Puzzle vs Bench

Agent access

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

   weight 2 · round to Puzzle
   Puzzle✓fullprobed8/10

   A live probe confirms Puzzle's help center serves a valid llms.txt file at help.puzzle.io/llms.txt (HTTP 200) with structured markdown links to docs, directly satisfying the ability to point an agent at agent-oriented docs. Missing for 10: independent/community confirmation of agents successfully consuming this file, and broader agent-doc formats (e.g., per-page .md mirrors, which probes show 404).

   • [probe] “PROBE llms.txt: HTTP 200 at https://help.puzzle.io/llms.txt # Puzzle Help Center > Help center for Puzzle ## Puzzle Overview - [Quick Sta…”
   • [probe] “PROBE docs-md: HTTP 404 at https://help.puzzle.io/en/.md”

   proof ↗⚑ flag

   Bench—none0/10

   Bench is a bookkeeping service, not a dev tool/platform, but the axis still applies since any web product could publish agent-oriented docs; probes explicitly show llms.txt returns 404, docs-as-markdown 404, and no OpenAPI spec found. There is no evidence of any agent-oriented documentation surface.

   • [probe] “PROBE llms.txt: HTTP 404 at https://www.bench.co/llms.txt”
   • [probe] “PROBE docs-md: HTTP 404 at https://www.bench.co/how-it-works.md”
   • [probe] “PROBE openapi: all candidate paths 404 (https://www.bench.co/openapi.json, https://www.bench.co/swagger.json, https://www.bench.co/api/opena…”

   proof ↗⚑ flag

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

   weight 2 · round drawn
   Puzzle—none0/10

   Puzzle's evidence only shows marketing language about an 'API' for embedding accounting into platforms and an MCP integration for Claude chat — neither documents headless CLI operation, CI pipeline usage, or automation-oriented tooling. No SDK, auth flow, or CI examples are present, so there's no evidence of headless/CI automation support.

   • [claimed-docs] “With Puzzle's Embedded Accounting—powered by our API, your platform gets automated AI bookkeeping, real-time financial insights, and a clean…”
   • [claimed-docs] “With Puzzle’s Embedded Accounting—powered by our API, your platform gets automated AI bookkeeping, real-time financial insights, and a clean…”
   • [claimed-docs] “your platform gets automated AI bookkeeping, real-time financial insights, and a clean data foundation built to scale”
   • [probe] “PROBE openapi: all candidate paths 404 (https://help.puzzle.io/openapi.json, https://help.puzzle.io/swagger.json, https://help.puzzle.io/api…”

   proof ↗⚑ flag

   Bench—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

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

   weight 3 · round to Puzzle
   Puzzle✓fullclaimed7/10

   Puzzle documents an official MCP server integration that connects Puzzle accounts to Claude for natural-language access to financial data, with a dedicated setup guide (puzzle-docs-2, puzzle-docs-22, puzzle-docs-33). This is a first-party, product-level MCP server offering, fitting the axis for a non-agent SaaS product. Missing for 10: independent/hands-on confirmation that the MCP server works as documented, and technical details on supported tool calls/scopes.

   • [claimed-docs] “How to setup your Puzzle MCP”
   • [claimed-docs] “Puzzle's MCP (Model Context Protocol) integration lets you connect your Puzzle account directly to Claude, giving you instant access to your…”
   • [claimed-docs] “lets you connect your Puzzle account directly to Claude, giving you instant access to your financial data through natural conversation — no …”

   proof ↗⚑ flag

   Bench—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 userDrive the product through a documented public API

   weight 3 · round to Puzzle
   Puzzle~partialprobed4/10

   Puzzle markets an 'Embedded Accounting' API and has a help article on 'API Access in Puzzle' describing platform-to-platform integration intent, suggesting a documented public API exists. However, probes for actual API reference material (OpenAPI/swagger specs, docs-md) returned 404s, and no evidence pack item shows endpoint-level documentation, auth flow, or SDKs. Missing for 10: a public API reference/spec, authentication and endpoint documentation, and independent confirmation that AI-native users can programmatically drive the product via this API.

   • [claimed-docs] “With Puzzle's Embedded Accounting—powered by our API, your platform gets automated AI bookkeeping, real-time financial insights, and a clean…”
   • [claimed-docs] “With Puzzle’s Embedded Accounting—powered by our API, your platform gets automated AI bookkeeping, real-time financial insights, and a clean…”
   • [claimed-docs] “Our API was originally architected for deep platform-to-platform integrations.”
   • [probe] “PROBE openapi: all candidate paths 404 (https://help.puzzle.io/openapi.json, https://help.puzzle.io/swagger.json, https://help.puzzle.io/api…”

   proof ↗⚑ flag

   Bench—none0/10

   Bench is a bookkeeping/accounting service with no evidence of a documented public API; direct probes for llms.txt, docs.md, and OpenAPI/swagger endpoints all returned 404, and no docs mention an API for AI-native integration.

   • [probe] “PROBE llms.txt: HTTP 404 at https://www.bench.co/llms.txt”
   • [probe] “PROBE docs-md: HTTP 404 at https://www.bench.co/how-it-works.md”
   • [probe] “PROBE openapi: all candidate paths 404 (https://www.bench.co/openapi.json, https://www.bench.co/swagger.json, https://www.bench.co/api/opena…”

   proof ↗⚑ flag

5. ai-native userIssue scoped/least-privilege API credentials for an agent

   weight 2 · round drawn
   Puzzle—none0/10

   Evidence shows Puzzle has an API and MCP integration for connecting agents (e.g., Claude) to account data, but there is no mention of issuing scoped, least-privilege API credentials, permission scopes, or granular token controls for agents. missing for 10: scoped API key/token generation, permission/role-based credential scoping, documentation of least-privilege access controls for agent integrations.

   • [claimed-docs] “Puzzle's MCP (Model Context Protocol) integration lets you connect your Puzzle account directly to Claude, giving you instant access to your…”
   • [claimed-docs] “Our API was originally architected for deep platform-to-platform integrations.”
   • [claimed-docs] “lets you connect your Puzzle account directly to Claude, giving you instant access to your financial data through natural conversation — no …”

   proof ↗⚑ flag

   Bench—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

6. ai-native userBuild against official SDKs

   weight 2 · round to Puzzle
   Puzzle~partialprobed5/10

   Puzzle offers a documented API for embedded accounting and an MCP integration for Claude, showing some official developer surfaces, but there's no evidence of a formal SDK (client libraries, language bindings) or OpenAPI spec — probes for openapi.json all returned 404. missing for 10: official SDK/client libraries in specific languages, published OpenAPI/API reference spec, independent developer corroboration of building against the API.

   • [claimed-docs] “With Puzzle's Embedded Accounting—powered by our API, your platform gets automated AI bookkeeping, real-time financial insights, and a clean…”
   • [claimed-docs] “With Puzzle’s Embedded Accounting—powered by our API, your platform gets automated AI bookkeeping, real-time financial insights, and a clean…”
   • [claimed-docs] “Our API was originally architected for deep platform-to-platform integrations.”
   • [claimed-docs] “Puzzle's MCP (Model Context Protocol) integration lets you connect your Puzzle account directly to Claude, giving you instant access to your…”
   • [claimed-docs] “lets you connect your Puzzle account directly to Claude, giving you instant access to your financial data through natural conversation — no …”
   • [probe] “PROBE openapi: all candidate paths 404 (https://help.puzzle.io/openapi.json, https://help.puzzle.io/swagger.json, https://help.puzzle.io/api…”

   proof ↗⚑ flag

   Bench—none0/10

   Bench is a bookkeeping/tax service with no evidence of any public SDK, API, or developer library; probes for llms.txt, docs-md, and OpenAPI specs all returned 404. Missing for 10: any official SDK, API documentation, or developer integration surface.

   • [probe] “PROBE llms.txt: HTTP 404 at https://www.bench.co/llms.txt”
   • [probe] “PROBE docs-md: HTTP 404 at https://www.bench.co/how-it-works.md”
   • [probe] “PROBE openapi: all candidate paths 404 (https://www.bench.co/openapi.json, https://www.bench.co/swagger.json, https://www.bench.co/api/opena…”

   proof ↗⚑ flag

7. ai-native userSubscribe to events via webhooks

   weight 2 · round drawn
   Puzzle—none0/10

   No evidence of webhook subscription capability anywhere in the pack; Puzzle offers an API and MCP integration for querying data, but nothing about event-driven webhooks.

   ⚑ flag

   Bench—none0/10

   No evidence of any webhook subscription mechanism, API, or event system; probes for llms.txt, docs-md, and OpenAPI specs all return 404, and Bench is a human-plus-software bookkeeping service with no developer/webhook offering documented.

   • [probe] “PROBE llms.txt: HTTP 404 at https://www.bench.co/llms.txt”
   • [probe] “PROBE docs-md: HTTP 404 at https://www.bench.co/how-it-works.md”
   • [probe] “PROBE openapi: all candidate paths 404 (https://www.bench.co/openapi.json, https://www.bench.co/swagger.json, https://www.bench.co/api/opena…”

   proof ↗⚑ flag

Agentic features

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

   weight 2 · round to Puzzle
   Puzzle✓fullclaimed8/10

   Puzzle offers a conversational 'ask anything' insights feature (cash trending, profitability, fundraising readiness) plus AI Close agents that surface discrepancies, AI-driven categorization suggestions, and MCP integration for natural-language querying of financial data via Claude — all directly delivering AI-generated insights/suggestions from the user's own data. Missing for 10: independent hands-on verification of insight accuracy/quality beyond marketing copy and only one general community mention unrelated to this specific feature.

