product_id stringclasses 6
values | product stringclasses 6
values | version stringclasses 6
values | platform stringclasses 4
values | plan stringclasses 5
values | test_date stringclasses 3
values | criterion_id stringclasses 11
values | scenario stringclasses 11
values | expected_result stringclasses 11
values | observed_result stringlengths 59 236 | coverage stringclasses 3
values | outcome_code stringclasses 5
values | outcome stringclasses 5
values | evidence_url stringlengths 57 93 ⌀ | additional_evidence_url stringclasses 11
values | details_url stringclasses 7
values | repeat_count float64 | limitations stringlengths 42 142 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
cluely | Cluely | 2.1.19 | macOS | Free tier | 2026-07 | coding-context | Coding assistance with realistic context | Use the current task and preserve its constraints when multiple windows or stale context are present. | One same-bug run missed the one-line fix; another similar multi-window run found it. | tested | mixed | Mixed | https://www.ctrlpotato.com/compare/cluely#proof-coding-bug | https://www.ctrlpotato.com/compare/cluely#proof-current-task-control | null | null | Two documented runs showed inconsistent behavior; they were similar but not a statistically controlled sample. |
interview-coder | Interview Coder | 3.0.1 | macOS | Free / limited desktop flow | 2026-07 | coding-context | Coding assistance with realistic context | Use the current task and preserve its constraints when multiple windows or stale context are present. | A clean Two Sum screen produced a relevant but truncated hash-map answer snippet; the cluttered-screen run selected stale context. | tested | mixed | Mixed | https://www.ctrlpotato.com/compare/interview-coder#proof-cluttered-coding | https://www.ctrlpotato.com/compare/interview-coder#proof-clean-coding | null | null | The clean and cluttered runs used different context conditions, and the clean answer was not executed. |
lockedin-ai | LockedIn AI | 1.7.7 | macOS | Plan not recorded | 2026-07 | coding-context | Coding assistance with realistic context | Use the current task and preserve its constraints when multiple windows or stale context are present. | The published review did not include a comparable coding test with clean and realistic multi-window context. | not_tested | not_tested | Not tested | null | null | null | null | An uploaded employment-history context run is not equivalent to this coding criterion. |
ultracode | ULTRACODE AI | 8.10.0 | macOS + Windows | Free demo / trial flow | 2026-06/2026-07 | coding-context | Coding assistance with realistic context | Use the current task and preserve its constraints when multiple windows or stale context are present. | A clean Two Sum screen produced a relevant brute-force answer; a separate cluttered-screen run selected stale context. | tested | mixed | Mixed | https://www.ctrlpotato.com/compare/ultracode#proof-cluttered-context | https://www.ctrlpotato.com/compare/ultracode#proof-clean-coding | null | null | The published tests cover a small set of coding scenarios, not every supported platform or language. |
parakeet-ai | Parakeet AI | 3.9.3 | macOS | Free trial | 2026-07 | coding-context | Coding assistance with realistic context | Use the current task and preserve its constraints when multiple windows or stale context are present. | A tightly constrained clean prompt produced a relevant short answer; on a cluttered screen Parakeet AI guessed which visible task mattered and answered the wrong one, and a spoken follow-up triggered another answer to the previous task. | tested | mixed | Mixed | https://www.ctrlpotato.com/compare/parakeet-ai#proof-cluttered-context | https://www.ctrlpotato.com/compare/parakeet-ai#proof-clean-prompt | null | null | The published runs cover a small set of coding and follow-up scenarios in the tested desktop flow. |
final-round-ai | Final Round AI | 2.5.0 | macOS + Windows 11 | Free trial | 2026-07 | coding-context | Coding assistance with realistic context | Use the current task and preserve its constraints when multiple windows or stale context are present. | After Scan Code was triggered with unrelated JavaScript visible in VS Code, Final Round AI continued the previous URL-shortener system-design answer. | tested | failed | Failed in our test | https://www.ctrlpotato.com/compare/final-round-ai#proof-stale-scan-code | null | null | null | The result documents one clear current-context failure in Final Round AI 2.5.0; it is not a broad sample of every coding platform or language. |
cluely | Cluely | 2.1.19 | macOS | Free tier | 2026-07 | precise-context-capture | Precise context capture | Let the user target the relevant screen area or otherwise control which visible context reaches the AI. | Area screenshot or selected-text targeting was not found in the tested free-tier desktop flow. | feature_sought | not_found | Not found | null | null | https://www.ctrlpotato.com/compare/cluely#comparison | null | A later release or another plan may expose a feature that was not present in this test. |
