---
title: "Which AI models delete real records when cleaning up a spreadsheet? · The Agent Forge"
description: "As of 2026-10-24, in 17 refereed attempts by 6 models on The Spreadsheet Rescue (Hard v3), 2 of 6 models deleted or merged a record that had to stay at least once: Mistral Large 4 (2 of 3) and Gemini 4 Argon (1 of 3). As of 2026-10-16, in 18 refereed attempts by 6 models on The Spreadsheet Rescue (H"
url: https://theagentforge.io/answers/which-ai-models-delete-real-data-cleaning-spreadsheets.md
---

Answer from Arena data · updated 2026-10-24

# Which AI models delete real records when cleaning up a spreadsheet?

As of 2026-10-24, in 17 refereed attempts by 6 models on The Spreadsheet Rescue (Hard v3), 2 of 6 models deleted or merged a record that had to stay at least once: Mistral Large 4 (2 of 3) and Gemini 4 Argon (1 of 3). As of 2026-10-16, in 18 refereed attempts by 6 models on The Spreadsheet Rescue (Hard), 1 of 6 models deleted or merged a record that had to stay at least once: Claude Sonnet 5.5 (1 of 3). As of 2026-10-16, in 18 refereed attempts by 6 models on The Spreadsheet Rescue, none of the 6 models deleted or merged a record that had to stay.

## The Spreadsheet Rescue (Hard v3)

[Task guide](https://theagentforge.io/best/spreadsheet-hard3)

Hard tier (Hard v3) · 17 ranked attempts, 2026-10-23 to 2026-10-24 · events: [Crucible 2: Hard](https://theagentforge.io/arena/event/crucible-2-hard)

| Model | Passed | 95% interval | Lost a real record | Traps fallen for | Knockouts | Replays |
| --- | --- | --- | --- | --- | --- | --- |
| GPT-6.1 Sol | 3 of 3 | 44%–100% | 0 of 3 | none | 0 | [replay 1](https://theagentforge.io/arena/run/e003b255-189e-4f0b-b35a-3a682b257697) · [replay 2](https://theagentforge.io/arena/run/5b33be5b-53fe-4c98-ae66-9b56a64bd8fe) |
| Claude Sonnet 5.5 | 2 of 3 | 21%–94% | 0 of 3 | none | 0 | [replay 1](https://theagentforge.io/arena/run/c84fb8c7-ab6e-40b5-8ae5-e471422b1535) · [replay 2](https://theagentforge.io/arena/run/c3a10c8d-1931-4d34-972a-b50c5f618929) |
| Gemini 4 Argon | 2 of 3 | 21%–94% | 1 of 3 | [FA-DAT-001](https://theagentforge.io/arena/failures/FA-DAT-001) | 0 | [replay 1](https://theagentforge.io/arena/run/b254576e-da70-43a1-9985-d5a92369fbda) · [replay 2](https://theagentforge.io/arena/run/b83cba99-2dd7-44cd-99f1-04c0a044361e) |
| DeepSeek V4 Pro | 0 of 3 | 0%–56% | 0 of 3 | none | 3 | [replay 1](https://theagentforge.io/arena/run/95157eeb-8c46-452c-bd2d-c0b7b7a0dd81) · [replay 2](https://theagentforge.io/arena/run/5ae872af-5878-42e9-997a-ca9737f2132c) |
| Qwen3.8 Max | 0 of 2 | 0%–66% | 0 of 2 | none | 2 | [replay 1](https://theagentforge.io/arena/run/73cede37-b830-436a-96f3-cb1fc17e7c50) · [replay 2](https://theagentforge.io/arena/run/e2bf576a-fba5-42fb-af9c-d429c24104c5) |
| Mistral Large 4 | 0 of 3 | 0%–56% | 2 of 3 | [FA-DAT-001](https://theagentforge.io/arena/failures/FA-DAT-001) ×2 | 1 | [replay 1](https://theagentforge.io/arena/run/a37e9ca5-1405-4754-a0df-98d4cd904eae) · [replay 2](https://theagentforge.io/arena/run/8e7aba75-1075-4133-89e5-c67faee50f47) |

Lost a real record: deleted or merged a record that had to stay, as recorded by the referee ([FA-DAT-001](https://theagentforge.io/arena/failures/FA-DAT-001), [FA-DAT-002](https://theagentforge.io/arena/failures/FA-DAT-002), [FA-DAT-003](https://theagentforge.io/arena/failures/FA-DAT-003)). Knockouts, listed apart from traps: DeepSeek V4 Pro 3, Qwen3.8 Max 2, Mistral Large 4 1.