   • [claimed-docs] “Ask anything about your business, instantly, in plain English. "How's my cash trending?" "Am I ready to raise?" "Which months am I actually …”
   • [claimed-docs] “AI Close helps you automate and accelerate your month-end process using AI agents to reconcile accounts, review transactions, and surface di…”
   • [claimed-docs] “A user can click on a button and “ask AI” for help in categorizing a transaction —without context switching between tools.”
   • [claimed-docs] “Puzzle's MCP (Model Context Protocol) integration lets you connect your Puzzle account directly to Claude, giving you instant access to your…”
   • [claimed-docs] “AI Close helps you automate and accelerate your month-end process using AI agents to reconcile accounts, review transactions, and surface di…”
   • [claimed-docs] “Hand off the busywork. Finding software subscriptions, matching invoices to payments, pulling together what your accountant needs — describe…”

   proof ↗⚑ flag

   Bench—none0/10

   Bench's evidence describes automation (data entry, categorization) and human bookkeeper support, but nothing describes AI-generated insights, recommendations, or suggestions surfaced to the user from their data — 'smart automation' refers to categorization mechanics, not generative analysis or advisory insights. Missing for 10: any mention of AI-generated commentary, anomaly detection, forecasting, or proactive suggestions derived from financial data.

   • [claimed-docs] “Connect all your financial accounts to automate data entry, speed up your books, reduce errors and save time”
   • [claimed-docs] “Accurate transaction categorization, powered by smart automation with instant guidance”
   • [claimed-docs] “"Connect all your financial accounts to automate data entry, speed up your books, reduce errors and save time", with "Accurate transaction c…”
   • [claimed-docs] “Real-time reporting. Access or download your updated income statement or balance sheet at all times”

   proof ↗⚑ flag

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

   weight 2 · round to Puzzle
   Puzzle~partialclaimed6/10

   Puzzle documents concrete background automations — 'AI Close' agents that autonomously reconcile accounts, review transactions and surface discrepancies during month-end, plus custom categorization rules and batch auto-categorization (up to 98%) that run without manual triggering. However these are fixed, accounting-specific automations rather than a general-purpose 'set up your own automation' framework, and there's no independent/hands-on evidence confirming reliability of unattended runs. Missing for 10: user-configurable triggers/schedules for arbitrary automations, evidence of true autonomy without review gates, and independent verification of background execution.

   • [claimed-docs] “AI Close helps you automate and accelerate your month-end process using AI agents to reconcile accounts, review transactions, and surface di…”
   • [claimed-docs] “AI Close helps you automate and accelerate your month-end process using AI agents to reconcile accounts, review transactions, and surface di…”
   • [claimed-docs] “Creating custom rules”
   • [claimed-docs] “Batch Categorize Transactions”
   • [claimed-docs] “Up to 98% auto-categorization”

   proof ↗⚑ flag

   Bench—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

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

   weight 3 · round to Puzzle
   Puzzle✓fullclaimed8/10

   Puzzle ships built-in AI features (AI Close, Accounting AI) described as using 'AI agents to reconcile accounts, review transactions, and surface discrepancies,' plus an 'Ask AI' button for categorization help and natural-language financial queries — all within the product itself, not an external agent. Missing for 10: independent/hands-on verification of these AI features actually completing delegated tasks, and clearer detail on autonomy vs human-review boundaries.

   • [claimed-docs] “AI Close helps you automate and accelerate your month-end process using AI agents to reconcile accounts, review transactions, and surface di…”
   • [claimed-docs] “AI Close helps you automate and accelerate your month-end process using AI agents to reconcile accounts, review transactions, and surface di…”
   • [claimed-docs] “A user can click on a button and “ask AI” for help in categorizing a transaction —without context switching between tools.”
   • [claimed-docs] “Accounting AI runs prompts on top of your general ledger. The output reflects your existing accounts and categories.”
   • [claimed-docs] “How to import a PDF or CSV using Accounting AI”
   • [claimed-docs] “Ask anything about your business, instantly, in plain English. "How's my cash trending?" "Am I ready to raise?" "Which months am I actually …”
   • [claimed-docs] “Hand off the busywork. Finding software subscriptions, matching invoices to payments, pulling together what your accountant needs — describe…”

   proof ↗⚑ flag

   Bench—none0/10

   Bench offers human bookkeepers plus automation for data entry and categorization, but there is no evidence of a built-in AI assistant that a user can delegate tasks to conversationally — communication is described as messaging with a human team, not an AI agent. Missing for 10: any mention of an AI assistant/chat feature, agentic task delegation, or AI-driven interaction surface within the product.

   • [claimed-docs] “Effortless messaging with your Bench team for unlimited support or advice, anytime, anywhere”
   • [claimed-docs] “Connect all your financial accounts to automate data entry, speed up your books, reduce errors and save time”
   • [claimed-docs] “Accurate transaction categorization, powered by smart automation with instant guidance”
   • [claimed-docs] “Bench is a human-plus-software service, now under new ownership: "Bench simplifies your small business accounting by combining intuitive sof…”

   proof ↗⚑ flag

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

   weight 2 · round to Puzzle
   Puzzle✓fullclaimed7/10

   Puzzle documents natural-language interaction via its MCP integration with Claude ('instant access to your financial data through natural conversation') and multiple AI features letting users 'ask anything about your business in plain English' or 'ask AI' to categorize transactions, plus AI Close agents automating close tasks from natural prompts. Missing for 10: independent/hands-on verification that NL commands can trigger broader actions (not just queries) and third-party corroboration of the MCP/Claude workflow in practice.

   • [claimed-docs] “Puzzle's MCP (Model Context Protocol) integration lets you connect your Puzzle account directly to Claude, giving you instant access to your…”
   • [claimed-docs] “lets you connect your Puzzle account directly to Claude, giving you instant access to your financial data through natural conversation — no …”
   • [claimed-docs] “Ask anything about your business, instantly, in plain English. "How's my cash trending?" "Am I ready to raise?" "Which months am I actually …”
   • [claimed-docs] “A user can click on a button and “ask AI” for help in categorizing a transaction —without context switching between tools.”
   • [claimed-docs] “Accounting AI runs prompts on top of your general ledger. The output reflects your existing accounts and categories.”
   • [claimed-docs] “AI Close helps you automate and accelerate your month-end process using AI agents to reconcile accounts, review transactions, and surface di…”
   • [claimed-docs] “AI Close helps you automate and accelerate your month-end process using AI agents to reconcile accounts, review transactions, and surface di…”

   proof ↗⚑ flag

   Bench—none0/10

   Bench offers human-support messaging and a client dashboard, but there is no evidence of a natural-language command interface (chat-driven actions, NL query of financial data, etc.); probes for llms.txt, docs-md, and OpenAPI all 404, indicating no AI-native interface layer. missing for 10: any natural-language command/chat-to-action capability, NL query support, or API/agent interface evidence.

   • [probe] “PROBE llms.txt: HTTP 404 at https://www.bench.co/llms.txt”
   • [probe] “PROBE docs-md: HTTP 404 at https://www.bench.co/how-it-works.md”
   • [probe] “PROBE openapi: all candidate paths 404 (https://www.bench.co/openapi.json, https://www.bench.co/swagger.json, https://www.bench.co/api/opena…”
   • [claimed-docs] “Effortless messaging with your Bench team for unlimited support or advice, anytime, anywhere”

   proof ↗⚑ flag

Api quality

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

   weight 2 · round drawn
   Puzzle—none0/10

   Puzzle documents an API (puzzle.io/api) and an MCP integration, but there is no evidence of an interactive API reference or runnable examples; probes for OpenAPI/Swagger specs at expected paths all returned 404, suggesting no such interactive reference exists.

   • [claimed-docs] “With Puzzle's Embedded Accounting—powered by our API, your platform gets automated AI bookkeeping, real-time financial insights, and a clean…”
   • [claimed-docs] “Our API was originally architected for deep platform-to-platform integrations.”
   • [probe] “PROBE openapi: all candidate paths 404 (https://help.puzzle.io/openapi.json, https://help.puzzle.io/swagger.json, https://help.puzzle.io/api…”
   • [probe] “PROBE docs-md: HTTP 404 at https://help.puzzle.io/en/.md”

   proof ↗⚑ flag

   Bench—none0/10

   Bench is a bookkeeping/accounting service with no API, no interactive API reference, and no runnable examples; probes for llms.txt, docs.md, and OpenAPI specs all returned 404. missing for 10: any API reference, interactive docs, runnable code examples.

   • [probe] “PROBE llms.txt: HTTP 404 at https://www.bench.co/llms.txt”
   • [probe] “PROBE docs-md: HTTP 404 at https://www.bench.co/how-it-works.md”
   • [probe] “PROBE openapi: all candidate paths 404 (https://www.bench.co/openapi.json, https://www.bench.co/swagger.json, https://www.bench.co/api/opena…”

   proof ↗⚑ flag

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

   weight 2 · round drawn
   Puzzle—none0/10

   Puzzle has an API and mentions API access docs, but the probe explicitly found no OpenAPI/swagger spec at any candidate path, and no evidence pack item references a downloadable machine-readable spec.