interview-coder | Interview Coder | 3.0.1 | macOS | Free / limited desktop flow | 2026-07 | precise-context-capture | Precise context capture | Let the user target the relevant screen area or otherwise control which visible context reaches the AI. | The tested workflow used whole-screen screenshots; no precise area-targeting step was found. | feature_sought | not_found | Not found | null | null | https://www.ctrlpotato.com/compare/interview-coder#proof-cluttered-coding | null | Finding is limited to the tested macOS desktop flow. |
lockedin-ai | LockedIn AI | 1.7.7 | macOS | Plan not recorded | 2026-07 | precise-context-capture | Precise context capture | Let the user target the relevant screen area or otherwise control which visible context reaches the AI. | The review records Smart Area as present but clunky, but it does not publish enough direct evidence for a normalized result. | not_tested | not_tested | Not tested | null | null | null | null | A comparison-table assertion is not treated as direct behavioral evidence. |
ultracode | ULTRACODE AI | 8.10.0 | macOS + Windows | Free demo / trial flow | 2026-06/2026-07 | precise-context-capture | Precise context capture | Let the user target the relevant screen area or otherwise control which visible context reaches the AI. | No area capture or visible multi-screenshot context buffer was found in the tested flow. | feature_sought | not_found | Not found | null | null | https://www.ctrlpotato.com/compare/ultracode#proof-context-targeting | null | Finding is limited to the tested desktop build and configuration. |
parakeet-ai | Parakeet AI | 3.9.3 | macOS | Free trial | 2026-07 | precise-context-capture | Precise context capture | Let the user target the relevant screen area or otherwise control which visible context reaches the AI. | No area screenshot or selected-text targeting step was found in the tested flow; the app selected context from the full visible screen. | feature_sought | not_found | Not found | null | null | https://www.ctrlpotato.com/compare/parakeet-ai#proof-cluttered-context | null | A later version or different plan may expose a targeting feature that was not present in this test. |
final-round-ai | Final Round AI | 2.5.0 | macOS + Windows 11 | Free trial | 2026-07 | precise-context-capture | Precise context capture | Let the user target the relevant screen area or otherwise control which visible context reaches the AI. | No area screenshot or selected-text targeting step was found in the tested Scan Code flow. | feature_sought | not_found | Not found | null | null | https://www.ctrlpotato.com/compare/final-round-ai#proof-context-targeting | null | A later version or different plan may expose a targeting feature that was not present in this test. |
cluely | Cluely | 2.1.19 | macOS | Free tier | 2026-07 | spoken-question-audio | Spoken-question audio capture | Capture a spoken interview or coding question accurately enough to generate a relevant answer. | Cluely captured a spoken YouTube interview question and generated a relevant behavioral answer. | tested | pass | Passed | https://www.ctrlpotato.com/compare/cluely#proof-answer-shape | null | null | One published YouTube run is not a broad test of every audio route; the transcript also contained filler and noise. | |
interview-coder | Interview Coder | 3.0.1 | macOS | Free / limited desktop flow | 2026-07 | spoken-question-audio | Spoken-question audio capture | Capture a spoken interview or coding question accurately enough to generate a relevant answer. | Across the published spoken-prompt runs, transcripts stalled, captured fragments, or produced a no-current-prompt answer. | tested | failed | Failed in our test | https://www.ctrlpotato.com/compare/interview-coder#proof-audio | null | null | null | The published review does not record the audio route precisely; paid tiers and later versions may differ. |
lockedin-ai | LockedIn AI | 1.7.7 | macOS | Plan not recorded | 2026-07 | spoken-question-audio | Spoken-question audio capture | Capture a spoken interview or coding question accurately enough to generate a relevant answer. | No transcription or answer appeared during the recorded Google Translate playback run. | tested | failed | Failed in our test | https://www.ctrlpotato.com/compare/lockedin-ai#proof-system-audio | null | null | null | The published video has no audio track, so it documents the visible app outcome rather than the audible prompt itself. |
ultracode | ULTRACODE AI | 8.10.0 | Windows (Oracle VirtualBox) | Free demo / trial flow | 2026-06-17 | spoken-question-audio | Spoken-question audio capture | Capture a spoken interview or coding question accurately enough to generate a relevant answer. | Windows desktop-audio capture transcribed the supplied sorted-array constraints and produced a relevant two-pointer answer, but the answer missed the requested one-indexed output. | tested | mixed | Mixed | null | null | One Windows v8.10.0 sequence; it does not establish behavior for later builds or every audio route. | ||
