapping and threshold tuning. However, no evidence explicitly mentions 3DS step-up as an action type, and an older founder comment (sift-comm-8) claims 'no rules' which is in tension with the current Workflows rules-engine framing, though this appears to reflect product evolution rather than a live contradiction. missing for 10: explicit 3DS/step-up action documentation, independent hands-on confirmation of threshold customization in practice.

   • [claimed-docs] “Sift represents this risk with a score between 0 and 100, where risky events have higher scores.”
   • [claimed-docs] “Workflows enable you to make risk-based decisions in real time during key events. Follow this guide to create your own custom Workflow.”
   • [claimed-docs] “Learn to set up custom Decisions, like Ban Account or Cancel Order, that are connected to real business actions in your backend.”
   • [claimed-docs] “Sift offers Workflows, a powerful rules automation platform, and Review Queues.”
   • [community] “Sift Science founder on false positives: 'Most of our customers review each user we flag, and have a human make a final go/no-go decision ab…”
   • [community] “CEO explains no-rules ML approach: 'we do not have any rules in our product... rules are rather easy for fraudsters to circumvent, and they …”

   proof ↗⚑ flag

Scoring api

1. developerGet a machine-learning risk score for a transaction in real time — synchronously, before authorization completes — through a documented API

   weight 3 · round drawn
   Stripe Radar~partialclaimed6/10

   Docs confirm Radar attaches a risk level/score to each charge (e.g. test card 4000000000004954 'Results in a charge with a risk level of highest') and that rules/reviews act on this score, implying the score is computed as part of normal payment processing and exposed on the charge object via the API. However, the pack never explicitly documents the exact API field (e.g. charge.outcome.risk_score) or explicitly states the score is available synchronously before authorization completes, and no independent/hands-on confirmation of real-time synchronous scoring is present. missing for 10: explicit API reference/schema for risk_score/risk_level field, explicit documentation stating the score is returned before/at authorization time, independent corroboration of real-time synchronous behavior.

   • [claimed-docs] “4000000000004954 | Results in a charge with a risk level of `highest`”
   • [claimed-docs] “Radar Standard: Out-of-the-box fraud protection for all payment methods to detect and prevent transaction fraud, and identify fraudulent acc…”
   • [claimed-docs] “Request 3D Secure (3DS) for all payments that support it and are made by a new customer”
   • [claimed-docs] “POST /v1/reviews/:id/approve”

   proof ↗⚑ flag

   Sift~partialcommunity6/10

   Sift's documented Score API returns a 0-100 ML risk score via REST, and Workflows enable real-time risk-based decisions during key events, supporting synchronous scoring before a decision. However, the evidence doesn't explicitly confirm the Score API is called synchronously pre-authorization (vs. async/webhook-based scoring), and community feedback notes many customers use human review queues rather than pure real-time automated decisions. Missing for 10: explicit documentation of synchronous call/response latency guarantees for pre-authorization use, and independent confirmation of real-time performance under transaction-blocking scenarios.

   • [claimed-docs] “Sift represents this risk with a score between 0 and 100, where risky events have higher scores.”
   • [claimed-docs] “Workflows enable you to make risk-based decisions in real time during key events. Follow this guide to create your own custom Workflow.”
   • [claimed-docs] “Block unauthorized access in real time.”
   • [community] “Sift Science founder on false positives: 'Most of our customers review each user we flag, and have a human make a final go/no-go decision ab…”

   proof ↗⚑ flag

Backtesting

1. risk analystBacktest a rule against my historical traffic before deploying it, seeing exactly what it would have blocked, flagged, and cost

   weight 2 · round drawn
   Stripe Radar—none0/10

   The evidence pack covers rule creation, value lists, risk settings, reviews, and analytics dashboards, but no documentation or community evidence describes a backtesting/simulation feature that shows what a new rule would have blocked/flagged/cost against historical traffic before deployment.

   • [claimed-docs] “you can also set up rules that are unique to your business using the supported attributes”
   • [claimed-docs] “Visualize trends in transaction volume and fraud rates over time.”
   • [claimed-docs] “Risk settings let you balance authorization and fraud on your account by using risk controls.”

   proof ↗⚑ flag

   Sift—none0/10

   The evidence describes Workflows as a rules automation platform and mentions real-time decisioning, but nothing in the pack describes a backtest/simulation capability against historical traffic showing what a rule would have blocked/flagged/cost.

   • [claimed-docs] “Sift offers Workflows, a powerful rules automation platform, and Review Queues.”
   • [claimed-docs] “Workflows enable you to make risk-based decisions in real time during key events. Follow this guide to create your own custom Workflow.”

   proof ↗⚑ flag

Lists

1. ops userMaintain allow and block lists — emails, cards, devices, IPs — and velocity limits, managed through the dashboard and programmatically

   weight 2 · round to Stripe Radar
   Stripe Radar!disputedcontradicted5/10

   Stripe documents value lists for allow/block lists (customers, cards, emails, IPs) manageable via API and dashboard (stripe-radar-docs-10, docs-14) plus rules for velocity/custom attributes (docs-3, docs-1), giving vendor-side coverage of the story. However hands-on community reports contradict smooth programmatic control: one user cites 'lack of easy programmatic control' causing false-positive grief, and another found the dashboard's block-override button non-functional, requiring a support email to whitelist transactions; Stripe itself confirms Allow Rules are gated behind manual support enablement for newer accounts, not self-serve. Missing for 10: clear documentation of dedicated device/IP allow-and-block management (only cards/customers/emails are explicit), and no evidence the programmatic control gap or override bug has been resolved.