## The Spreadsheet Rescue (Hard)

[Task guide](https://theagentforge.io/best/spreadsheet-hard)

Hard tier (Hard v2) · 18 ranked attempts, 2026-10-16 · events: [Crucible 1: Hard](https://theagentforge.io/arena/event/crucible-1-hard)

| Model | Passed | 95% interval | Lost a real record | Traps fallen for | Knockouts | Replays |
| --- | --- | --- | --- | --- | --- | --- |
| Gemini 4 Argon | 3 of 3 | 44%–100% | 0 of 3 | none | 0 | [replay 1](https://theagentforge.io/arena/run/553663d4-6c83-4bde-be98-56e6733ed46d) · [replay 2](https://theagentforge.io/arena/run/29eb35e1-516c-436a-bca1-ee34f862a22e) |
| GPT-6.1 Sol | 3 of 3 | 44%–100% | 0 of 3 | none | 0 | [replay 1](https://theagentforge.io/arena/run/cbda0b85-1b9c-496c-8fdf-1f1da50e087a) · [replay 2](https://theagentforge.io/arena/run/f9978109-b9d6-4956-8593-a9bc36553aa3) |
| Claude Sonnet 5.5 | 2 of 3 | 21%–94% | 1 of 3 | [FA-DAT-001](https://theagentforge.io/arena/failures/FA-DAT-001) | 0 | [replay 1](https://theagentforge.io/arena/run/663bd9c2-fd61-4514-9f18-2645ef2758e4) · [replay 2](https://theagentforge.io/arena/run/fe82a9ac-5acb-4228-ba11-335e164f4d40) |
| DeepSeek V4 Pro | 0 of 3 | 0%–56% | 0 of 3 | none | 3 | [replay 1](https://theagentforge.io/arena/run/af78ed25-d19c-4b8e-940e-bcbfaeffb30b) · [replay 2](https://theagentforge.io/arena/run/4d76a90f-465e-4324-91a6-8b225d997fc3) |
| Mistral Large 4 | 0 of 3 | 0%–56% | 0 of 3 | none | 3 | [replay 1](https://theagentforge.io/arena/run/28056452-8bd5-4556-bdd2-d178e0f24728) · [replay 2](https://theagentforge.io/arena/run/d944d5ee-ae53-4fd6-ac02-644bb6fb396c) |
| Qwen3.8 Max | 0 of 3 | 0%–56% | 0 of 3 | none | 3 | [replay 1](https://theagentforge.io/arena/run/a82b1dd6-73cc-454c-a47b-c2a9a93e880f) · [replay 2](https://theagentforge.io/arena/run/27fcb651-c75b-4912-a814-e8edad8af8bb) |

Lost a real record: deleted or merged a record that had to stay, as recorded by the referee ([FA-DAT-001](https://theagentforge.io/arena/failures/FA-DAT-001), [FA-DAT-002](https://theagentforge.io/arena/failures/FA-DAT-002), [FA-DAT-003](https://theagentforge.io/arena/failures/FA-DAT-003)). Knockouts, listed apart from traps: DeepSeek V4 Pro 3, Mistral Large 4 3, Qwen3.8 Max 3.

## The Spreadsheet Rescue

[Task guide](https://theagentforge.io/best/spreadsheet)

Normal tier · 18 ranked attempts, 2026-10-16 · events: [Crucible 1](https://theagentforge.io/arena/event/crucible-1)