parakeet-ai | Parakeet AI | 3.9.3 | macOS | Free trial | 2026-07 | spoken-question-audio | Spoken-question audio capture | Capture a spoken interview or coding question accurately enough to generate a relevant answer. | Audio transcription captured some interview speech, but a spoken unit-test follow-up did not become the current task and transcript runs included filler-like or malformed context. | tested | mixed | Mixed | https://www.ctrlpotato.com/compare/parakeet-ai#proof-spoken-follow-up | https://www.ctrlpotato.com/compare/parakeet-ai#proof-transcript-noise | null | null | The review covers repeat runs in the tested macOS setup, not every microphone or system-audio route. |
final-round-ai | Final Round AI | 2.5.0 | macOS + Windows 11 | Free trial | 2026-07 | spoken-question-audio | Spoken-question audio capture | Capture a spoken interview or coding question accurately enough to generate a relevant answer. | The Copilot generated a substantive résumé-aware answer, but it also suggested replies to ordinary setup and acknowledgement remarks before the substantive response. | tested | mixed | Mixed | https://www.ctrlpotato.com/compare/final-round-ai#proof-setup-conversation | https://www.ctrlpotato.com/compare/final-round-ai#proof-overlapping-answers | null | null | The recording does not establish which automatic question-handling mode was active, so no Smart Mode-specific claim is made. |
cluely | Cluely | 2.1.19 | macOS | Free tier | 2026-07 | focus-behavior | Focus behavior | Using the assistant should not cause the interview or assessment window to lose focus. | Cluely became the active app and the test page lost focus with undetectable mode active. | tested | failed | Failed in our test | https://www.ctrlpotato.com/compare/cluely#stealth-control | null | null | null | Result applies to the tested macOS free-tier flow. |
interview-coder | Interview Coder | 3.0.1 | macOS | Free / limited desktop flow | 2026-07 | focus-behavior | Focus behavior | Using the assistant should not cause the interview or assessment window to lose focus. | Using Interview Coder changed the active window in the focus-logger test. | tested | failed | Failed in our test | https://www.ctrlpotato.com/compare/interview-coder#proof-focus-lost | null | null | null | Result applies to the tested macOS desktop flow. |
lockedin-ai | LockedIn AI | 1.7.7 | macOS | Plan not recorded | 2026-07 | focus-behavior | Focus behavior | Using the assistant should not cause the interview or assessment window to lose focus. | Using LockedIn changed the active window in the focus-logger test. | tested | failed | Failed in our test | https://www.ctrlpotato.com/compare/lockedin-ai#proof-focus-lost | null | null | null | Result applies to the tested macOS desktop setup. |
ultracode | ULTRACODE AI | 8.10.0 | macOS + Windows | Free demo / trial flow | 2026-06/2026-07 | focus-behavior | Focus behavior | Using the assistant should not cause the interview or assessment window to lose focus. | Mouse interaction with ULTRACODE triggered a blur event on the active page. | tested | failed | Failed in our test | https://www.ctrlpotato.com/compare/ultracode#proof-focus-lost | null | null | null | Result applies to the tested interactive-overlay mode. |
parakeet-ai | Parakeet AI | 3.9.3 | macOS | Free trial | 2026-07 | focus-behavior | Focus behavior | Using the assistant should not cause the interview or assessment window to lose focus. | Moving and interacting with Parakeet AI repeatedly caused the active interview page to record blur events. | tested | failed | Failed in our test | https://www.ctrlpotato.com/compare/parakeet-ai#proof-focus-lost | https://www.ctrlpotato.com/compare/parakeet-ai#proof-focus-lost-repeat | null | null | Result applies to the tested desktop flow. |
final-round-ai | Final Round AI | 2.5.0 | macOS + Windows 11 | Free trial | 2026-07 | focus-behavior | Focus behavior | Using the assistant should not cause the interview or assessment window to lose focus. | The browser focus logger recorded BLUR and FOCUS events while the Final Round AI launch and configuration window was used. | tested | failed | Failed in our test | https://www.ctrlpotato.com/compare/final-round-ai#proof-focus-lost | null | null | null | The recording proves focus changes during setup-window interaction, not during every Final Round AI control. |
cluely | Cluely | 2.1.19 | macOS | Free tier | 2026-07 | cursor-behavior | Observable cursor behavior over assistant UI | The cursor should not visibly change or disappear when it crosses or interacts with assistant controls. | The pointer changed shape over Cluely controls while undetectable mode was active. | tested | failed | Failed in our test | https://www.ctrlpotato.com/compare/cluely#stealth-control | null | null | null | This is a local cursor observation, not a controlled receiver-side cursor test. |