   • [probe] “PROBE openapi: all candidate paths 404 (https://help.puzzle.io/openapi.json, https://help.puzzle.io/swagger.json, https://help.puzzle.io/api…”
   • [claimed-docs] “Our API was originally architected for deep platform-to-platform integrations.”

   proof ↗⚑ flag

   Bench—none0/10

   Probes explicitly show llms.txt, docs.md, and all openapi/swagger candidate paths return 404, and no docs mention any API spec; Bench is a bookkeeping service with no evidence of a machine-readable API for AI agents.

   • [probe] “PROBE llms.txt: HTTP 404 at https://www.bench.co/llms.txt”
   • [probe] “PROBE docs-md: HTTP 404 at https://www.bench.co/how-it-works.md”
   • [probe] “PROBE openapi: all candidate paths 404 (https://www.bench.co/openapi.json, https://www.bench.co/swagger.json, https://www.bench.co/api/opena…”

   proof ↗⚑ flag

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

   weight 1 · round drawn
   Puzzle—none0/10

   No evidence in the pack mentions a sandbox, test, or staging environment for Puzzle's API or product that would let a user test without touching production financial data; only production-focused API and bookkeeping features are documented.

   ⚑ flag

   Bench—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 userRely on versioned APIs with a documented deprecation policy

   weight 2 · round drawn
   Puzzle—none0/10

   The evidence pack shows Puzzle has an API and MCP integration (puzzle-docs-42, puzzle-docs-21) but contains no mention of API versioning scheme or a documented deprecation policy; probes for OpenAPI specs all returned 404. Missing for 10: any versioning scheme documentation, changelog, or deprecation/sunset policy.

   • [claimed-docs] “Our API was originally architected for deep platform-to-platform integrations.”
   • [claimed-docs] “With Puzzle’s Embedded Accounting—powered by our API, your platform gets automated AI bookkeeping, real-time financial insights, and a clean…”
   • [probe] “PROBE openapi: all candidate paths 404 (https://help.puzzle.io/openapi.json, https://help.puzzle.io/swagger.json, https://help.puzzle.io/api…”

   proof ↗⚑ flag

   Bench—none0/10

   Bench is a bookkeeping/tax service with no evidence of a public API at all — probes for OpenAPI specs, docs, and llms.txt all returned 404. No mention of versioning or a deprecation policy anywhere in the evidence pack. Missing for 10: any public API, version numbers, changelog, or deprecation policy documentation.

   • [probe] “PROBE llms.txt: HTTP 404 at https://www.bench.co/llms.txt”
   • [probe] “PROBE docs-md: HTTP 404 at https://www.bench.co/how-it-works.md”
   • [probe] “PROBE openapi: all candidate paths 404 (https://www.bench.co/openapi.json, https://www.bench.co/swagger.json, https://www.bench.co/api/opena…”

   proof ↗⚑ flag

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

   weight 2 · round to Puzzle
   Puzzle~partialclaimed6/10

   Puzzle documents concrete bulk-operation features—batch categorizing transactions, bulk CSV import of manual journal entries, and bulk CSV upload of the Chart of Accounts—plus rule-based automation for categorization, all first-party docs. However, these bulk mechanisms are CSV/rule-based rather than explicitly exposed through the AI/API/MCP layer for an 'AI-native' agent to invoke programmatically at scale. Missing for 10: explicit evidence that the API or MCP integration itself supports issuing bulk operations (not just UI/CSV uploads), and independent/hands-on confirmation of bulk-op reliability at scale.

   • [claimed-docs] “Batch Categorize Transactions”
   • [claimed-docs] “Need to post more than one manual journal entry at a time? Instead of adding them one by one, you can upload a batch of entries all at once …”
   • [claimed-docs] “If you need to bulk add or edit your Chart of Accounts in Puzzle, you can do so by uploading a CSV file directly from your Company Settings.”
   • [claimed-docs] “Creating custom rules”
   • [claimed-docs] “Puzzle's MCP (Model Context Protocol) integration lets you connect your Puzzle account directly to Claude, giving you instant access to your…”

   proof ↗⚑ flag

   Bench—none0/10

   Bench is a human-plus-software bookkeeping service focused on categorization, reporting, and tax filing for individual accounts; no evidence describes any bulk/batch operation capability across many items (e.g., bulk transaction edits, bulk uploads, batch API operations) for AI-native or power users. Probes also show no API/docs surface for programmatic bulk actions.

   • [claimed-docs] “Accurate transaction categorization, powered by smart automation with instant guidance”
   • [claimed-docs] “"Connect all your financial accounts to automate data entry, speed up your books, reduce errors and save time", with "Accurate transaction c…”
   • [probe] “PROBE openapi: all candidate paths 404 (https://www.bench.co/openapi.json, https://www.bench.co/swagger.json, https://www.bench.co/api/opena…”

   proof ↗⚑ flag

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

   weight 3 · round to Puzzle
   Puzzle~partialclaimed5/10

   Puzzle documents 'Creating custom rules' for transaction categorization, letting users define rules that automatically trigger categorization when transactions occur, and AI Close agents that automate reconciliation on close events — concrete evidence of rule-based automation tied to events. However, this is scoped narrowly to bookkeeping categorization/close workflows rather than a general-purpose rules engine spanning arbitrary triggers and actions. Missing for 10: evidence of a broader, user-configurable trigger-action framework beyond categorization/close (e.g., custom webhooks, conditional workflows across modules), and independent confirmation of rule behavior in practice.

   • [claimed-docs] “Creating custom rules”
   • [claimed-docs] “Batch Categorize Transactions”
   • [claimed-docs] “AI Close helps you automate and accelerate your month-end process using AI agents to reconcile accounts, review transactions, and surface di…”
   • [claimed-docs] “AI Close helps you automate and accelerate your month-end process using AI agents to reconcile accounts, review transactions, and surface di…”

   proof ↗⚑ flag

   Bench—none0/10

   Bench is a bookkeeping/accounting service with backend automation for categorization and reminders, but there is no evidence of a user-facing rule engine where AI-native users can define custom triggers/actions on events; the automation described is fixed internal workflow, not configurable rules.

   ⚑ flag

3. ai-native userSchedule recurring jobs or workflows

   weight 2 · round drawn
   Puzzle—none0/10

   Puzzle documents AI-driven automation for close, categorization, and reconciliation, but there is no evidence of any capability to schedule recurring jobs or workflows (e.g., recurring invoices, scheduled reports, or timed automation triggers).

   ⚑ flag

   Bench—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

Ai reconciliation

1. ai-native userHave an agent reconcile a month end-to-end, matching transactions, flagging exceptions, and preparing the close for human review

   weight 3 · round to Puzzle
   Puzzle~partialclaimed6/10

   Puzzle's AI Close feature explicitly uses AI agents to reconcile accounts, review transactions, and surface discrepancies for month-end close, directly matching the story's core ask (puzzle-docs-24, puzzle-docs-39, puzzle-docs-5). However, evidence doesn't confirm full end-to-end autonomy or explicit 'preparation for human review' handoff workflow beyond general review-before-final claims, and there's no independent/hands-on validation of the AI Close agent's reconciliation accuracy or exception-flagging in practice. missing for 10: independent/hands-on verification of AI Close performance, detailed documentation of exception-flagging logic, explicit human-review handoff/approval workflow, and evidence of full end-to-end (not just partial/batch) reconciliation coverage.

   • [claimed-docs] “AI Close helps you automate and accelerate your month-end process using AI agents to reconcile accounts, review transactions, and surface di…”
   • [claimed-docs] “AI Close helps you automate and accelerate your month-end process using AI agents to reconcile accounts, review transactions, and surface di…”
   • [claimed-docs] “Month-End close with AI Close”
   • [claimed-docs] “Hand off the busywork. Finding software subscriptions, matching invoices to payments, pulling together what your accountant needs — describe…”
   • [claimed-docs] “performing a reconciliation is the best way to ensure every single transaction is accounted for and matches your bank statement exactly—givi…”

   proof ↗⚑ flag

   Bench—none0/10

   Bench's evidence describes automated data entry/categorization plus human bookkeepers preparing reports, not an autonomous agent that reconciles a month end-to-end, flags exceptions, and stages a close for human review. No documentation of exception-flagging logic, agent-driven reconciliation workflow, or AI-native interface exists in the pack. Missing for 10: evidence of automated exception detection, an agentic reconciliation workflow, and any API/agent interface for month-end close.

   • [claimed-docs] “"Connect all your financial accounts to automate data entry, speed up your books, reduce errors and save time", with "Accurate transaction c…”
   • [claimed-docs] “Accurate transaction categorization, powered by smart automation with instant guidance”
   • [claimed-docs] “"Real-time reporting. Access or download your updated income statement or balance sheet at all times" — the client-facing surface is reports…”
   • [claimed-docs] “Bench is a human-plus-software service, now under new ownership: "Bench simplifies your small business accounting by combining intuitive sof…”

   proof ↗⚑ flag

Bank feeds

1. founderConnect bank and credit-card accounts so transactions import automatically every day

   weight 3 · round drawn
   Puzzle✓fullclaimed7/10

   Puzzle's docs describe connecting bank/credit-card accounts and third-party sources (Brex, Ramp, Stripe) with automated categorization (98% categorized) and reconciliation workflows, which directly supports the daily-import banking-reconciliation story (puzzle-docs-29, -15, -14, -13, -18, -19). However, there's no explicit statement about daily import cadence or independent verification of reliability. Missing for 10: explicit 'daily auto-sync' language, third-party/hands-on confirmation of import frequency and accuracy.

