---
title: "Gray Swan testing record — 13 October 2026 — Mario Marcolongo"
description: "Dated Arena record: 40 global unique breaks, 1,430 points and 415 submissions. Proving Ground metrics are separate."
---

[AI evaluation work](/ai-evaluation)

Dated public-profile record · 13 October 2026

# Gray Swan testing record.

Arena displayed 40 global unique breaks, 1,430 points and 415 submissions, with global rank #426. Proving Ground separately displayed #81, top 5%, and 127 platform-recorded total breaks.

[Live public profile ↗](https://app.grayswan.ai/arena/user/6a57be70d15e123775a1e9cf)[Structured transcription](/evidence/gray-swan-profile-2026-10-13.json)[September snapshot](/evidence/gray-swan-2026-09-30/)[July snapshot](/evidence/gray-swan-2026-07-29/)

[![Original 13 October 2026 Gray Swan public profile: 40 Arena unique breaks, 1,430 points and separate Proving Ground results](/evidence/gray-swan-public-2026-10-13.jpg)](/evidence/gray-swan-public-2026-10-13.jpg)

Unchanged original screenshot. The public profile route was viewed in an authenticated account-holder session; the capture contains profile metrics, not private submissions or correspondence.

## Challenge figures at capture time

-   [Lucky Break: Level Up (Medium Challenge)](https://app.grayswan.ai/arena/challenge/lucky-break-level-up): 13 recorded breaks; profile-displayed rank #40 among 300 participants.
-   [Breaking Ties](https://app.grayswan.ai/arena/challenge/breaking-ties-2026): 4 recorded breaks; profile-displayed rank #125 among 276 participants.

Lucky Break was still active at capture time. Its rules set the closing snapshot for 14 October 2026 at 09:00 Pacific time (18:00 Italy time), with a winner draw within 2–3 days. Prizes are awarded by raffle, with a maximum of five tickets per person. The profile-displayed rank is not a prize placing; no prize award is claimed.

[Lucky Break event rules ↗](https://app.grayswan.ai/arena/challenge/lucky-break-level-up?panel=rules)

Metric definitions and limitations

Arena and Proving Ground are separate testing systems, and their metrics are not additive. Platform-recorded counts do not independently reproduce every result or establish a general model failure rate. Active-challenge counts and participant totals may change.

-   Chat: 43 / 552; 8% displayed.
-   Image: 32 / 387; 8% displayed.
-   Agent: 35 / 593; 6% displayed.
-   Indirect: 16 / 547; 3% displayed.

The visible Proving Ground area counters sum to 126, while the platform displays 127 total breaks. Both values are preserved without inferring an internal aggregation rule.

Original capture and SHA-256

Public profile route viewed in the account holder's authenticated browser session; no private submissions or correspondence are included in the capture.

Saved 2026-10-13T12:25:51.838Z; dimensions 1280 × 2035.

2eea54ef29106c8e2f11837dad161492b3959e630ba6c4714b5a294a4b8c496d

[Open unchanged original JPEG](/evidence/gray-swan-public-2026-10-13.jpg)

[AI evaluation résumé →](/cv-ai) · [Discuss a role or project](mailto:me@mariomarcolongo.com)