What changed with Qloo?

Same shelf. Same tastes. Same constraints.

Quiet evenings, different tastes

Two selection policies using the same shelf and tastes. The Qloo agent receives ranking tools; the catalog-only model receives a different context and policy. This is not a matched model comparison.

Choices

film

Changed

With Qloo

Amélie

Without Qloo

Before Sunrise

book

Changed

With Qloo

The Ocean at the End of the Lane

Without Qloo

Invisible Cities

artist

Changed

With Qloo

BADBADNOTGOOD

Without Qloo

Nujabes

Evidence

Amélie

The Ocean at the End of the Lane

BADBADNOTGOOD

Technical details

Source: live · Model: claude-haiku-5-5 · Run: 2026-10-20T03:39:47.163Z

Metadata: 3092943f0a3e2ab2b78e4b39dd49b3b31a481347e29d6c2cce0c0db487cd3a17 · Configuration: 5945bad175c8ff1ec247a68bc85c13d4097bdfac9b75767a40e467fcba312bf7

model: claude-haiku-5-5
baseline: catalog-only model
steps: 4
totalInputTokens: 32000
totalOutputTokens: 4096
inputTokens: 8000
outputTokens: 1024
deadlineMs: 45000
reservationUsd: 0.02
maxRetries: 0

With Qloo

movie: B7A0731F-EC92-496A-9C40-47D9770489B6
book: 4A176261-3FB2-4479-BA7D-A34DEC79AECA
artist: 5DE0ACFB-A7A9-4CE3-B83E-4B1F41D48D55

5387 ms · Reserved $0.02 · 2026-10-20T03:39:43.066Z

Model usage

Step 1: 1453 input tokens · 43 output tokens · input bound 3997
Step 2: 1886 input tokens · 380 output tokens · input bound 4893
Step 3: 1886 input tokens · 319 output tokens · input bound 4893

Without Qloo

movie: EDA54D37-08C7-457D-83F8-1BCE21FEDC63
book: 1FEAADCD-E6C3-456D-8FFA-7C1DFF5A4189
artist: E11125AF-936B-4935-B607-F2DA2E6A6B92

4092 ms · Reserved $0.02 · 2026-10-20T03:39:47.161Z

Model usage

Step 1: 2431 input tokens · 173 output tokens · input bound 5046
Step 2: 2443 input tokens · 345 output tokens · input bound 5266

Signals: Spirited Away, Piranesi, Bon Iver

Exclusions: 9B33A620-68CA-49B3-BFCF-F09CCF8AD595, 55908017-4CFD-4688-9A7E-7B1FF04B171D, E51DB8F2-7A93-4746-A283-E55B5216FDE1

Locks: None

Actual billing and recipient preference have not been measured.

Different choices do not prove better choices.

All cases

Opening comparison…

Quiet evenings, different tastes

Two selection policies using the same shelf and tastes. The Qloo agent receives ranking tools; the catalog-only model receives a different context and policy. This is not a matched model comparison.

Choices

film

Changed

With Qloo

Amélie

Without Qloo

Before Sunrise

book

Changed

With Qloo

The Ocean at the End of the Lane

Without Qloo

Invisible Cities

artist

Changed

With Qloo

BADBADNOTGOOD

Without Qloo

Nujabes

Evidence

Amélie

The Ocean at the End of the Lane

BADBADNOTGOOD

Technical details

Source: live · Model: claude-haiku-5-5 · Run: 2026-10-20T03:39:47.163Z

Metadata: 3092943f0a3e2ab2b78e4b39dd49b3b31a481347e29d6c2cce0c0db487cd3a17 · Configuration: 5945bad175c8ff1ec247a68bc85c13d4097bdfac9b75767a40e467fcba312bf7

model: claude-haiku-5-5
baseline: catalog-only model
steps: 4
totalInputTokens: 32000
totalOutputTokens: 4096
inputTokens: 8000
outputTokens: 1024
deadlineMs: 45000
reservationUsd: 0.02
maxRetries: 0

With Qloo

movie: B7A0731F-EC92-496A-9C40-47D9770489B6
book: 4A176261-3FB2-4479-BA7D-A34DEC79AECA
artist: 5DE0ACFB-A7A9-4CE3-B83E-4B1F41D48D55

5387 ms · Reserved $0.02 · 2026-10-20T03:39:43.066Z

Model usage

Step 1: 1453 input tokens · 43 output tokens · input bound 3997
Step 2: 1886 input tokens · 380 output tokens · input bound 4893
Step 3: 1886 input tokens · 319 output tokens · input bound 4893

Without Qloo

movie: EDA54D37-08C7-457D-83F8-1BCE21FEDC63
book: 1FEAADCD-E6C3-456D-8FFA-7C1DFF5A4189
artist: E11125AF-936B-4935-B607-F2DA2E6A6B92

4092 ms · Reserved $0.02 · 2026-10-20T03:39:47.161Z

Model usage

Step 1: 2431 input tokens · 173 output tokens · input bound 5046
Step 2: 2443 input tokens · 345 output tokens · input bound 5266

Signals: Spirited Away, Piranesi, Bon Iver

Exclusions: 9B33A620-68CA-49B3-BFCF-F09CCF8AD595, 55908017-4CFD-4688-9A7E-7B1FF04B171D, E51DB8F2-7A93-4746-A283-E55B5216FDE1

Locks: None

Actual billing and recipient preference have not been measured.

Different choices do not prove better choices.