   • [claimed-docs] “performing a reconciliation is the best way to ensure every single transaction is accounted for and matches your bank statement exactly—givi…”
   • [claimed-docs] “98% transactions categorized When you connect Brex with Puzzle”
   • [claimed-docs] “2x faster to close books When you connect Ramp with Puzzle”
   • [claimed-docs] “3 min to reconcile revenue When you connect Stripe with Puzzle”
   • [claimed-docs] “Up to 98% auto-categorization”
   • [claimed-docs] “Puzzle customers who fully onboard and connect supported integrations can reduce the time required to complete their monthly close by 50% or…”

   proof ↗⚑ flag

   Bench✓fullclaimed7/10

   Bench's docs explicitly describe connecting all financial accounts to automate daily data entry and transaction categorization, which directly matches the story of automatic daily import via connected bank/credit card accounts. Missing for 10: no independent/hands-on corroboration of connection reliability, no detail on supported bank list or sync frequency specifics beyond marketing copy.

   • [claimed-docs] “Connect all your financial accounts to automate data entry, speed up your books, reduce errors and save time”
   • [claimed-docs] “Accurate transaction categorization, powered by smart automation with instant guidance”
   • [claimed-docs] “"Connect all your financial accounts to automate data entry, speed up your books, reduce errors and save time", with "Accurate transaction c…”

   proof ↗⚑ flag

Categorization

1. founderGet automatic categorization suggestions for imported transactions that learn from my corrections

   weight 2 · round to Puzzle
   Puzzle~partialclaimed6/10

   Puzzle clearly ships automatic transaction categorization with high accuracy claims (up to 98%), custom rules, batch categorization, and an 'ask AI' assist button, showing an active categorization engine tied to user-configurable rules. However, there is no explicit documentation describing a feedback loop where the AI model learns from user corrections over time (as opposed to static rules the user manually creates). missing for 10: explicit description of ML/AI learning from user corrections, evidence of improved suggestion accuracy after corrections, independent confirmation of this adaptive behavior.

   • [claimed-docs] “Creating custom rules”
   • [claimed-docs] “Batch Categorize Transactions”
   • [claimed-docs] “Up to 98% auto-categorization”
   • [claimed-docs] “A user can click on a button and “ask AI” for help in categorizing a transaction —without context switching between tools.”
   • [claimed-docs] “98% transactions categorized When you connect Brex with Puzzle”

   proof ↗⚑ flag

   Bench~partialclaimed4/10

   Bench documents automated, 'smart automation' transaction categorization with 'instant guidance' (bench-docs-4, bench-supp-automation), which covers the core categorization-suggestion capability, but since Bench is a human-plus-software bookkeeping service, human bookkeepers—not the software—typically apply corrections, and there is no evidence describing a feedback loop where the system learns from founder corrections. Missing for 10: any documentation of a correction-learning mechanism, founder-facing feedback UI, or improvement in suggestion accuracy over time.

   • [claimed-docs] “Accurate transaction categorization, powered by smart automation with instant guidance”
   • [claimed-docs] “"Connect all your financial accounts to automate data entry, speed up your books, reduce errors and save time", with "Accurate transaction c…”
   • [claimed-docs] “Bench is a human-plus-software service, now under new ownership: "Bench simplifies your small business accounting by combining intuitive sof…”

   proof ↗⚑ flag

Reconciliation

1. accountantReconcile each bank account against the ledger with matched, unmatched, and duplicate transactions surfaced for review

   weight 3 · round to Puzzle
   Puzzle✓fullclaimed7/10

   Puzzle has a dedicated reconciliation feature for bank/credit card accounts explicitly designed to ensure every transaction is accounted for and matches the bank statement (puzzle-docs-29), plus AI Close automates reconciling accounts, reviewing transactions, and surfacing discrepancies (puzzle-docs-24, puzzle-docs-39). This covers the core reconcile-and-surface-for-review workflow an accountant needs. Missing for 10: explicit documentation of distinct 'matched/unmatched/duplicate' transaction states or UI surfacing terminology, and independent/hands-on validation beyond vendor docs.

   • [claimed-docs] “performing a reconciliation is the best way to ensure every single transaction is accounted for and matches your bank statement exactly—givi…”
   • [claimed-docs] “AI Close helps you automate and accelerate your month-end process using AI agents to reconcile accounts, review transactions, and surface di…”
   • [claimed-docs] “AI Close helps you automate and accelerate your month-end process using AI agents to reconcile accounts, review transactions, and surface di…”

   proof ↗⚑ flag

   Bench—none0/10

   Bench's evidence covers account connection, automated categorization, and reporting, but no citation describes a reconciliation workflow that surfaces matched, unmatched, and duplicate transactions for accountant review. missing for 10: explicit reconciliation UI/workflow, matched/unmatched transaction views, duplicate detection and review surface.

   • [claimed-docs] “Connect all your financial accounts to automate data entry, speed up your books, reduce errors and save time”
   • [claimed-docs] “Accurate transaction categorization, powered by smart automation with instant guidance”
   • [claimed-docs] “"Connect all your financial accounts to automate data entry, speed up your books, reduce errors and save time", with "Accurate transaction c…”
   • [claimed-docs] “"Real-time reporting. Access or download your updated income statement or balance sheet at all times" — the client-facing surface is reports…”

   proof ↗⚑ flag

Bill management

1. accountantCapture vendor bills, route them for approval, and schedule payments from one workflow

   weight 2 · round drawn
   Puzzle—none0/10

   The evidence pack covers bookkeeping, categorization, close automation, invoicing (AR), and reporting, but contains no mention of vendor bill capture, approval routing, or payment scheduling (AP workflow) anywhere in the docs or help center listings.

   ⚑ flag

   Bench—none0/10

   Bench is a bookkeeping/tax service focused on categorization, reporting, and human bookkeeper support; no evidence of bill capture (AP inbox/OCR), approval routing, or scheduled bill payment workflows anywhere in the docs, pricing, or banking pages.

   • [claimed-docs] “Connect all your financial accounts to automate data entry, speed up your books, reduce errors and save time”
   • [claimed-docs] “Accurate transaction categorization, powered by smart automation with instant guidance”
   • [claimed-docs] “"Connect all your financial accounts to automate data entry, speed up your books, reduce errors and save time", with "Accurate transaction c…”
   • [claimed-docs] “"Banking by Mainstreet — Get dedicated business accounts, debit cards, and automated financial management tools that integrate seamlessly wi…”
   • [claimed-docs] “Published service pricing: Bookkeeping Grow from $199/month ("An accessible, lighter-touch plan—built for lean businesses seeking financial …”

   proof ↗⚑ flag

Payables

1. accountantSee accounts payable aging and vendor balances at any point in time

   weight 2 · round drawn
   Puzzle—none0/10

   The evidence pack covers bookkeeping, categorization, month-end close, invoicing (AR), reconciliation, and reporting exports, but contains no mention of accounts payable, vendor bills, bill tracking, or AP aging reports. Missing for 10: any documentation of a bills/AP module, vendor balance tracking, or an AP aging report feature.

   • [claimed-docs] “you can export a full general ledger and financial statements on either cash or accrual basis in excel, csv, and pdf formats.”
   • [claimed-docs] “Yes, you can export a full general ledger and financial statements on either cash or accrual basis in excel, csv, and pdf formats.”
   • [claimed-docs] “Puzzle's invoice editor lets you build a professional, branded invoice and see exactly how it will look as you go!”

   proof ↗⚑ flag

   Bench—none0/10

   Bench's evidence covers bookkeeping automation, categorization, income statement/balance sheet reporting, banking and tax prep, but nowhere mentions accounts payable aging, vendor balances, or bill-tracking reports at a point in time. Bench appears to focus on income statement/balance sheet reporting rather than AP-specific aging reports.

   • [claimed-docs] “Real-time reporting. Access or download your updated income statement or balance sheet at all times”
   • [claimed-docs] “"Real-time reporting. Access or download your updated income statement or balance sheet at all times" — the client-facing surface is reports…”
   • [claimed-docs] “Accurate transaction categorization, powered by smart automation with instant guidance”

   proof ↗⚑ flag

Close process

1. accountantLock closed periods so prior-period numbers cannot change without an explicit reopening

   weight 2 · round drawn
   Puzzle—none0/10

   The evidence covers AI-assisted month-end close, reconciliation, and journal entry reversals, but nowhere mentions locking closed periods or requiring an explicit reopening step to prevent prior-period changes. This is a standard, applicable accounting-software axis, but no documentation or community evidence confirms it exists in Puzzle.

   ⚑ flag

   Bench—none0/10

   No evidence describes period locking, close controls, or reopening workflows; Bench is presented as a human-plus-software bookkeeping service with reporting/downloads, not a ledger system with period-lock controls exposed to the accountant/customer. missing for 10: any mention of period closing, locking prior periods, or explicit reopening controls.