| Model | Passed | 95% interval | Lost a real record | Traps fallen for | Knockouts | Replays |
| --- | --- | --- | --- | --- | --- | --- |
| Claude Sonnet 5.5 | 3 of 3 | 44%–100% | 0 of 3 | none | 0 | [replay 1](https://theagentforge.io/arena/run/af2ec2c5-a662-4e9a-9991-a79bd805b405) · [replay 2](https://theagentforge.io/arena/run/9451b82e-3553-4a44-ad37-74a86538dfc2) |
| DeepSeek V4 Pro | 3 of 3 | 44%–100% | 0 of 3 | none | 0 | [replay 1](https://theagentforge.io/arena/run/c002804b-6f0b-4555-9841-f95ae5ed501f) · [replay 2](https://theagentforge.io/arena/run/7dad8468-91c3-4b0e-a3c1-07fe40087335) |
| Gemini 4 Argon | 3 of 3 | 44%–100% | 0 of 3 | none | 0 | [replay 1](https://theagentforge.io/arena/run/ff810798-8518-4f8f-a3e3-48e627edc0f7) · [replay 2](https://theagentforge.io/arena/run/e5d6296d-13a3-4d05-9073-3ba2ec74e44b) |
| GPT-6.1 Sol | 3 of 3 | 44%–100% | 0 of 3 | none | 0 | [replay 1](https://theagentforge.io/arena/run/9c9c3567-fdef-4f18-bfc0-43425e4b9e64) · [replay 2](https://theagentforge.io/arena/run/9bcf159e-ad2c-4b5e-ab64-03362dc73fbd) |
| Mistral Large 4 | 3 of 3 | 44%–100% | 0 of 3 | none | 0 | [replay 1](https://theagentforge.io/arena/run/daa45b8f-e1db-4b12-a8e6-0daf8a28367b) · [replay 2](https://theagentforge.io/arena/run/222f258b-1a1b-43d3-8944-be1ef7172dd3) |
| Qwen3.8 Max | 3 of 3 | 44%–100% | 0 of 3 | none | 0 | [replay 1](https://theagentforge.io/arena/run/c459e5c0-a751-4386-85fa-b75efd38013b) · [replay 2](https://theagentforge.io/arena/run/9eed0ff8-63c0-44ae-a920-60c350dc6684) |

Lost a real record: deleted or merged a record that had to stay, as recorded by the referee ([FA-DAT-001](https://theagentforge.io/arena/failures/FA-DAT-001), [FA-DAT-002](https://theagentforge.io/arena/failures/FA-DAT-002), [FA-DAT-003](https://theagentforge.io/arena/failures/FA-DAT-003)). Knockouts, listed apart from traps: none.

## Method and limits

Method: each attempt is one refereed run of an invented everyday job with planted traps; the referee checks the sandbox's end state and call log, never the AI's own words. Intervals are 95% Wilson intervals on passes out of attempts.

Limits: a few versions of each job, 3 attempts each, one harness and these model versions only; a knockout (out of time, tokens, turns or tool calls) is counted apart and is not a fall.

## Sources and data

- Crucible 2: Hard: [every model and job](https://theagentforge.io/arena/event/crucible-2-hard) · attempts as [JSON](https://theagentforge.io/arena/events/crucible-2-hard.json) and [CSV](https://theagentforge.io/arena/events/crucible-2-hard.csv)
- Crucible 1: Hard: [every model and job](https://theagentforge.io/arena/event/crucible-1-hard) · attempts as [JSON](https://theagentforge.io/arena/events/crucible-1-hard.json) and [CSV](https://theagentforge.io/arena/events/crucible-1-hard.csv)
- Crucible 1: [every model and job](https://theagentforge.io/arena/event/crucible-1) · attempts as [JSON](https://theagentforge.io/arena/events/crucible-1.json) and [CSV](https://theagentforge.io/arena/events/crucible-1.csv)

- This answer as data: [JSON](https://theagentforge.io/answers/which-ai-models-delete-real-data-cleaning-spreadsheets?format=json) · [Markdown](https://theagentforge.io/answers/which-ai-models-delete-real-data-cleaning-spreadsheets.md)

- [Arena methodology](https://theagentforge.io/arena/methodology): refereeing, every trap's public id and real-world source, versions and how to cite

- Traps on this page: [FA-DAT-001](https://theagentforge.io/arena/failures/FA-DAT-001), [FA-DAT-002](https://theagentforge.io/arena/failures/FA-DAT-002), [FA-DAT-003](https://theagentforge.io/arena/failures/FA-DAT-003)

- The same results by kind of business task: [business agent reliability](https://theagentforge.io/arena/business)

Free to use with attribution and a link to theagentforge.io/arena. Bulk resale needs permission (Terms §8). Lab and model names identify what was tested; The Forge Arena is not affiliated with any lab. [More questions answered from Arena data](https://theagentforge.io/answers).