Recommender Performance Index v3.1 is live

NSL-1: A RECOMMENDER YOU CAN SEE INSIDE

Most recommenders are a black box with a score coming out. NSL-1 is a foundation model with the lid off: the audience it has learned, the content it has read, and why any one result sits where it does.

Every leading architecture trained on your behaviour, promoted only when it beats the last one and rolled back when it stops. Then boost, suppress or pin whatever the business needs on top.

Deployed across media, commerce, and publishing

  • Streaming
  • Marketplace
  • Publishing
  • Retail media
  • Music
  • Games
  • Classifieds
  • News
  • Podcasts
  • Grocery
  • Fashion
  • Learning

Meet NSL-1

The foundational recommender

Audience map

3 communities
  • Long-form documentary
  • Weeknight comedy
  • Live sport

Meet NSL-1

Understand your audience. Test the best models. Control every recommendation. NSL-1 is where each deployment starts - and behind it, a field of architectures measured against your own behaviour, so every surface runs whichever one wins.

  • A living map of your audience

    Users, content, interests and behavioural clusters in one space you can actually look at. The communities in your data are found rather than defined, so the groups you did not know to look for turn up beside the ones you did.

    Audience map

    3 communities
    • Long-form documentary
    • Weeknight comedy
    • Live sport
  • Interests that change over time

    Follow one person's interests as they move, and tell a settled taste apart from a single unusual session. Fragmented signals become one current picture of what somebody wants now - not an average of everything they have ever done.

    Interests · user u_41902

    12 weeks
    • Established tasteNature documentary
    • This sessionMotorsport · 41 min

    One odd evening does not rewrite a profile

  • Deep understanding of the content

    Multimodal embeddings - TwelveLabs' Marengo, OpenAI, or your own - let ranking read scenes, actions, dialogue, sound, objects and themes rather than titles and tags. A new item can be placed the day it is published, with no interaction history at all.

    Content · ep_4417

    42:18 · published today
    • Scenecoastal cliff · dusk
    • Actionclimbing, rope work
    • Dialogueexpedition planning
    • Soundwind, no score
    • Objectsharness, sea, gull
    EmbeddingsTwelveLabs · MarengoOpenAIYour own

    Rankable before it has a single view

  • The best model for your data

    Collaborative, content-based, sequential, graph and generative architectures, trained on your actual behaviour and measured against each other on the same footing. Towards your outcome - watch time, completion, retention, purchases - not a fixed list of generic signals.

    Candidates · your data

    Optimising watch time
    • Sequentialpromoted · 0.92
    • Graph0.86
    • Two-tower0.81
    • Generative0.77
    • Content-based0.68
    • Collaborative0.61

    No architecture wins every catalogue · this one wins yours

  • Automated testing, promotion and rollback

    A candidate only goes live when it beats the incumbent on the agreed metrics and stays inside your quality, diversity, latency and cost guardrails. Shadow first, then a canary on a slice of traffic, with automatic rollback if the live numbers move the wrong way.

    Release · nsl-1-retail v5

    Day 2 of 3
    • Shadow24h · no traffic served
    • Canary5% of homepage_rail
    • Promoteheld · awaiting day 3
    Relevance+4.8%
    Diversity−0.4%
    p95 latency41 ms
    Cost / 1k+2%

    Rolls back on its own if live performance drops

  • Complete control over what is recommended

    Boost, suppress, exclude, pin or guarantee content, and set how hard an editorial decision pushes against personalisation. Every result traces back to the interests, interactions, similarities and rules that produced it.

    Control · homepage_rail

    4 rules
    • Ifstock = 0exclude
    • Iftag = new-seasonboost ×1.4
    • Ifcreator = in-houseguarantee 2
    • Ifmargin < 8%suppress
    Editorial override0.35 of ranking

    Every result traceable to the signal or rule that moved it

Ranking that works on behalf of catalogues

Can your recommender do this?

Catalogues of any size can fine-tune NSL-1.

CapabilityNSL-1Off-the-shelf engine

Content it can see

Scenes, dialogue, sound and objects read from the item itself, so a new one is rankable before it has a single view.

Model selection

Leading architectures trained on your behaviour and compared like for like, rather than one fixed model for every catalogue.

Safe releases

Shadow, then canary, then promotion - and automatic rollback the moment live performance drops.

Editorial control

Boost, suppress, pin or guarantee at request time, with the strength of the override set by you.

A traceable result

Every position decomposed into the interests, similarities and rules that produced it, with confidence attached.

Adaptive relevance, in the open

Most engines guess.
NSL-1 ranks.

This sandbox is a real feed. A cursor hovers, adds to cart, and favourites items, emitting weak and strong signals that re-rank the grid as they land. Switch between two users to see the same catalogue resolve to two different front pages.

Live sandboxhome-feed
Personalised home
Trending
Wireless Headset
0%
Wireless Headset
4K Action Cam
0%
4K Action Cam
Smart Speaker
0%
Smart Speaker
Noise Buds Pro
0%
Noise Buds Pro
For you
relevance updated
4K Action Cam
0%
4K Action Cam
Fitness Watch
0%
Fitness Watch
Gaming Mouse
0%
Gaming Mouse
Mirrorless Camera
0%
Mirrorless Camera
Noise Buds Pro
0%
Noise Buds Pro
Portable Projector
0%
Portable Projector
Smart Speaker
0%
Smart Speaker
Wireless Headset
0%
Wireless Headset

Operate the whole system in natural language

An MCP server over the same control surface the console uses. An AI operator can inspect analytics and catalogue data, or work with pipelines, rules and training runs directly - no separate, weaker API to fall back on.

Read the docs

Recommender Performance Index · v3.1

Put your catalogue to work

23 systems, scored against a frozen field on the same datasets and the same budget. Every number below is reproducible, and the ones we did not measure ourselves say so.

Find the next one

Eight histories.
One next item each.

Every clue is what somebody did. The answer is what they reached for next, hidden in the grid, and the history is enough to work it out: the rest of a director’s run, the writer behind the last three, the tool the previous purchases were building towards. Mark the first letter and the last to claim one.

0 of 8 found
  • Watched

    Breaking Bad · Better Call Saul (8)

  • Watched

    Arrival · Blade Runner 2049 · Sicario (4)

  • Watched

    Black Mirror · Devs · Mr. Robot (9)

  • Watched

    Planet Earth · Blue Planet II · Frozen Planet (9)

  • Purchased

    Espresso machine · Burr grinder · Milk jug (6)

  • Watched

    The Wire · Generation Kill · Show Me a Hero (5)

  • Read

    Dune · Neuromancer · Snow Crash (8)

  • Purchased

    Cast-iron skillet · Carbon steel chef's knife · Dutch oven (9)

Running at scale across media, commerce, and publishing

Index frozen 23 August 2026 · 23 systems measured