BOOK 1 · VERSION 1.0

Concept map

Who does what at ATLAS, and how data, permissions, technical work and evaluation connect.

The map separates each person’s role from what they know. Its links describe tasks or dependencies, not intimacy or access to private information. ATLAS, Nexo and ARGOS are fictional; the Kaggle challenge is real.

A ROUTE THROUGH THE MAP

From the team to its decisions

Choose a branch to meet the people or follow the challenge concepts. The lines organise the reading; the captions state the documented relationships.

START HEREATLAS Laboratory

Fictional university team working on the visual matching challenge.

01 · PEOPLE AND ROLES

Who does what

02 · CHALLENGE AND METHOD

Matching challenge

ATLAS takes part in the challenge; Kaggle hosts it.

Kaggle · real platform

  1. Matching challenge
  2. Patient-level separation
  3. Authorised data
  4. Provenance and versions
  5. Experiment
  6. Validation
  7. Mean F1 per query
  8. Agents and specifications
  9. Limited resources

03 · THE TEAM AROUND ATLAS

Funding and comparison

Nexo

Funds and reviews ATLAS and ARGOS.

ARGOS

Another team compared with ATLAS.

Team and world

Person · Narrated or established in the cast brief

Leo Martín

Role
Models and experimentation tools
What they deliver or decide
Implements hypotheses, compares models and records the conditions under which each experiment works.
Knowledge
Programming, model comparison and testable specifications; he is still developing his method and can make mistakes.
What they need
He needs traceable data, computing resources and independent evaluation.

Open character profile →

Sources: Cast brief · Responsibility matrix · Chapter 1. Read the full book

Person · Narrated or established in the cast brief

Alicia Valcárcel

Role
Analysis, validation and operational coordination
What they deliver or decide
Turns an operational question into analysis and conclusions bounded by the evidence.
Knowledge
Mathematics and data science; distinguishes measured results from claims the team can support.
What they need
She needs Nadia’s evaluation design and an exact record of each test.

Open character profile →

Sources: Cast brief · Responsibility matrix · Chapter 1. Read the full book

Person · Narrated or established in the cast brief

Mara Quispe Salazar

Role
Data engineering and lifecycle
What they deliver or decide
Prepares repeatable transformations, versions and a deliverable package with clear provenance.
Knowledge
Provenance, preparation, versions and packages that someone else can reproduce.
What they need
She needs permission, an agreed scope and time to rerun and review the work.

Open character profile →

Sources: Cast brief · Responsibility matrix · Chapter 1. Read the full book

Person · Narrated or established in the cast brief

Nico Serrano

Role
Development and integration
What they deliver or decide
Integrates services and interface, then tests the whole path to a usable result.
Knowledge
Services, interface and tests of the complete workflow.
What they need
He needs clear component contracts and access to the test system.

Open character profile →

Sources: Cast brief · Responsibility matrix · Chapter 1. Read the full book

Person · Narrated or established in the cast brief

Jun Park Etxeberria

Role
MLOps, computing and resources
What they deliver or decide
Organises environments, runs, queues and logs; sets the bounds for each agent.
Knowledge
Environments, runs, queues, bounded agents and operational records.
What they need
He needs approved data, priorities and a computing budget.

Open character profile →

Sources: Cast brief · Responsibility matrix · Chapter 1. Read the full book

Person · Narrated or established in the cast brief

Nadia Vidal

Role
Postdoctoral researcher; methodology and evaluation
What they deliver or decide
Designs comparisons, evaluates uncertainty and reviews the assumptions behind each result.
Knowledge
Statistics, comparison, uncertainty and scrutiny of assumptions; she works on ATLAS part time.
What they need
She needs time and the artifacts from the specific run being evaluated.

Open character profile →

Sources: Cast brief · Responsibility matrix · Chapter 1. Read the full book

Person · Narrated or established in the cast brief

Inés Valdés

Role
Scientific and laboratory lead
What they deliver or decide
Sets scope and priorities, arranges permissions and is accountable for funding agreements.
Knowledge
Machine learning, project scope, priorities, people and funding.
What they need
She needs candid information about results, resources and working conditions.

Open character profile →

Sources: Cast brief · Responsibility matrix · Chapter 1. Read the full book

Person · Narrated or established in the cast brief

Álvaro Beltrán

Role
External communication; independent podcast
What they deliver or decide
Prepares interviews and outreach pieces while preserving context and publication boundaries.
Knowledge
Interviews, editing and context for listeners; his laboratory access is limited.
What they need
He needs specific permission and material cleared for publication.

Open character profile →

Sources: Cast brief · Responsibility matrix · Chapter 1. Read the full book

Team or organisation · Narrated or established in the cast brief

ATLAS Laboratory

Fictional university team working on the visual matching challenge.