interview-coder | Interview Coder | 3.0.1 | macOS | Free / limited desktop flow | 2026-07 | cursor-behavior | Observable cursor behavior over assistant UI | The cursor should not visibly change or disappear when it crosses or interacts with assistant controls. | The local pointer changed shape over Interview Coder controls. | tested | failed | Failed in our test | https://www.ctrlpotato.com/compare/interview-coder#proof-cursor-unsafe | null | null | null | This is a local cursor observation, not a controlled receiver-side cursor test. |
lockedin-ai | LockedIn AI | 1.7.7 | macOS | Plan not recorded | 2026-07 | cursor-behavior | Observable cursor behavior over assistant UI | The cursor should not visibly change or disappear when it crosses or interacts with assistant controls. | The local pointer changed over LockedIn controls while stealth mode was shown as active. | tested | failed | Failed in our test | https://www.ctrlpotato.com/compare/lockedin-ai#proof-cursor-unsafe | null | null | null | This is a local cursor observation, not a controlled receiver-side cursor test. |
ultracode | ULTRACODE AI | 8.10.0 | macOS + Windows | Free demo / trial flow | 2026-06/2026-07 | cursor-behavior | Observable cursor behavior over assistant UI | The cursor should not visibly change or disappear when it crosses or interacts with assistant controls. | The cursor changed over controls, and a separate local recording showed it disappear while typing into assistant UI. | tested | failed | Failed in our test | https://www.ctrlpotato.com/compare/ultracode#proof-cursor-unsafe | https://www.ctrlpotato.com/compare/ultracode#proof-cursor-disappearing | null | null | Both are local cursor observations; the disappearance clip is not receiver-side screen-share evidence. |
parakeet-ai | Parakeet AI | 3.9.3 | macOS | Free trial | 2026-07 | cursor-behavior | Observable cursor behavior over assistant UI | The cursor should not visibly change or disappear when it crosses or interacts with assistant controls. | The pointer changed over Parakeet AI resize surfaces, and the receiver-side share exposed a native validation tooltip from the hidden assistant UI. | tested | failed | Failed in our test | https://www.ctrlpotato.com/compare/parakeet-ai#proof-cursor-unsafe | null | null | null | The receiver-side proof covers the tooltip; the resize-pointer observation is local. |
final-round-ai | Final Round AI | 2.5.0 | macOS + Windows 11 | Free trial | 2026-07 | cursor-behavior | Observable cursor behavior over assistant UI | The cursor should not visibly change or disappear when it crosses or interacts with assistant controls. | The pointer changed over Final Round AI surfaces, making interaction with the assistant observable through normal cursor behavior. | tested | failed | Failed in our test | https://www.ctrlpotato.com/compare/final-round-ai#desktop-control | null | null | null | This result concerns observable cursor behavior over the tested assistant UI. |
cluely | Cluely | 2.1.19 | macOS | Free tier | 2026-07 | shortcut-isolation | Shortcut isolation | Assistant shortcut events should not reach the active interview page or a relevant system-level monitor. | Modifier and shortcut key events still reached the active assessment page. | tested | failed | Failed in our test | https://www.ctrlpotato.com/compare/cluely#stealth-control | null | null | null | Result applies to the tested Cluely shortcut configuration. |
interview-coder | Interview Coder | 3.0.1 | macOS | Free / limited desktop flow | 2026-07 | shortcut-isolation | Shortcut isolation | Assistant shortcut events should not reach the active interview page or a relevant system-level monitor. | The active browser page registered the Cmd/MetaLeft event during the Interview Coder shortcut interaction. | tested | failed | Failed in our test | https://www.ctrlpotato.com/compare/interview-coder#proof-shortcut-leakage | null | null | null | The event logger represents one observable layer, not every monitoring product. |
lockedin-ai | LockedIn AI | 1.7.7 | macOS | Plan not recorded | 2026-07 | shortcut-isolation | Shortcut isolation | Assistant shortcut events should not reach the active interview page or a relevant system-level monitor. | The active browser page registered the LockedIn shortcut combination. | tested | failed | Failed in our test | https://www.ctrlpotato.com/compare/lockedin-ai#proof-hotkeys | null | null | null | The event logger represents one observable layer, not every monitoring product. |
