Obelisk validation doctrine¶

The target construct¶

The Obelisk is trying to form broadly capable human beings, not merely high scorers, employable workers, AI operators, polymathic trivia stores, or people who are difficult to surprise.

Its target construct is deliberately multi-dimensional:

• a deeply connected inheritance of human knowledge;
• durable internal models and intellectual independence;
• generative transfer and the ability to enter new fields;
• judgment under uncertainty and value conflict;
• cultivated taste and discriminating quality judgment;
• practical, embodied and technical agency;
• relational competence and care;
• civic, economic and institutional competence;
• responsible use and governance of increasingly capable AI;
• authorship of one's commitments and accountability for delegated action;
• stewardship of culture, institutions, nature and future generations; and
• the capacity for lifelong renewal.

No finite assessment can prove the whole construct. Surprise or adversarial scenarios are only one stress-test of adaptive agency; they are not the curriculum's organizing purpose.

Validation is an argument, not a stamp¶

The Standards for Educational and Psychological Testing treat validity as evidence and theory supporting a particular interpretation and use, not as an intrinsic property of a test. The Obelisk adopts the same discipline.

Therefore we do not describe the whole curriculum as validated. The following are examples of claims to test, not findings established by this release:

• this graph is structurally coherent;
• this branch covers the problem families exposed by a particular external benchmark;
• completers retain foundational knowledge without AI;
• completers transfer models to unseen cases;
• completers learn unfamiliar domains faster;
• completers make better-calibrated decisions and retain authorship when using AI;
• completers perform competently in practical, relational and institutional domains;
• expert judges consistently recognize depth, judgment, taste and ownership in their work; or
• longitudinal completers show broad functional competence and continued formation.

Each claim has different evidence requirements and different falsifiers. See data/obelisk/validation-claims.json.

Why All Souls matters¶

All Souls is not valuable merely because it is prestigious. It is valuable because its Examination Fellowship is a severe human acceptance process for open-ended intellectual performance. Candidates write specialist and general papers, and shortlisted candidates face a viva. Published criteria emphasize depth, analytical power, independence, original-source knowledge, breadth, clarity, responsiveness to argument, awareness of unresolved difficulties and future intellectual potential.

This motivates a proposed criterion proxy for part of intellectual formation: can a learner generate, organize and defend serious thought without an answer key? Mapping a curriculum to published questions does not show that its learners can perform those tasks or meet the examining institution's standards.

It still does not test the whole person. It does not establish health agency, relationships, embodied skill, engineering execution, practical independence, long-run stewardship, or flourishing.

The imported All Souls mappings have not been established as an independent, uncontaminated evaluation. They do not constitute endorsement or external judgment of this curriculum by All Souls. A proposed next experiment would preregister and freeze the graph before testing additional Economics, English Literature, History, Law and Politics papers.

Seven converging evidence families¶

1. Construct and content evidence - deletion tests, branch saturation, life-pillar coverage, global/temporal audits, source stacks and civilizational anchors.
2. External human acceptance - All Souls-style expert judgment of open-ended work; professional standards; external portfolio review; user/community judgment where appropriate.
3. Response-process and ownership evidence - unaided reconstruction, viva, provenance, decision records, live practical demonstrations and human/AI condition contrasts.
4. Unseen transfer and learning evidence - unfamiliar cases, delayed tests, hidden perturbations, learning sprints and new-domain acquisition.
5. Judgment and adaptive-agency evidence - decisions under uncertainty, competing objectives, adversarial AI, incomplete evidence, pressure, reversibility and accountability. Surprise scenarios live here as one method.
6. Real-world functional evidence - projects, craft, fieldwork, household/practical performance, institutional participation, teaching, relationships/care and service to real stakeholders.
7. Longitudinal natural experiment - eventually follow serious completers through real lives. Observe learning, decision calibration, work/craft depth, practical agency, civic/relational functioning, responsible AI use, cultural participation, stewardship and plural self-defined meaning while treating selection effects explicitly.

No single evidence family substitutes for the others.

The Obelisk General Examination¶

The project should maintain a demanding capstone examination modeled on the logic of All Souls rather than its exact format. It combines broad essay judgment, mixed-source synthesis, specialist depth, systems/institution cases, live practical stations, unfamiliar-domain learning, AI stewardship and viva voce.

This is an assessment of learners, not external validation of the curriculum. Its credibility must come from frozen unseen banks, independent examiners, inter-rater study, calibration against external criteria and evidence that performance relates to later real functioning.

The role of adversarial and surprise scenarios¶

Life contains novelty, but novelty is not the central educational construct. An adversarial scenario is useful because it prevents template matching and reveals whether knowledge can be mobilized when conditions change.

The existing surprise-resilience.json battery is therefore retained for backward compatibility but explicitly demoted to a subordinate adaptive-agency stress-test. It should never crowd out whole-work reading, deep disciplinary study, relationships, craft, health, culture, taste, service or long-term formation.

The claims we refuse to fake¶

We should not promise that completing the Obelisk guarantees maximum potential, a good life, success, happiness, wealth, virtue, superior moral character, or immunity to tragedy. Good lives are plural; circumstances matter; luck matters; and many of the most important human outcomes are not reducible to a scalar.

The long-term educational aspiration is:

A serious completer should possess an unusually broad and deeply connected inheritance of human knowledge, practiced competence across the critical pillars of life, strong independent and AI-augmented judgment, cultivated taste, practical and relational agency, and the capacity to continue forming themselves while exercising responsible authorship and stewardship in a high-capability civilization.

Whether the curriculum produces these outcomes requires prospective evidence, including long-term follow-up. The aspiration is not a promise to current learners or a demonstrated result.

Current evidence state¶

The project is an unfinished public preview. See the generated publication status and evidence for the current candidate and inventory.

Structural checks establish record consistency within their tested scope, not scholarly completeness. Reading-route counts include unselected reading areas and do not establish bibliographic quality. Imported prompt mappings remain unverified as independent hold-outs; proposed assessments have not established learner efficacy or broad life outcomes.