   • [claimed-docs] “"Real-time reporting. Access or download your updated income statement or balance sheet at all times" — the client-facing surface is reports…”
   • [claimed-docs] “Real-time reporting. Access or download your updated income statement or balance sheet at all times”

   proof ↗⚑ flag

2. accountantRun a month-end close against a checklist that tracks reconciliations, accruals, and review sign-offs

   weight 3 · round to Puzzle
   Puzzle~partialclaimed5/10

   Puzzle's 'AI Close' explicitly automates month-end close by reconciling accounts, reviewing transactions, and surfacing discrepancies (puzzle-docs-24/39), and separate reconciliation docs describe matching bank/credit card transactions (puzzle-docs-29). Cash and accrual accounting are supported (puzzle-docs-20), but there's no evidence of a structured checklist UI, accrual-specific tasks, or multi-person review/sign-off workflow tracking. Missing for 10: explicit checklist/task-list feature, accrual entry tracking, and reviewer sign-off/approval workflow evidence.

   • [claimed-docs] “AI Close helps you automate and accelerate your month-end process using AI agents to reconcile accounts, review transactions, and surface di…”
   • [claimed-docs] “AI Close helps you automate and accelerate your month-end process using AI agents to reconcile accounts, review transactions, and surface di…”
   • [claimed-docs] “performing a reconciliation is the best way to ensure every single transaction is accounted for and matches your bank statement exactly—givi…”
   • [claimed-docs] “Cash **_and_** accrual accounting”
   • [claimed-docs] “Month-End close with AI Close”

   proof ↗⚑ flag

   Bench—none0/10

   Bench markets itself as a done-for-you bookkeeping/tax service with messaging, document upload, categorization, and real-time reports, but nothing in the evidence describes a month-end close checklist tracking reconciliations, accruals, or review sign-offs — only generic 'timely reminders' are mentioned, which is not the same as a structured close workflow.

   • [claimed-docs] “Get timely reminders to stay on top of your financial tasks and deadlines”
   • [claimed-docs] “"Connect all your financial accounts to automate data entry, speed up your books, reduce errors and save time", with "Accurate transaction c…”
   • [claimed-docs] “"Real-time reporting. Access or download your updated income statement or balance sheet at all times" — the client-facing surface is reports…”

   proof ↗⚑ flag

Collaboration

1. founderInvite my external accountant or bookkeeper with role-scoped access to review and adjust the books

   weight 3 · round drawn
   Puzzle—none0/10

   Puzzle markets an accountant-facing page (migration, real-time reports, startup metrics) but no evidence describes actual user invitation, permission levels, or role-scoped access controls for external collaborators. Missing for 10: explicit invite flow, role/permission settings, and any documentation of accountant-specific access restrictions.

   • [claimed-docs] “Migrate from QuickBooks without lifting a finger or losing your client's data”
   • [claimed-docs] “Instantly see startup metrics like ARR, burn, & runway”
   • [claimed-docs] “Get real-time financial reports with one click”

   proof ↗⚑ flag

   Bench—none0/10

   Bench's evidence describes a done-for-you bookkeeping service where Bench's own staff manage the books and communicate with the founder, with no mention of inviting an external accountant/bookkeeper as a role-scoped collaborator with adjustable permissions inside a shared ledger. The 'Add a bookkeeper to your team' claim refers to Bench's own hire-a-bookkeeper offering working inside the founder's existing systems, not the founder granting scoped access to their own outside accountant.

   • [claimed-docs] “Add a bookkeeper to your team who works with your existing systems”
   • [claimed-docs] “Published service pricing: Bookkeeping Grow from $199/month ("An accessible, lighter-touch plan—built for lean businesses seeking financial …”
   • [claimed-docs] “Bench is a human-plus-software service, now under new ownership: "Bench simplifies your small business accounting by combining intuitive sof…”

   proof ↗⚑ flag

Data access

1. developerPull the full general ledger and trial balance through a documented API in open formats

   weight 3 · round to Puzzle
   Puzzle~partialprobed4/10

   Puzzle documents that users can export a full general ledger and financial statements (cash or accrual) in Excel, CSV, and PDF (open formats), and separately references an API 'architected for deep platform-to-platform integrations,' but no evidence ties GL/trial-balance export specifically to the documented API — exports appear to be a UI feature, and probes for OpenAPI/API docs (openapi.json, swagger.json) all returned 404, suggesting no publicly discoverable API reference. Missing for 10: explicit API endpoint/documentation for pulling GL and trial balance, confirmation of trial balance as a report type, and a working/discoverable OpenAPI spec.

   • [claimed-docs] “you can export a full general ledger and financial statements on either cash or accrual basis in excel, csv, and pdf formats.”
   • [claimed-docs] “you can export a full general ledger and financial statements on either cash or accrual basis in excel, csv, and pdf formats”
   • [claimed-docs] “Yes, you can export a full general ledger and financial statements on either cash or accrual basis in excel, csv, and pdf formats.”
   • [claimed-docs] “Our API was originally architected for deep platform-to-platform integrations.”
   • [probe] “PROBE openapi: all candidate paths 404 (https://help.puzzle.io/openapi.json, https://help.puzzle.io/swagger.json, https://help.puzzle.io/api…”

   proof ↗⚑ flag

   Bench—none0/10

   Bench only offers report viewing/downloads (income statement, balance sheet) via its web app; there is no documented API, no OpenAPI/Swagger spec, and probes for llms.txt, markdown docs, and OpenAPI endpoints all returned 404s, so developers have no documented way to pull the GL or trial balance in open formats.

   • [probe] “PROBE llms.txt: HTTP 404 at https://www.bench.co/llms.txt”
   • [probe] “PROBE docs-md: HTTP 404 at https://www.bench.co/how-it-works.md”
   • [probe] “PROBE openapi: all candidate paths 404 (https://www.bench.co/openapi.json, https://www.bench.co/swagger.json, https://www.bench.co/api/opena…”
   • [claimed-docs] “"Real-time reporting. Access or download your updated income statement or balance sheet at all times" — the client-facing surface is reports…”

   proof ↗⚑ flag

Integrations

1. founderConnect my banking, payments, and payroll systems so their activity flows into the ledger automatically

   weight 2 · round drawn
   Puzzle~partialclaimed6/10

   Puzzle documents direct connections to banking/payment platforms like Stripe, Ramp, and Brex with concrete automation stats (transaction categorization, faster close, revenue reconciliation) and general bank/credit card reconciliation flows, showing banking and payments activity flows into the ledger automatically. However, no evidence names a payroll system integration (e.g., Gusto, Rippling) or shows payroll data flowing into the ledger. Missing for 10: explicit payroll system integrations, independent/hands-on confirmation of banking+payroll auto-sync working end-to-end.

   • [claimed-docs] “3 min to reconcile revenue When you connect Stripe with Puzzle”
   • [claimed-docs] “2x faster to close books When you connect Ramp with Puzzle”
   • [claimed-docs] “98% transactions categorized When you connect Brex with Puzzle”
   • [claimed-docs] “Is my revenue automatically ASC 606 compliant in Puzzle?”
   • [claimed-docs] “performing a reconciliation is the best way to ensure every single transaction is accounted for and matches your bank statement exactly—givi…”
   • [claimed-docs] “Our API was originally architected for deep platform-to-platform integrations.”

   proof ↗⚑ flag

   Bench~partialclaimed6/10

   Bench's docs claim customers can 'connect all your financial accounts to automate data entry' and offers integrated banking (Banking by Bench/Mainstreet), which supports bank/payment account linking flowing into bookkeeping. However, payroll integration is never explicitly mentioned, and the ledger itself is not customer-facing/editable — it's a human-plus-software service producing reports rather than a live self-serve ledger. missing for 10: explicit payroll system integration, technical detail on connection breadth (which banks/payment processors), independent/hands-on confirmation of automated data flow reliability.

   • [claimed-docs] “Connect all your financial accounts to automate data entry, speed up your books, reduce errors and save time”
   • [claimed-docs] “"Connect all your financial accounts to automate data entry, speed up your books, reduce errors and save time", with "Accurate transaction c…”
   • [claimed-docs] “Get dedicated business accounts, debit cards, and automated financial management tools that integrate seamlessly with your bookkeeping opera…”
   • [claimed-docs] “"Banking by Mainstreet — Get dedicated business accounts, debit cards, and automated financial management tools that integrate seamlessly wi…”
   • [claimed-docs] “"Real-time reporting. Access or download your updated income statement or balance sheet at all times" — the client-facing surface is reports…”

   proof ↗⚑ flag

Migration

1. founderMigrate historical books from another accounting system without losing transaction detail

   weight 1 · round to Puzzle
   Puzzle✓fullclaimed6/10

   Puzzle explicitly markets QuickBooks migration ('Migrate from QuickBooks without lifting a finger or losing your client's data', 'Switch in minutes. Free migration.'), directly addressing historical books migration without data loss for founders switching accounting systems. Missing for 10: independent/hands-on verification of transaction-level fidelity during migration, detail on which source systems beyond QuickBooks are supported, and documented handling of complex historical data edge cases (multi-entity, custom categories).