   • [claimed-docs] “Value lists allow you to group values together which can then be referenced in rules.”
   • [claimed-docs] “Customer IDs for trusted customers: Use this list to automatically allow payments by these customers.”
   • [claimed-docs] “you can also set up rules that are unique to your business using the supported attributes”
   • [claimed-docs] “Request 3D Secure (3DS) for all payments that support it and are made by a new customer”
   • [community] “We are seeing an opposite side: customers using Stripe that had very low fraud rates previously are now getting more false positives causing…”
   • [community] “We got a couple cases of false positives ourselves, and the Stripe UI wasn't very clear that we couldn't override the 'block' (the button wa…”
   • [community] “Allow Rules, if not implemented properly, could open a vector for fraud. That's why it's not enabled for newer businesses on Stripe—we ask b…”

   proof ↗⚑ flag

   Sift!disputedcontradicted3/10

   Docs describe a general 'Workflows' rules automation platform and Review Queues, but there's no explicit mention of dedicated allow/block lists for emails, cards, devices, or IPs, nor velocity limit configuration in the dashboard or API. Compounding this, a Sift co-founder is quoted stating 'we do not have any rules in our product' as a core design philosophy, directly conflicting with the rules-engine framing implied by the Workflows docs. missing for 10: explicit allow/block list management (emails, cards, devices, IPs), velocity limit configuration, dashboard UI evidence, and resolution of the rules-vs-no-rules contradiction.

   • [claimed-docs] “Sift offers Workflows, a powerful rules automation platform, and Review Queues.”
   • [claimed-docs] “Workflows enable you to make risk-based decisions in real time during key events. Follow this guide to create your own custom Workflow.”
   • [community] “CEO explains no-rules ML approach: 'we do not have any rules in our product... rules are rather easy for fraudsters to circumvent, and they …”
   • [community] “Critical take: 'The biggest problem with ML is that it takes time for it to react to new fraud patterns/schemes. While rules engines have li…”

   proof ↗⚑ flag

Rule authoring

1. risk analystAuthor custom rules that combine model scores, velocity counters, list matches, and transaction attributes into allow, block, or review decisions

   weight 3 · round to Stripe Radar
   Stripe Radar~partialcommunity7/10

   Stripe docs show a rich rules engine — custom rules using risk score, transaction attributes, value lists (for list matches), and allow/block/review outcomes (docs-1,3,10,13,14) — which covers most of the story. However, community evidence shows secondary caveats: Allow Rules are disabled by default for newer businesses and require manual support enablement (comm-7), and users report inability to override block decisions via UI plus false positives (comm-2,3), indicating real friction in the rule-authoring/decisioning workflow. Missing for 10: explicit documentation/example of velocity-counter attributes in rule syntax, and independent confirmation that all rule types (allow/block/review) work reliably without support intervention.

   • [claimed-docs] “Request 3D Secure (3DS) for all payments that support it and are made by a new customer”
   • [claimed-docs] “you can also set up rules that are unique to your business using the supported attributes”
   • [claimed-docs] “Value lists allow you to group values together which can then be referenced in rules.”
   • [claimed-docs] “Risk settings let you balance authorization and fraud on your account by using risk controls.”
   • [claimed-docs] “Customer IDs for trusted customers: Use this list to automatically allow payments by these customers.”
   • [community] “Allow Rules, if not implemented properly, could open a vector for fraud. That's why it's not enabled for newer businesses on Stripe—we ask b…”
   • [community] “We got a couple cases of false positives ourselves, and the Stripe UI wasn't very clear that we couldn't override the 'block' (the button wa…”
   • [community] “We are seeing an opposite side: customers using Stripe that had very low fraud rates previously are now getting more false positives causing…”

   proof ↗⚑ flag

   Sift!disputedcontradicted5/10

   Docs describe Workflows as a 'powerful rules automation platform' for real-time decisioning (allow/block/review) using scores and custom logic (sift-docs-6, sift-docs-7, sift-docs-12), suggesting the requested rule authoring capability. However, a Sift co-founder is on record stating 'we do not have any rules in our product' and that rules are easy to circumvent, directly contradicting the existence of a rules engine (sift-comm-8), and no evidence details velocity counters or list-match integration into rule logic. missing for 10: explicit documentation of velocity counters, list matching, and multi-attribute rule composition within Workflows; independent verification reconciling the rules-vs-no-rules discrepancy.

   • [claimed-docs] “Workflows enable you to make risk-based decisions in real time during key events. Follow this guide to create your own custom Workflow.”
   • [claimed-docs] “Learn to set up custom Decisions, like Ban Account or Cancel Order, that are connected to real business actions in your backend.”
   • [claimed-docs] “Sift offers Workflows, a powerful rules automation platform, and Review Queues.”
   • [community] “CEO explains no-rules ML approach: 'we do not have any rules in our product... rules are rather easy for fraudsters to circumvent, and they …”

   proof ↗⚑ flag