Sources: Cast brief · Chapter 1. Read the full book

Team or organisation · Narrated or established in the cast brief

Nexo Salud Digital

Fictional company funding and reviewing projects internally; it does not run the international competition.

Sources: Cast brief · Chapter 1. Read the full book

Team or organisation · Narrated or established in the cast brief

ARGOS

Another fictional team compared with ATLAS by Nexo; a score alone proves no wrongdoing.

Sources: Relationships and knowledge · Chapter 1. Read the full book

Real platform · Narrated or established in the cast brief

Kaggle

Platform hosting the real challenge that inspired the story; the teams and results in the narrative are fictional.

Sources: Challenge research · Chapter 1. Read the full book

Challenge and method

Concept · Narrated or established in the cast brief

Matching challenge

For each query image, find gallery samples from the same patient. There may be zero, one or several.

Sources: Challenge research · Chapter 1. Read the full book

Concept · Editorial explanation

Patient-level separation

Patients used for tuning must not reappear in an evaluation meant to measure new cases.

Sources: Challenge research · Chapter 1. Read the full book

Concept · Editorial explanation

Authorised data

Access and use have their own conditions; computing capacity does not grant permission to use data.

Sources: Responsibility matrix · Challenge research. Read the full book

Concept · Editorial explanation

Provenance and versions

Recording origin, transformations and version lets the team identify what ran and repeat a comparison.

Sources: Responsibility matrix · Chapter 1. Read the full book

Concept · Editorial explanation

Experiment

An implemented hypothesis is compared under known conditions; one isolated test does not establish generalisation.

Sources: Responsibility matrix. Read the full book

Concept · Editorial explanation

Validation

Comparison used to choose settings and examine errors; it is not an independent final evaluation.

Sources: Responsibility matrix · Chapter 8 · Chapter 11. Read the full book

Concept · Narrated or established in the cast brief

Mean F1 per query

The challenge metric. Scores are comparable only when measured on the same evaluation under the same rules.

Sources: Challenge research · Chapter 1 · Chapter 9. Read the full book

Concept · Editorial explanation

Agents and specifications

Tools that carry out bounded tasks from instructions and checks; their output still needs review.

Sources: Responsibility matrix · Cast brief. Read the full book

Concept · Editorial explanation

Limited resources

People's time, GPU, memory, storage and API allowance are managed separately.

Sources: Responsibility matrix · Chapter 1. Read the full book

How they connect

  1. Leo Martín is a member of ATLAS Laboratory.Fictional university team working on the visual matching challenge.Narrated or established in the cast brief

    Sources: Cast brief · Chapter 1. Read the full book

  2. Alicia Valcárcel is a member of ATLAS Laboratory.Fictional university team working on the visual matching challenge.Narrated or established in the cast brief

    Sources: Cast brief · Chapter 1. Read the full book

  3. Mara Quispe Salazar is a member of ATLAS Laboratory.Fictional university team working on the visual matching challenge.Narrated or established in the cast brief

    Sources: Cast brief · Chapter 1. Read the full book

  4. Nico Serrano is a member of ATLAS Laboratory.Fictional university team working on the visual matching challenge.Narrated or established in the cast brief

    Sources: Cast brief · Chapter 1. Read the full book

  5. Jun Park Etxeberria is a member of ATLAS Laboratory.Fictional university team working on the visual matching challenge.Narrated or established in the cast brief

    Sources: Cast brief · Chapter 1. Read the full book

  6. Nadia Vidal is a member of ATLAS Laboratory.Fictional university team working on the visual matching challenge.Narrated or established in the cast brief

    Sources: Cast brief · Chapter 1. Read the full book

  7. Inés Valdés leads ATLAS Laboratory.Fictional university team working on the visual matching challenge.Narrated or established in the cast brief

    Sources: Cast brief · Chapter 1. Read the full book

  8. Álvaro Beltrán covers from outside ATLAS Laboratory.Fictional university team working on the visual matching challenge.Narrated or established in the cast brief

    Sources: Cast brief · Chapter 1. Read the full book

  9. Nexo Salud Digital funds and reviews ATLAS Laboratory.Fictional university team working on the visual matching challenge.Narrated or established in the cast brief

    Sources: Cast brief · Chapter 1. Read the full book

  10. Nexo Salud Digital funds and reviews ARGOS.Another fictional team compared with ATLAS by Nexo; a score alone proves no wrongdoing.Narrated or established in the cast brief

    Sources: Relationships and knowledge · Chapter 1. Read the full book

  11. ATLAS Laboratory is compared with ARGOS.Another fictional team compared with ATLAS by Nexo; a score alone proves no wrongdoing.Narrated or established in the cast brief