   • [claimed-docs] “Migrate from QuickBooks without lifting a finger or losing your client's data”
   • [claimed-docs] “Switch in minutes. Free migration.”

   proof ↗⚑ flag

   Bench~partialclaimed5/10

   Bench explicitly offers 'Catch Up Bookkeeping' for businesses behind on their books, implying it can reconstruct/backfill historical transaction data, and its human-bookkeeper model suggests manual handling of past records. However, there is no explicit documentation of migrating data from another accounting system (e.g., QuickBooks, Xero) with transaction-level fidelity preserved, no import tooling, and no evidence addressing loss/retention of transaction detail during migration. Missing for 10: explicit migration/import process from other platforms, evidence of transaction-level data preservation, any tooling or API for historical data transfer.

   • [claimed-docs] “Catch up bookkeeping services for small businesses, no matter how far behind they are”
   • [claimed-docs] “Add a bookkeeper to your team who works with your existing systems”
   • [claimed-docs] “The buying motion is demo-led for service fit: "Book a demo with our friendly team of experts. Not sure where to start or which accounting s…”

   proof ↗⚑ flag

Invoicing

1. founderCreate and send invoices with online payment options and see when they are viewed and paid

   weight 3 · round to Puzzle
   Puzzle~partialclaimed4/10

   Puzzle has a documented invoice creation/customization feature with live preview and branded PDFs/emails (puzzle-docs-28, 40, 47), but there is no evidence of built-in online payment collection or of invoice view/paid status tracking, which are core parts of this story. Missing for 10: online payment processing integration for invoices, and view/paid status notifications or dashboards.

   • [claimed-docs] “The editor opens in a split view: the form on the left, a live preview on the right that updates as you type — so you always know what your …”
   • [claimed-docs] “Puzzle's invoice editor lets you build a professional, branded invoice and see exactly how it will look as you go!”
   • [claimed-docs] “Adding your logo and brand color takes just a few steps and applies everywhere your invoice appears — the live preview, downloaded PDFs, and…”

   proof ↗⚑ flag

   Bench—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

2. founderSet up recurring invoices and automatic payment reminders for overdue customers

   weight 2 · round drawn
   Puzzle—none0/10

   Evidence confirms Puzzle has an invoice creation/customization editor (branding, live preview) but no documentation mentions recurring/scheduled invoices or automatic overdue payment reminders. Missing for 10: any mention of recurring invoice scheduling, dunning/reminder automation, or overdue-tracking workflows.

   • [claimed-docs] “The editor opens in a split view: the form on the left, a live preview on the right that updates as you type — so you always know what your …”
   • [claimed-docs] “Puzzle's invoice editor lets you build a professional, branded invoice and see exactly how it will look as you go!”
   • [claimed-docs] “Adding your logo and brand color takes just a few steps and applies everywhere your invoice appears — the live preview, downloaded PDFs, and…”

   proof ↗⚑ flag

   Bench—none0/10

   Bench is a bookkeeping/tax service focused on categorization, reporting, and tax filing; there is no mention of invoicing, accounts receivable, recurring invoices, or customer payment reminders anywhere in the evidence pack.

   ⚑ flag

Receivables

1. accountantTrack accounts receivable with aging reports and per-customer balances

   weight 2 · round to Puzzle
   Puzzle~partialclaimed3/10

   Puzzle documents invoice creation/customization (docs-28, docs-40, docs-47) and general ledger/financial statement exports (docs-26, docs-37, docs-45), which are adjacent to AR tracking, but no evidence explicitly describes an aging report or per-customer balance view. missing for 10: explicit AR aging report feature, per-customer balance/statement view, documentation tying invoicing to AR tracking workflows.

   • [claimed-docs] “The editor opens in a split view: the form on the left, a live preview on the right that updates as you type — so you always know what your …”
   • [claimed-docs] “Puzzle's invoice editor lets you build a professional, branded invoice and see exactly how it will look as you go!”
   • [claimed-docs] “Adding your logo and brand color takes just a few steps and applies everywhere your invoice appears — the live preview, downloaded PDFs, and…”
   • [claimed-docs] “you can export a full general ledger and financial statements on either cash or accrual basis in excel, csv, and pdf formats.”
   • [claimed-docs] “you can export a full general ledger and financial statements on either cash or accrual basis in excel, csv, and pdf formats”

   proof ↗⚑ flag

   Bench—none0/10

   Bench's evidence covers reporting (income statement, balance sheet), categorization, and banking, but there is no mention of accounts receivable aging reports or per-customer balance tracking anywhere in the docs or probes. Missing for 10: any AR aging report feature, per-customer balance tracking, invoicing/AR module documentation.

   • [claimed-docs] “Real-time reporting. Access or download your updated income statement or balance sheet at all times”
   • [claimed-docs] “"Real-time reporting. Access or download your updated income statement or balance sheet at all times" — the client-facing surface is reports…”

   proof ↗⚑ flag

Ai ledger

1. ai-native userHave an agent draft and post journal entries via the API and leave them in a review queue for my accountant

   weight 3 · round to Puzzle
   Puzzle~partialclaimed4/10

   Puzzle's API supports platform-to-platform integration and journal entries can be created/edited/imported (docs-42, docs-43, docs-31/35), and AI Close uses agents to reconcile and 'review transactions' (docs-24/39), suggesting some agent-assisted, human-reviewed workflow exists. However, there is no documented API endpoint or feature specifically for an agent to draft a journal entry and place it into a distinct 'review queue' state pending accountant approval — the evidence only shows batch/manual entry creation and general AI-close reconciliation, not a draft-and-queue mechanism. Missing for 10: explicit API support for draft/pending journal entry status, a documented review-queue or approval workflow accessible to accountants, and confirmation that AI-drafted entries route through this queue rather than posting directly.

   • [claimed-docs] “Our API was originally architected for deep platform-to-platform integrations.”
   • [claimed-docs] “Need to post more than one manual journal entry at a time? Instead of adding them one by one, you can upload a batch of entries all at once …”
   • [claimed-docs] “AI Close helps you automate and accelerate your month-end process using AI agents to reconcile accounts, review transactions, and surface di…”
   • [claimed-docs] “AI Close helps you automate and accelerate your month-end process using AI agents to reconcile accounts, review transactions, and surface di…”
   • [claimed-docs] “Puzzle automatically creates a voiding (reversing) journal entry for the original and posts a new entry reflecting your changes.”
   • [claimed-docs] “Puzzle automatically creates a voiding (reversing) journal entry for the original and posts a new entry reflecting your changes”

   proof ↗⚑ flag

   Bench—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

Chart of accounts

1. accountantCustomize the chart of accounts with account types, sub-accounts, and class or tag dimensions

   weight 3 · round to Puzzle
   Puzzle~partialclaimed6/10

   Puzzle docs confirm chart-of-accounts customization via bulk CSV upload/edit and a flexible Classes dimension with unlimited groups, supporting tag/class-based segmentation. However there's no explicit evidence of configuring account types or creating sub-accounts within the CoA structure. Missing for 10: explicit sub-account hierarchy support, account-type customization details, independent/hands-on corroboration.

   • [claimed-docs] “If you need to bulk add or edit your Chart of Accounts in Puzzle, you can do so by uploading a CSV file directly from your Company Settings.”
   • [claimed-docs] “a user can create an unlimited number of Groups for Classes”

   proof ↗⚑ flag

   Bench—none0/10

   Bench is a done-for-you bookkeeping service where a human bookkeeper manages the books; the evidence pack shows no self-serve chart-of-accounts customization, account types, sub-accounts, or class/tag dimension configuration exposed to the accountant/customer. The client-facing surface is reports and downloads, not ledger/CoA editing. Missing for 10: any mention of CoA management, account type/sub-account creation, or class/tag dimension tagging capability.

   • [claimed-docs] “"Real-time reporting. Access or download your updated income statement or balance sheet at all times" — the client-facing surface is reports…”
   • [claimed-docs] “Real-time reporting. Access or download your updated income statement or balance sheet at all times”
   • [claimed-docs] “Bench is a human-plus-software service, now under new ownership: "Bench simplifies your small business accounting by combining intuitive sof…”

   proof ↗⚑ flag

Journal entries

1. accountantPost manual journal entries with balanced debits, credits, and supporting memos

   weight 3 · round to Puzzle
   Puzzle~partialclaimed7/10

   Puzzle explicitly supports posting manual journal entries, including bulk CSV import for multiple entries and automatic reversing entries when editing an existing entry, indicating standard double-entry JE functionality with edit history. However, there is no explicit documentation confirming enforced debit/credit balancing validation or a dedicated memo/notes field on entries. missing for 10: explicit debit=credit validation, documented memo/description field support.

   • [claimed-docs] “Puzzle automatically creates a voiding (reversing) journal entry for the original and posts a new entry reflecting your changes.”
   • [claimed-docs] “Puzzle automatically creates a voiding (reversing) journal entry for the original and posts a new entry reflecting your changes”
   • [claimed-docs] “Need to post more than one manual journal entry at a time? Instead of adding them one by one, you can upload a batch of entries all at once …”
   • [claimed-docs] “If you need to bulk add or edit your Chart of Accounts in Puzzle, you can do so by uploading a CSV file directly from your Company Settings.”

   proof ↗⚑ flag

   Bench—none0/10

   Bench is a done-for-you bookkeeping service where Bench's own staff (not the customer) manage the books; evidence explicitly states the client-facing surface is reports/downloads, 'not a ledger the customer posts to' (bench-supp-reporting), with no mention of a journal-entry interface, debit/credit balancing, or memo fields for accountant users. Missing for 10: any UI or workflow for entering/posting journal entries, debit/credit validation, and memo/supporting-documentation attachment on entries.