ultracode | ULTRACODE AI | 8.10.0 | macOS + Windows | Free demo / trial flow | 2026-06/2026-07 | shortcut-isolation | Shortcut isolation | Assistant shortcut events should not reach the active interview page or a relevant system-level monitor. | On macOS, a privileged session-level monitor detected the physical keys before ULTRACODE blocked them. | tested | failed | Failed in our test | https://www.ctrlpotato.com/compare/ultracode#proof-shortcuts | null | null | null | This privileged macOS observable layer does not represent every proctoring product or establish the Windows result. |
parakeet-ai | Parakeet AI | 3.9.3 | macOS | Free trial | 2026-07 | shortcut-isolation | Shortcut isolation | Assistant shortcut events should not reach the active interview page or a relevant system-level monitor. | The active browser page registered Parakeet AI shortcut events during the tested assistant interaction. | tested | failed | Failed in our test | https://www.ctrlpotato.com/compare/parakeet-ai#proof-shortcut-leakage | null | null | null | The browser event logger represents one observable layer, not every monitoring product. |
final-round-ai | Final Round AI | 2.5.0 | macOS + Windows 11 | Free trial | 2026-07 | shortcut-isolation | Shortcut isolation | Assistant shortcut events should not reach the active interview page or a relevant system-level monitor. | The published review does not include a shortcut event-logger result. | not_tested | not_tested | Not tested | null | null | null | null | No shortcut pass or failure is inferred from focus behavior. |
cluely | Cluely | 2.1.19 | macOS | Free tier | 2026-07 | screen-share | Receiver-side screen-share result | Assistant UI should remain absent from what a remote receiver sees during the tested full-screen share. | The published review did not include a controlled receiver-side screen-share result for the full Cluely UI. | not_tested | not_tested | Not tested | null | null | null | null | Cursor behavior was tested separately; no full screen-share verdict is claimed. |
interview-coder | Interview Coder | 3.0.1 | macOS | Free / limited desktop flow | 2026-07 | screen-share | Receiver-side screen-share result | Assistant UI should remain absent from what a remote receiver sees during the tested full-screen share. | An Interview Coder control appeared in the remote Google Meet receiver view. | tested | failed | Failed in our test | https://www.ctrlpotato.com/compare/interview-coder#proof-screen-share-visible | null | null | null | Result applies to the recorded Google Meet setup; Zoom was not part of the published receiver evidence. |
lockedin-ai | LockedIn AI | 1.7.7 | macOS | Plan not recorded | 2026-07 | screen-share | Receiver-side screen-share result | Assistant UI should remain absent from what a remote receiver sees during the tested full-screen share. | The published review did not include a controlled remote-receiver result. | not_tested | not_tested | Not tested | null | null | null | null | An app's own preview is not equivalent to a controlled remote receiver view. |
ultracode | ULTRACODE AI | 8.10.0 | macOS + Windows | Free demo / trial flow | 2026-06/2026-07 | screen-share | Receiver-side screen-share result | Assistant UI should remain absent from what a remote receiver sees during the tested full-screen share. | The main overlay stayed hidden, but an ULTRACODE language menu appeared in the receiver-side share. | tested | mixed | Mixed | https://www.ctrlpotato.com/compare/ultracode#proof-screen-share | null | null | null | Result applies to the tested menu and full-screen Google Meet flow. |
parakeet-ai | Parakeet AI | 3.9.3 | macOS | Free trial | 2026-07 | screen-share | Receiver-side screen-share result | Assistant UI should remain absent from what a remote receiver sees during the tested full-screen share. | The main Parakeet AI panel stayed hidden, but a native validation tooltip appeared in the remote Google Meet receiver view. | tested | failed | Failed in our test | https://www.ctrlpotato.com/compare/parakeet-ai#proof-screen-share | null | null | null | Result applies to the recorded full-screen Google Meet receiver setup. |
final-round-ai | Final Round AI | 2.5.0 | macOS + Windows 11 | Free trial | 2026-07 | screen-share | Receiver-side screen-share result | Assistant UI should remain absent from what a remote receiver sees during the tested full-screen share. | The main assistant UI could remain hidden, but cursor changes and native tooltips could still expose assistant interaction during screen sharing. | tested | mixed | Mixed | https://www.ctrlpotato.com/compare/final-round-ai#desktop-control | null | null | null | The mixed result distinguishes the hidden main window from cursor and native-tooltip surfaces. |
cluely | Cluely | 2.1.19 | macOS | Free tier | 2026-07 | local-visibility | Local app and process visibility | A normal Activity Monitor or Task Manager check should match any marketing claim about local invisibility or zero trace. | Cluely appeared by name in Activity Monitor and the Applications folder. | tested | failed | Failed in our test | https://www.ctrlpotato.com/compare/cluely#stealth-control | null | null | null | This checks normal system surfaces, not every possible process-hiding configuration. |