    Sources: Relationships and knowledge · Chapter 1. Read the full book

  12. Kaggle hosts Matching challenge.For each query image, find gallery samples from the same patient. There may be zero, one or several.Narrated or established in the cast brief

    Sources: Challenge research · Chapter 1. Read the full book

  13. ATLAS Laboratory takes part in Matching challenge.For each query image, find gallery samples from the same patient. There may be zero, one or several.Narrated or established in the cast brief

    Sources: Chapter 1. Read the full book

  14. Leo Martín develops Experiment.An implemented hypothesis is compared under known conditions; one isolated test does not establish generalisation.Narrated or established in the cast brief

    Sources: Responsibility matrix. Read the full book

  15. Leo Martín tests Agents and specifications.Tools that carry out bounded tasks from instructions and checks; their output still needs review.Narrated or established in the cast brief

    Sources: Cast brief · Responsibility matrix. Read the full book

  16. Alicia Valcárcel analyses Validation.Comparison used to choose settings and examine errors; it is not an independent final evaluation.Narrated or established in the cast brief

    Sources: Responsibility matrix. Read the full book

  17. Alicia Valcárcel interprets Mean F1 per query.The challenge metric. Scores are comparable only when measured on the same evaluation under the same rules.Narrated or established in the cast brief

    Sources: Responsibility matrix · Chapter 1. Read the full book

  18. Mara Quispe Salazar prepares and controls Provenance and versions.Recording origin, transformations and version lets the team identify what ran and repeat a comparison.Narrated or established in the cast brief

    Sources: Responsibility matrix. Read the full book

  19. Mara Quispe Salazar prepares and controls Patient-level separation.Patients used for tuning must not reappear in an evaluation meant to measure new cases.Narrated or established in the cast brief

    Sources: Responsibility matrix · Chapter 1. Read the full book

  20. Jun Park Etxeberria manages runs and resources for Limited resources.People's time, GPU, memory, storage and API allowance are managed separately.Narrated or established in the cast brief

    Sources: Responsibility matrix. Read the full book

  21. Jun Park Etxeberria operates Agents and specifications.Tools that carry out bounded tasks from instructions and checks; their output still needs review.Narrated or established in the cast brief

    Sources: Responsibility matrix. Read the full book

  22. Nadia Vidal reviews the method of Experiment.An implemented hypothesis is compared under known conditions; one isolated test does not establish generalisation.Narrated or established in the cast brief

    Sources: Responsibility matrix. Read the full book

  23. Nadia Vidal designs and reviews Validation.Comparison used to choose settings and examine errors; it is not an independent final evaluation.Narrated or established in the cast brief

    Sources: Responsibility matrix. Read the full book

  24. Inés Valdés prioritises Limited resources.People's time, GPU, memory, storage and API allowance are managed separately.Narrated or established in the cast brief

    Sources: Responsibility matrix. Read the full book

  25. Álvaro Beltrán needs for coverage Authorised data.Access and use have their own conditions; computing capacity does not grant permission to use data.Narrated or established in the cast brief

    Sources: Cast brief · Responsibility matrix. Read the full book

  26. Authorised data sets conditions for Provenance and versions.Recording origin, transformations and version lets the team identify what ran and repeat a comparison.Editorial explanation

    Sources: Responsibility matrix. Read the full book

  27. Provenance and versions makes repeatable Experiment.An implemented hypothesis is compared under known conditions; one isolated test does not establish generalisation.Editorial explanation

    Sources: Responsibility matrix. Read the full book

  28. Patient-level separation protects Validation.Comparison used to choose settings and examine errors; it is not an independent final evaluation.Editorial explanation

    Sources: Challenge research · Chapter 1. Read the full book

  29. Leo Martín debates methods with Jun Park Etxeberria.MLOps, computing and resourcesEditorial explanation

    Sources: Relationships and knowledge · Chapter 1. Read the full book

  30. Alicia Valcárcel works with Mara Quispe Salazar.Data engineering and lifecycleEditorial explanation

    Sources: Relationships and knowledge · Chapter 1. Read the full book

  31. Mara Quispe Salazar works with Nico Serrano.Development and integrationEditorial explanation

    Sources: Relationships and knowledge · Chapter 1. Read the full book

  32. Jun Park Etxeberria reviews runs with Nadia Vidal.Postdoctoral researcher; methodology and evaluationEditorial explanation

    Sources: Relationships and knowledge · Responsibility matrix. Read the full book

What each person knows

A projected score does not tell a viewer which version, permissions or causes produced it; the relevant records are needed.

Known by: Determined from records · Do not attribute it to: Álvaro Beltrán

Sources: Relationships and knowledge · Responsibility matrix. Read the full book