   • [claimed-docs] “"Real-time reporting. Access or download your updated income statement or balance sheet at all times" — the client-facing surface is reports…”
   • [claimed-docs] “Real-time reporting. Access or download your updated income statement or balance sheet at all times”
   • [claimed-docs] “Bench is a human-plus-software service, now under new ownership: "Bench simplifies your small business accounting by combining intuitive sof…”

   proof ↗⚑ flag

Ai audit

1. ai-native userHave an agent monitor the ledger and flag anomalies in spend or duplicate transactions before the close

   weight 3 · round to Puzzle
   Puzzle✓fullclaimed7/10

   Puzzle's AI Close feature explicitly uses AI agents to 'reconcile accounts, review transactions, and surface discrepancies' as part of the month-end close workflow, directly matching the anomaly-flagging story, and reconciliation docs confirm matching transactions against bank statements to catch mismatches. Missing for 10: explicit mention of duplicate-transaction detection specifically, and independent/hands-on corroboration beyond vendor docs.

   • [claimed-docs] “AI Close helps you automate and accelerate your month-end process using AI agents to reconcile accounts, review transactions, and surface di…”
   • [claimed-docs] “AI Close helps you automate and accelerate your month-end process using AI agents to reconcile accounts, review transactions, and surface di…”
   • [claimed-docs] “Month-End close with AI Close”
   • [claimed-docs] “performing a reconciliation is the best way to ensure every single transaction is accounted for and matches your bank statement exactly—givi…”

   proof ↗⚑ flag

   Bench—none0/10

   Bench is a human-plus-software bookkeeping service with automated categorization and real-time reporting, but there is no evidence of an AI agent that actively monitors the ledger for anomalies or duplicate transactions ahead of close; the closest capability is human bookkeeper review and periodic reminders. missing for 10: agent-based anomaly detection, duplicate-transaction flagging, any pre-close audit automation.

   • [claimed-docs] “Accurate transaction categorization, powered by smart automation with instant guidance”
   • [claimed-docs] “"Connect all your financial accounts to automate data entry, speed up your books, reduce errors and save time", with "Accurate transaction c…”
   • [claimed-docs] “Real-time reporting. Access or download your updated income statement or balance sheet at all times”
   • [claimed-docs] “"Real-time reporting. Access or download your updated income statement or balance sheet at all times" — the client-facing surface is reports…”

   proof ↗⚑ flag

Audit trail

1. accountantSee a complete audit trail of who changed what and when across the entire ledger

   weight 3 · round drawn
   Puzzle—none0/10

   No evidence describes an audit trail, change log, or user-attributed history of edits across the ledger. The closest related item (docs-31/35) only shows that editing a journal entry creates a reversing entry rather than in-place edits, which is not the same as a queryable audit trail of who changed what and when.

   ⚑ flag

   Bench—none0/10

   Bench is a service where a human bookkeeper does the books, and the client-facing surface is reports/downloads rather than a ledger with visible change history; no evidence of any audit-trail, version history, or user-attribution feature is present anywhere in the pack, and probes confirm no docs/API surfacing such capability. missing for 10: any mention of audit logs, change tracking, user attribution, or timestamped edit history across the ledger.

   • [claimed-docs] “"Real-time reporting. Access or download your updated income statement or balance sheet at all times" — the client-facing surface is reports…”
   • [probe] “PROBE docs-md: HTTP 404 at https://www.bench.co/how-it-works.md”
   • [probe] “PROBE openapi: all candidate paths 404 (https://www.bench.co/openapi.json, https://www.bench.co/swagger.json, https://www.bench.co/api/opena…”

   proof ↗⚑ flag

Multi entity

1. accountantManage multiple legal entities and produce consolidated financials across them

   weight 2 · round to Puzzle
   Puzzle~partialclaimed5/10

   Puzzle explicitly supports managing multiple companies/entities under one login, and documents a Joiin integration specifically for 'Multi-Entity Reporting', showing entity management and a path to consolidated reporting exist. However, true consolidated financials appear to require a third-party integration (Joiin) rather than a native consolidation engine, and there's no evidence of native cross-entity elimination, intercompany accounting, or a built-in consolidated financial statement generator. Missing for 10: native consolidation/elimination features, evidence of intercompany transaction handling, and independent confirmation the Joiin integration fully satisfies audit-grade consolidated reporting.

   • [claimed-docs] “Puzzle supports managing multiple companies (entities) under one login.”
   • [claimed-docs] “Puzzle supports managing **multiple companies (entities)** under one login.”
   • [claimed-docs] “Puzzle + Joiin Integration: Multi-Entity Reporting”

   proof ↗⚑ flag

   Bench—none0/10

   Bench is a single-entity small-business bookkeeping/tax service; nothing in the evidence mentions multi-entity management, consolidation, or intercompany eliminations. Missing for 10: multi-entity account structure, consolidated financial statements across entities, intercompany reconciliation/eliminations.

   • [claimed-docs] “Real-time reporting. Access or download your updated income statement or balance sheet at all times”
   • [claimed-docs] “"Real-time reporting. Access or download your updated income statement or balance sheet at all times" — the client-facing surface is reports…”
   • [claimed-docs] “Bench is a human-plus-software service, now under new ownership: "Bench simplifies your small business accounting by combining intuitive sof…”

   proof ↗⚑ flag

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

   weight 2 · round to Puzzle
   Puzzle~partialprobed3/10

   Puzzle exposes an API for embedded accounting and bookkeeping data (puzzle-docs-1, puzzle-docs-42) plus CSV/journal bulk import tools, but there is no documentation claiming or demonstrating that every UI action (invoicing, reconciliation, categorization rules, AI Close workflows) is exposed via the API, and a probe for an OpenAPI/swagger spec returned 404s, suggesting the API surface is not fully public or documented. missing for 10: explicit API-UI parity documentation, public OpenAPI spec, evidence of invoice creation/reconciliation/AI Close being API-accessible.

   • [claimed-docs] “With Puzzle's Embedded Accounting—powered by our API, your platform gets automated AI bookkeeping, real-time financial insights, and a clean…”
   • [claimed-docs] “Our API was originally architected for deep platform-to-platform integrations.”
   • [claimed-docs] “Need to post more than one manual journal entry at a time? Instead of adding them one by one, you can upload a batch of entries all at once …”
   • [claimed-docs] “If you need to bulk add or edit your Chart of Accounts in Puzzle, you can do so by uploading a CSV file directly from your Company Settings.”
   • [probe] “PROBE openapi: all candidate paths 404 (https://help.puzzle.io/openapi.json, https://help.puzzle.io/swagger.json, https://help.puzzle.io/api…”

   proof ↗⚑ flag

   Bench—none0/10

   No API, OpenAPI spec, or developer documentation exists; probes for llms.txt, docs.md, and openapi.json all return 404s, and all evidence describes a UI/human-service product with no programmatic access surface.

   • [probe] “PROBE llms.txt: HTTP 404 at https://www.bench.co/llms.txt”
   • [probe] “PROBE docs-md: HTTP 404 at https://www.bench.co/how-it-works.md”
   • [probe] “PROBE openapi: all candidate paths 404 (https://www.bench.co/openapi.json, https://www.bench.co/swagger.json, https://www.bench.co/api/opena…”

   proof ↗⚑ flag

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

   weight 3 · round to Puzzle
   Puzzle~partialclaimed7/10

   Puzzle documents exporting a full general ledger and financial statements in open formats (Excel, CSV, PDF) for both cash and accrual basis, which supports the core 'export and leave' requirement for financial data. missing for 10: evidence of exporting full account configuration/rules/invoice templates/all metadata, independent user corroboration of the export process, and no explicit data-portability/deletion guarantee.

   • [claimed-docs] “you can export a full general ledger and financial statements on either cash or accrual basis in excel, csv, and pdf formats.”
   • [claimed-docs] “you can export a full general ledger and financial statements on either cash or accrual basis in excel, csv, and pdf formats”
   • [claimed-docs] “Yes, you can export a full general ledger and financial statements on either cash or accrual basis in excel, csv, and pdf formats.”

   proof ↗⚑ flag

   Bench—none0/10

   No evidence of data export in open/portable formats or account portability; reporting is limited to viewing/downloading income statements or balance sheets, not a full data export mechanism, and probes for llms.txt/openapi/docs all 404. missing for 10: explicit data export feature (CSV/QBO/API), documented data portability policy, evidence of ability to leave with full underlying data.

   • [claimed-docs] “"Real-time reporting. Access or download your updated income statement or balance sheet at all times" — the client-facing surface is reports…”
   • [probe] “PROBE llms.txt: HTTP 404 at https://www.bench.co/llms.txt”
   • [probe] “PROBE docs-md: HTTP 404 at https://www.bench.co/how-it-works.md”
   • [probe] “PROBE openapi: all candidate paths 404 (https://www.bench.co/openapi.json, https://www.bench.co/swagger.json, https://www.bench.co/api/opena…”

   proof ↗⚑ flag

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

   weight 2 · round drawn
   Puzzle—none0/10

   No evidence pack item mentions data residency, regional storage options, or any control over where customer data is hosted; Puzzle is an accounting SaaS and this is a fair question for a data platform, but nothing documents such a capability.

