Policy Decisions Ledger

How this ledger is read

Each entry has an ID (P-01, P-02, etc.).

IDs are never reused once assigned, even if an entry is retired and replaced by a successor entry.

Each entry names: the decision, the scope it binds across, the rationale (in layers), the canonical sources that document the underlying analysis, the conditions under which the entry would be reopened, and the operational consequence.

Status values are LOCKED (in force), UNDER REVIEW (a review event has fired and reopening is being assessed), or RETIRED (no longer in force, with a pointer to the successor entry where applicable).

The ledger is updated on a six-month cadence and within thirty days of any triggering event — a new use-case request, a regulatory change, a mission-constraint re-litigation in writing.

P-01 — B2B selection, hiring, and personnel use cases excluded

Status: LOCKED.

Decision dated 2026-05-19; terms-of-service language fixed 2026-05-21; ledger entry filed 2026-05-22.

The decision

Metryval excludes from its product scope any use case in which the instrument’s output is consumed as input to a hiring, promotion, termination, performance-review, or similar personnel-selection decision about the respondent.

The exclusion covers individual selection (single-candidate evaluation), comparative selection (ranking candidates), and team-formation selection (composition optimization).

It binds across every current and planned surface: the consumer Lite and Blueprint, the Couples Pack, the Connected tier, the Founding 500, the Phase-2 institutional bot (Discord and Slack), the Personality OS for AI agents API, and any B2B contract.

What this exclusion permits

Voluntary self-administration by individuals.

Coaching and therapy contexts where the respondent retains full agency over what to disclose, with no downstream gate consequential to a decision someone else is making.

B2B substrate buyers in therapy, wellness, coaching, and research per the impact-gates framework.

Agent-to-agent personality modeling in multi-agent AI systems, where the parties measured are software agents and no human-selection downstream gate exists.

The rationale, in three layers

Layer 1 — regulatory. The European Union AI Act, Annex III, classifies AI systems used for employment, worker management, and access to self-employment as high-risk.

Systems in that class require formal conformity assessment, transparency obligations, human oversight, technical documentation, post-market monitoring, and a documented risk- management system.

Metryval’s architecture, governance, and bootstrapping resources are not built for that compliance burden, nor will they be in any current roadmap horizon.

The terms of service exclude the use case before the bot ships.

Layer 2 — measurement validity. The conditions under which personality measurement remains scientifically defensible are voluntary self-report with the respondent retaining disclosure agency, the absence of a consequential downstream gate that respondents have incentive to manage impressions against, and reference-centroid framing rather than categorical assignment.

Selection contexts invert all three.

The gate is the entire purpose of the instance.

Respondents have direct incentive to manage impressions.

Comparative decisions push toward categorical assignment because categories are what comparison runs on.

The instrument then degrades into a proxy for the decision someone else is making about the person who took it.

Population-level validity scales mitigate the effect at the cohort level but do not restore individual-respondent defensibility.

Layer 3 — mission alignment. Metryval has committed publicly to refusing adversarial-hiring use cases as a structural posture, not a Y1-conditional one.

The instrument exists to serve the person experiencing personality variance, not the institution selecting against it.

This layer is independent of the other two and would hold even in the absence of the regulatory and measurement-validity constraints.

Review conditions

The entry is reopened only if all three of the following are true simultaneously, each documented in writing: the EU AI Act Annex III scope narrows such that voluntary self-report personality instruments are formally exempted; a peer-reviewed empirical finding demonstrates that personality measurement retains its standard validity properties under consequential downstream-gate conditions; and Metryval’s mission constraints are re-litigated in writing per the impact-gates pressure-point trigger framework.

Any single condition met in isolation does not warrant reopening.

The three layers compound; removing any one does not unlock the exclusion.

Operational consequence

The exclusion is named in any sales conversation, design- partner pitch, advisor outreach, press interview, or product copy where the use case might otherwise be assumed compatible.

The institutional-tier bot terms of service surface the exclusion at install time.

The Personality OS for AI agents API documentation surfaces it in the use-case-restrictions section.

Related defense surfaces

The score-coordinate-level guard that operates inside the consumer self-reflection scope this entry binds is documented at Methodology / Anti-Barnum threshold.

The Anti-Barnum gate is the validity guard within scope; this entry is the structural exclusion that prevents the instrument being repurposed into scopes where the gate’s assumptions invert.

Both surfaces apply at once.

Pending future entries

The following policy decisions are documented internally and binding on current product scope but have not yet been formalized into ledger entries with the full structure above.

Each will receive a complete entry on its next scheduled review cycle, or earlier if a public-facing citation need fires.

• P-02 (candidate). Total-anonymity operator posture — the Tomas Cantor pseudonym is the public byline; the operator’s real name is not surfaced publicly. The brand discloses the legal entity (Metryval Pte Ltd, Singapore) where required.
• P-03 (candidate). Open-methodology lock — the methodology vocabulary is open and citable; the framework moves toward an open-license footing on a dated trajectory; validation datasets open on the longer horizon. The brand name is the only trademarked term.
• P-04 (candidate). Affordable and subsidized tier permanence — the free Lite tier remains free; the universal Read tier sits at a fixed cap; a financial- hardship tier is reserved; institutional partnerships sit at cost-recovery rates.
• P-05 (candidate). Anti-extractive brand posture — no engineered scarcity, no countdown timers, no charismatic-founder marketing, no dark-pattern retention mechanics.
• P-06 (candidate). Cross-cultural scalar measurement-invariance commitment — the Y3 methodology paper publishes results including failures, and the validation roadmap is staged across populations on a dated trajectory.

The pending status reflects the formalization queue, not the binding status.

Each candidate is already in force as a product commitment; the entries surface them with the full rationale + sources + review-conditions structure as the ledger matures.

Update protocol

This ledger is updated:

• Every six months, on scheduled review cadence.
• Within thirty days of any triggering event — a new use-case request that warrants a recorded refusal or acceptance, a regulatory change that affects an existing entry, a mission-constraint re-litigation in writing.
• At pre-launch of any major product-scope expansion that changes which surfaces an existing entry binds across.

When updated, the ledger version increments (v1 → v2 → ...) and the change log below documents what changed.

Change log

• v1 (2026-05-22). Ledger created. P-01 B2B selection exclusion filed as the first entry. Five candidate future entries enumerated (P-02 through P-06).