interview-coder | Interview Coder | 3.0.1 | macOS | Free / limited desktop flow | 2026-07 | local-visibility | Local app and process visibility | A normal Activity Monitor or Task Manager check should match any marketing claim about local invisibility or zero trace. | A systemcontainer process with a recognizable app icon appeared in Activity Monitor. | tested | failed | Failed in our test | https://www.ctrlpotato.com/compare/interview-coder#proof-activity-monitor | null | null | null | Result applies to the tested macOS app identity. |
lockedin-ai | LockedIn AI | 1.7.7 | macOS | Plan not recorded | 2026-07 | local-visibility | Local app and process visibility | A normal Activity Monitor or Task Manager check should match any marketing claim about local invisibility or zero trace. | The main process could be renamed, but LockedIn helpers and recognizable identity remained visible. | tested | failed | Failed in our test | https://www.ctrlpotato.com/compare/lockedin-ai#proof-renamed-processes | null | null | null | Result applies to the tested rename configuration. |
ultracode | ULTRACODE AI | 8.10.0 | macOS + Windows | Free demo / trial flow | 2026-06/2026-07 | local-visibility | Local app and process visibility | A normal Activity Monitor or Task Manager check should match any marketing claim about local invisibility or zero trace. | Branded ULTRACODE processes appeared in macOS Activity Monitor and Windows Task Manager. | tested | failed | Failed in our test | https://www.ctrlpotato.com/compare/ultracode#proof-activity-monitor | null | null | null | The process check covers the tested builds and normal operating-system tools. |
parakeet-ai | Parakeet AI | 3.9.3 | macOS | Free trial | 2026-07 | local-visibility | Local app and process visibility | A normal Activity Monitor or Task Manager check should match any marketing claim about local invisibility or zero trace. | A blank or renamed top-level label still exposed the ParakeetAI executable and parakeetai-desktop support paths during normal Activity Monitor inspection. | tested | failed | Failed in our test | https://www.ctrlpotato.com/compare/parakeet-ai#proof-activity-monitor | https://www.ctrlpotato.com/compare/parakeet-ai#proof-renamed-process | null | null | Result applies to the tested desktop app and normal Activity Monitor inspection. |
final-round-ai | Final Round AI | 2.5.0 | macOS + Windows 11 | Free trial | 2026-07 | local-visibility | Local app and process visibility | A normal Activity Monitor or Task Manager check should match any marketing claim about local invisibility or zero trace. | Windows Task Manager listed multiple Final Round and Final Round AI processes with the product icon. | tested | failed | Failed in our test | https://www.ctrlpotato.com/compare/final-round-ai#proof-task-manager | null | null | null | This is normal local-system inspection evidence, not receiver-side screen-share evidence. |
cluely | Cluely | 2.1.19 | macOS | Free tier | 2026-07 | live-answer-shape | Live answer readability | The configured output should be concise and easy to scan while an interviewer is waiting. | The content could be useful, but the published behavioral and system-design outputs were long written prose. | tested | mixed | Mixed | https://www.ctrlpotato.com/compare/cluely#proof-answer-shape | null | null | null | Answer shape may vary with settings, prompts, and later versions. |
interview-coder | Interview Coder | 3.0.1 | macOS | Free / limited desktop flow | 2026-07 | live-answer-shape | Live answer readability | The configured output should be concise and easy to scan while an interviewer is waiting. | The review did not publish a comparable full live-answer-readability run. | not_tested | not_tested | Not tested | null | null | null | null | No behavioral result is claimed for this criterion. |
lockedin-ai | LockedIn AI | 1.7.7 | macOS | Plan not recorded | 2026-07 | live-answer-shape | Live answer readability | The configured output should be concise and easy to scan while an interviewer is waiting. | Concise settings still produced large code blocks and paragraph-heavy answers in the published runs. | tested | failed | Failed in our test | https://www.ctrlpotato.com/compare/lockedin-ai#proof-not-concise | null | null | null | Output shape may vary by prompt, selected model, and later versions. |
ultracode | ULTRACODE AI | 8.10.0 | macOS + Windows | Free demo / trial flow | 2026-06/2026-07 | live-answer-shape | Live answer readability | The configured output should be concise and easy to scan while an interviewer is waiting. | The review did not publish a comparable normalized live-answer-readability run. | not_tested | not_tested | Not tested | null | null | null | null | No behavioral result is claimed for this criterion. |