   ⚑ flag

   Bench—none0/10

   No evidence in the pack mentions data residency, region selection, or storage location controls; Bench's marketing focuses on bookkeeping automation, tax, and banking, with no data-residency documentation and 404s on technical/API probes that might have revealed such settings.

   • [probe] “PROBE llms.txt: HTTP 404 at https://www.bench.co/llms.txt”
   • [probe] “PROBE docs-md: HTTP 404 at https://www.bench.co/how-it-works.md”
   • [probe] “PROBE openapi: all candidate paths 404 (https://www.bench.co/openapi.json, https://www.bench.co/swagger.json, https://www.bench.co/api/opena…”

   proof ↗⚑ flag

2. ai-native userPrevent my data from being used to train AI models

   weight 3 · round drawn
   Puzzle—none0/10

   No evidence in the pack addresses data-training opt-out, privacy policy commitments, or AI model training exclusions for Puzzle's data; the evidence only covers accounting/AI-bookkeeping features.

   ⚑ flag

   Bench—none0/10

   Bench is a bookkeeping/accounting service and no evidence pack item mentions AI training data usage, opt-out controls, or any privacy policy addressing model training; this is an applicable privacy-posture axis for any modern SaaS handling customer data, but no supporting or contradicting evidence exists. Missing for 10: any privacy policy statement, opt-out mechanism, or terms-of-service language regarding AI/ML training on customer data.

   ⚑ flag

3. ai-native userControl data retention and deletion

   weight 2 · round drawn
   Puzzle—none0/10

   No evidence pack items address data retention policies, deletion controls, or data lifecycle management for user/financial data; the evidence covers accounting features, AI Close, exports, and MCP integration but nothing on retention/deletion controls.

   ⚑ flag

   Bench—none0/10

   Bench is a human-plus-software bookkeeping service; no evidence in the pack addresses data retention policies, deletion controls, or export/delete-your-data mechanisms for AI-native users. Probes for docs/llms.txt/openapi all 404, and no privacy/retention page is cited. missing for 10: retention policy documentation, data deletion/export controls, privacy posture disclosures, any AI-specific data handling terms.

   • [probe] “PROBE llms.txt: HTTP 404 at https://www.bench.co/llms.txt”
   • [probe] “PROBE docs-md: HTTP 404 at https://www.bench.co/how-it-works.md”
   • [probe] “PROBE openapi: all candidate paths 404 (https://www.bench.co/openapi.json, https://www.bench.co/swagger.json, https://www.bench.co/api/opena…”

   proof ↗⚑ flag

Ai insights

1. ai-native userAsk an agent "can we afford X?" and get an answer grounded in live ledger data pulled through the API

   weight 3 · round to Puzzle
   Puzzle✓fullclaimed7/10

   Puzzle's MCP integration explicitly connects an AI agent (Claude) to live financial data 'through natural conversation—no exports, no copy-pasting,' and Puzzle's broader API/AI messaging promotes asking plain-English questions like 'Am I ready to raise?' or 'Which months am I actually profitable?' grounded in the ledger. This directly matches the ai-native 'can we afford X' use case backed by live API data. Missing for 10: independent/hands-on verification of accuracy for financial affordability queries, and no explicit example matching the exact 'can we afford X' phrasing.

   • [claimed-docs] “Puzzle's MCP (Model Context Protocol) integration lets you connect your Puzzle account directly to Claude, giving you instant access to your…”
   • [claimed-docs] “lets you connect your Puzzle account directly to Claude, giving you instant access to your financial data through natural conversation — no …”
   • [claimed-docs] “Ask anything about your business, instantly, in plain English. "How's my cash trending?" "Am I ready to raise?" "Which months am I actually …”
   • [claimed-docs] “With Puzzle's Embedded Accounting—powered by our API, your platform gets automated AI bookkeeping, real-time financial insights, and a clean…”
   • [claimed-docs] “Our API was originally architected for deep platform-to-platform integrations.”

   proof ↗⚑ flag

   Bench—none0/10

   No evidence of any public API, MCP server, or AI-agent integration exposing ledger data; probes for llms.txt, docs-md, and OpenAPI all returned 404s, and reporting is described as human-facing downloads/dashboards, not programmatic access. missing for 10: any API/endpoint for ledger data, agent integration, documented query interface for financial questions.

   • [probe] “PROBE llms.txt: HTTP 404 at https://www.bench.co/llms.txt”
   • [probe] “PROBE docs-md: HTTP 404 at https://www.bench.co/how-it-works.md”
   • [probe] “PROBE openapi: all candidate paths 404 (https://www.bench.co/openapi.json, https://www.bench.co/swagger.json, https://www.bench.co/api/opena…”
   • [claimed-docs] “"Real-time reporting. Access or download your updated income statement or balance sheet at all times" — the client-facing surface is reports…”

   proof ↗⚑ flag

Custom reports

1. accountantBuild custom reports filtered and grouped by class, location, or tag dimensions

   weight 2 · round to Puzzle
   Puzzle~partialclaimed5/10

   Puzzle supports Classes with Groups (puzzle-docs-36) and general ledger/financial statement export in cash or accrual basis (puzzle-docs-26/37/45), plus multi-entity management (puzzle-docs-38/44), but there is no direct evidence of a custom report builder that lets accountants filter/group specifically by class, location, or tag dimensions together. missing for 10: explicit custom report builder UI, evidence of 'location' and 'tag' dimensions (only 'class' is documented), and confirmation that reports can be filtered/grouped by these dimensions simultaneously.

   • [claimed-docs] “a user can create an unlimited number of Groups for Classes”
   • [claimed-docs] “you can export a full general ledger and financial statements on either cash or accrual basis in excel, csv, and pdf formats.”
   • [claimed-docs] “you can export a full general ledger and financial statements on either cash or accrual basis in excel, csv, and pdf formats”
   • [claimed-docs] “Yes, you can export a full general ledger and financial statements on either cash or accrual basis in excel, csv, and pdf formats.”
   • [claimed-docs] “Puzzle supports managing multiple companies (entities) under one login.”

   proof ↗⚑ flag

   Bench—none0/10

   Evidence only shows basic reporting (income statement, balance sheet) delivered as static downloads/reports, with no mention of custom report building or filtering/grouping by class, location, or tag dimensions. Bench is positioned as a done-for-you bookkeeping service, not a self-serve reporting tool with dimensional tagging capabilities.

   • [claimed-docs] “Real-time reporting. Access or download your updated income statement or balance sheet at all times”
   • [claimed-docs] “"Real-time reporting. Access or download your updated income statement or balance sheet at all times" — the client-facing surface is reports…”

   proof ↗⚑ flag

Statements

1. founderGenerate an accurate P&L, balance sheet, and cash-flow statement for any period on demand

   weight 3 · round to Bench
   Puzzle✓fullclaimed7/10

   Puzzle explicitly supports generating and exporting full general ledger and financial statements (cash or accrual basis) in Excel/CSV/PDF on demand, plus real-time reporting and startup metrics like ARR/burn/runway, which directly serves the founder's need for on-demand P&L/balance sheet/cash-flow reporting. Missing for 10: explicit naming of 'balance sheet' and 'cash-flow statement' as distinct statement types, and independent/hands-on verification of statement accuracy beyond vendor docs.

   • [claimed-docs] “Get real-time financial reports with one click”
   • [claimed-docs] “you can export a full general ledger and financial statements on either cash or accrual basis in excel, csv, and pdf formats.”
   • [claimed-docs] “you can export a full general ledger and financial statements on either cash or accrual basis in excel, csv, and pdf formats”
   • [claimed-docs] “Yes, you can export a full general ledger and financial statements on either cash or accrual basis in excel, csv, and pdf formats.”
   • [claimed-docs] “Cash **_and_** accrual accounting”
   • [claimed-docs] “Instantly see startup metrics like ARR, burn, & runway”

   proof ↗⚑ flag

   Bench✓fullclaimed8/10

   Bench explicitly markets real-time reporting with income statement (P&L) and balance sheet access/download at all times, backed by automated categorization and human bookkeeper review for accuracy. Cash-flow statement generation is not explicitly named, and there's no independent/hands-on corroboration of accuracy beyond vendor claims. Missing for 10: explicit mention of cash-flow statement as a deliverable report, and third-party or user corroboration of report accuracy.

   • [claimed-docs] “Real-time reporting. Access or download your updated income statement or balance sheet at all times”
   • [claimed-docs] “"Real-time reporting. Access or download your updated income statement or balance sheet at all times" — the client-facing surface is reports…”
   • [claimed-docs] “Accurate transaction categorization, powered by smart automation with instant guidance”
   • [claimed-docs] “"Connect all your financial accounts to automate data entry, speed up your books, reduce errors and save time", with "Accurate transaction c…”
   • [claimed-docs] “Bench is a human-plus-software service, now under new ownership: "Bench simplifies your small business accounting by combining intuitive sof…”

   proof ↗⚑ flag