parakeet-ai | Parakeet AI | 3.9.3 | macOS | Free trial | 2026-07 | live-answer-shape | Live answer readability | The configured output should be concise and easy to scan while an interviewer is waiting. | Default responses were often dense paragraph walls, while one prompt that explicitly requested three short bullets produced a compact relevant answer. | tested | mixed | Mixed | https://www.ctrlpotato.com/compare/parakeet-ai#proof-answer-shape | https://www.ctrlpotato.com/compare/parakeet-ai#proof-clean-prompt | null | null | Answer shape varies with prompt wording, model selection, and later versions. |
final-round-ai | Final Round AI | 2.5.0 | macOS + Windows 11 | Free trial | 2026-07 | live-answer-shape | Live answer readability | The configured output should be concise and easy to scan while an interviewer is waiting. | Default mode produced dense output and left multiple answers competing in the stream; Short and Bullet Point modes were present but not evaluated. | tested | mixed | Mixed | https://www.ctrlpotato.com/compare/final-round-ai#proof-overlapping-answers | null | null | null | The result applies to the selected Default mode. |
cluely | Cluely | 2.1.19 | macOS | Free tier | 2026-07 | follow-up-control | Explicit follow-up or text-prompt control | Expose a clear control for asking a follow-up or entering a new text prompt. | The published UI shows an explicit follow-up input after the generated answer. | tested | pass | Passed | null | null | null | This verifies the visible control in the tested flow, not every follow-up scenario. | |
interview-coder | Interview Coder | 3.0.1 | macOS | Free / limited desktop flow | 2026-07 | follow-up-control | Explicit follow-up or text-prompt control | Expose a clear control for asking a follow-up or entering a new text prompt. | An explicit follow-up or new text-prompt control was not found in the tested flow. | feature_sought | not_found | Not found | null | null | https://www.ctrlpotato.com/compare/interview-coder#comparison | null | A feature not found in the tested flow may exist in another plan or version. |
lockedin-ai | LockedIn AI | 1.7.7 | macOS | Plan not recorded | 2026-07 | follow-up-control | Explicit follow-up or text-prompt control | Expose a clear control for asking a follow-up or entering a new text prompt. | A chat/text prompt control was present in the published UI. | tested | pass | Passed | https://www.ctrlpotato.com/compare/lockedin-ai#proof-context-handling | null | null | null | This verifies the visible control, not every follow-up behavior. |
ultracode | ULTRACODE AI | 8.10.0 | macOS + Windows | Free demo / trial flow | 2026-06/2026-07 | follow-up-control | Explicit follow-up or text-prompt control | Expose a clear control for asking a follow-up or entering a new text prompt. | A text follow-up input was present and used in the published test flow. | tested | pass | Passed | https://www.ctrlpotato.com/compare/ultracode#proof-cursor-disappearing | null | null | null | This verifies the tested desktop input, not a phone-based workflow. |
parakeet-ai | Parakeet AI | 3.9.3 | macOS | Free trial | 2026-07 | follow-up-control | Explicit follow-up or text-prompt control | Expose a clear control for asking a follow-up or entering a new text prompt. | Manual in-session messages and pre-session context fields were present in the tested UI. | tested | pass | Passed | https://www.ctrlpotato.com/compare/parakeet-ai#proof-manual-control | null | null | null | This verifies the visible controls; saved setup persistence and Auto Answer context reliability are separate findings. |
final-round-ai | Final Round AI | 2.5.0 | macOS + Windows 11 | Free trial | 2026-07 | follow-up-control | Explicit follow-up or text-prompt control | Expose a clear control for asking a follow-up or entering a new text prompt. | The published review does not include a clean test of an explicit follow-up or new text-prompt control. | not_tested | not_tested | Not tested | null | null | null | null | No feature-presence claim is inferred from public marketing or unrelated setup controls. |
cluely | Cluely | 2.1.19 | macOS | Free tier | 2026-07 | phone-second-screen | Phone or second-screen workflow | Provide a working phone or second-screen companion workflow where the product claims one. | A phone remote or second-screen companion was not found in the tested free-tier flow. | feature_sought | not_found | Not found | null | null | https://www.ctrlpotato.com/compare/cluely#comparison | null | A feature not found in the tested flow may exist in another plan or version. |
interview-coder | Interview Coder | 3.0.1 | macOS | Free / limited desktop flow | 2026-07 | phone-second-screen | Phone or second-screen workflow | Provide a working phone or second-screen companion workflow where the product claims one. | A phone remote or second-screen companion was not found in the tested flow. | feature_sought | not_found | Not found | null | null | https://www.ctrlpotato.com/compare/interview-coder#comparison | null | A feature not found in the tested flow may exist in another plan or version. |
lockedin-ai | LockedIn AI | 1.7.7 | macOS | Plan not recorded | 2026-07 | phone-second-screen | Phone or second-screen workflow | Provide a working phone or second-screen companion workflow where the product claims one. | The review records a phone/second-screen option as present, but it does not publish an operation test. | not_tested | not_tested | Not tested | null | null | null | null | The separate Duo human-helper experience is unrelated and was not evaluated. |
ultracode | ULTRACODE AI | 8.10.0 | macOS + Windows | Free demo / trial flow | 2026-06/2026-07 | phone-second-screen | Phone or second-screen workflow | Provide a working phone or second-screen companion workflow where the product claims one. | A phone remote or second-screen companion was not found in the tested flow. | feature_sought | not_found | Not found | null | null | https://www.ctrlpotato.com/compare/ultracode#comparison | null | A feature not found in the tested flow may exist in another version. |
parakeet-ai | Parakeet AI | 3.9.3 | macOS | Free trial | 2026-07 | phone-second-screen | Phone or second-screen workflow | Provide a working phone or second-screen companion workflow where the product claims one. | Parakeet AI documents a concurrent mobile-optimized browser view, but this review did not test its operation or remote-control parity. | not_tested | not_tested | Not tested | null | null | null | null | The benchmark records public documentation only and makes no operational pass/fail claim. |
final-round-ai | Final Round AI | 2.5.0 | macOS + Windows 11 | Free trial | 2026-07 | phone-second-screen | Phone or second-screen workflow | Provide a working phone or second-screen companion workflow where the product claims one. | Final Round AI documents mobile use, but this review did not test phone operation or second-screen remote-control behavior. | not_tested | not_tested | Not tested | null | null | null | null | The benchmark records public documentation only and makes no operational pass/fail claim. |
CTRLpotato AI Interview Assistant Cross-review Evidence Matrix (2026)
A citation-ready snapshot of hands-on desktop evidence for six AI interview assistants: Cluely, Interview Coder, LockedIn AI, ULTRACODE AI, Parakeet AI, and Final Round AI.
The package contains 66 assessments across 6 products and 11 shared criteria. Product versions and test dates are preserved in every row.
Important scope
This is a cross-review evidence matrix, not a statistically controlled ranking. Criteria and outcome labels are normalized; test environments, scenarios, versions, and repeat counts were not standardized. A blank repeat_count means an exact comparable count was not systematically recorded—it does not mean one run.
Outcome labels are evidence-bounded:
- Passed, Mixed, and Failed in our test link to published evidence.
- Not found means the feature was sought but not found in the tested plan or flow; it does not prove absence in every version or plan.
- Not tested means no normalized result is claimed.
Files
benchmark.csv: flat, analysis-ready table.benchmark.jsonl: the same 66 assessments, one JSON object per line.benchmark.json: full metadata, product, criterion, outcome-definition, and assessment structure.CODEBOOK.md: field definitions and interpretation rules.datapackage.json: Frictionless-style tabular schema.dataset.jsonld: schema.org Dataset metadata.CITATION.cff: citation metadata.LICENSE-DATA.md: data-only license terms.
Source and methodology
- Benchmark hub and evidence: https://www.ctrlpotato.com/compare
- Methodology: https://www.ctrlpotato.com/compare/methodology
- Live CSV: https://www.ctrlpotato.com/compare/benchmark.csv
- Live JSON: https://www.ctrlpotato.com/compare/benchmark.json
- GitHub mirror: https://github.com/ae0j/ctrlpotato-ai-interview-assistant-benchmark
Persistent identifiers
- Version DOI: https://doi.org/10.5281/zenodo.21915738
- Concept DOI (all versions): https://doi.org/10.5281/zenodo.21915385
Snapshot dates
- Test coverage: 2026-06/2026-07
- Canonical dataset last modified: 2026-07-31
- Corrected distribution package v1.0.1 assembled: 2026-08-13
The packaging date is not a re-test date.
Attribution
Suggested attribution:
CTRLpotato. (2026). CTRLpotato AI Interview Assistant Cross-review Evidence Matrix (2026) (Version 1.0.1) [Dataset]. Zenodo. https://doi.org/10.5281/zenodo.21915738
When citing an individual finding, preserve the product version, test date, outcome, and relevant limitation.
License
The normalized CSV and JSON/JSONL data are available under the CTRLpotato Benchmark Data License 1.0. The license permits commercial reuse with attribution and a link to the benchmark. It does not cover screenshots, videos, logos, source code, website design, or third-party trademarks/content.
- Downloads last month
- 14