Peira

Litepaper: The Assessment Layer for the Onchain World

Litepaper v0.1 · 2026-08-11

Status: draft for public review. This litepaper makes the case for a research-validated assessment layer built on proof-of-personhood. The research bibliographies behind the instruments, plus the platform's identity and ZK design notes, are published on this site (linked in §12); the full engineering specification is still internal.

1. Executive summary

The onchain world proved identity. It has not proved character.

World ID and its peers solved sybil-resistance: a wallet can be bound to one unique human. But every onchain organization is still flying blind on the thing that matters most: the character and capability of the human who holds the wallet. Their moral reasoning. Their creative capacity. Their judgment under trade-offs.

Peira is a research-validated moral and creative assessment built for this gap. One test per human, taken once, owned forever, shared with organizations of the candidate's choosing. It is grounded in decades of published psychometrics, engineered to resist LLM-assisted cheating, and explicitly not a medical or diagnostic tool. It ships first as an operated service with on-chain verifiable integrity, and is architected to decentralize: canonical instruments committed on-chain, any operator able to deliver, scoring that is deterministic and therefore trustlessly verifiable.

This document is the short version. The claim is modest and specific: the onchain world deserves an assessment layer that is scientifically credible, privacy-respecting, and cannot be gamed by an LLM. The pieces to build it already exist.

2. The problem

2.1 Identity was solved. Trust was not.

Proof-of-personhood primitives (World ID, Passport, biometric uniqueness) answer "is this a real, unique human?" They do not answer "how does that human reason and create?" A sybil-resistant wallet can still belong to someone who lies under pressure, caves to pressure, or cannot reason across a hard moral trade-off. For roles with real consequence (treasury signers, moderators, governors, hires), that gap is the whole ballgame.

2.2 Legacy integrity tests are broken.

The dominant "integrity" and "values" tests are direct Likert self-report questionnaires: "I have never told a lie." They are fakeable, carry documented adverse impact, and have not meaningfully improved since the U.S. Office of Technology Assessment flagged their validity threats in 1990: adverse impact, privacy invasion, low fidelity, and faking. They are the reference class this project is positioned against.

2.3 The result.

Onchain organizations currently select humans using identity infrastructure plus vibes. There is no content assessment in the stack. The dimension most consequential to the role is the one never measured.

2.4 Why these dimensions

Morality and creativity are the two capacities AI imitates most convincingly and possesses least. They are also the two that decide outcomes in roles of real consequence.

In high-stakes work you need people who can think laterally when the playbook runs out, and whose judgment you trust when nobody is watching. Lateral thinking finds the move no one scripted. Alignment with the mission means the work gets done, and done right, without anyone looking over a shoulder.

Both qualities have to attach to a person. On the internet, people show exactly as much of themselves as they choose, so a unique human with real ability still needs a way to prove they are the kind of candidate these roles require, and to do it without surrendering anonymity. That is why Peira measures these two dimensions, and why each result is bound to a verified human rather than to a name or a wallet.

As the world fills with generated content, the genuinely human qualities become the scarce ones. Peira exists to keep them measurable and trustworthy.

3. The solution, in plain terms

A candidate takes a single, one-time battery of six research instruments (~3 hours; split across sittings). The result is keyed to their World ID nullifier. Never their wallet. Never their name. They own it. They grant each organization access to it, and revoke that access at will.

The battery measures two domains:

Creativity: creative achievement (lifetime recognition), divergent thinking (idea-generation capacity), cognitive style (adapter vs. innovator), everyday creativity.
Morality: moral foundations (which values you weigh when they conflict), and moral-dilemma reasoning (how you balance outcomes, duties, and restraint).

Organizations see consented candidates' subscores with configurable, role-tuned weighting. They never see raw responses, scoring keys, or candidates who have not opted in. Selection decisions remain solely the organization's; the platform provides descriptive data.

4. Architecture & trust model

(Full detail: implementation plan §3.0, §3.0.1, §3.0.2, §3.4.)

The platform is a canonical assessment protocol: publicly-defined instruments, deterministic scoring, verifiable administration. It is operated by a single steward in v1 and decentralizes in v2.

4.1 Verifiable integrity in v1 (no oracle required)

The canonical item bank is content-committed on-chain: each instrument's signed item set is registered as a hash in a minimal content registry. At delivery time, the candidate's client verifies that the items it is about to render match the on-chain commitment. A tampered administration cannot reach a candidate without detection.

Uniqueness and instrument integrity are fused into a single signature: the World ID verify call's signal is set to hash(canonical item set ∥ per-candidate perturbation seed), so the proof-of-personhood also attests "this unique human was presented this exact item set." No third-party oracle is needed for instrument integrity.

4.2 Deterministic scoring is the trust-minimized core

Five of the six instruments score deterministically: the same responses always produce the same score, against fixed, published scoring keys. That makes scoring reproducible and verifiable: anyone can recompute a score from committed responses and confirm it. This is the layer that survives the v1→v2 transition unchanged, because it does not depend on who delivered the test.

The one non-deterministic element is the AI-assisted scoring on the divergent-thinking task (semantic-distance originality, LLM-categorized flexibility). This stays an operator-attested, model-pinned off-chain computation, the irreducibly trusted piece, isolated and labeled as such.

4.3 The v1 → v2 transition is a delivery-layer change, not a re-architecture

v1 keeps canonical plaintext delivery single-operator, preserving the item secrecy that keeps most instruments valid. v2 publishes the canonical bank openly and lets any registered operator deliver it, gated on perturbation being load-bearing enough that openness is non-fatal to validity (§8). The scoring core, identity layer, and integrity primitives are identical across the transition by design.

5. The instruments & why validity matters

Each instrument is a structural replica of a peer-reviewed measure, not an invented questionnaire. This is the entire basis of the platform's credibility, and the reason the team pushes back on "just LLM-generate the items" (see §10).

Code	Instrument	Measures	Grounded in
CAI	Creative Achievement	Pro-c / Big-C lifetime creative recognition	Carson, Peterson & Higgins (2005), CAQ
AUT	Divergent Thinking	Capacity to generate many, varied, unusual ideas	Guilford (1967); scored via Beketayev & Runco (2016)
CPSS	Cognitive Style	Adapter–innovator problem-solving orientation	Kirton (1976) adaption–innovation tradition
ECI	Everyday Creativity	Little-c creative practice in daily life	Kaufman & Beghetto (2009), Four-C model
MVP	Moral Foundations	Which moral foundations you weigh under conflict	Atari et al. (2023), MFQ-2
MTRS	Dilemma Reasoning	Outcomes vs. duties vs. restraint in hard trade-offs	Gawronski et al. (2017), CNI/ODR model

Non-clinical framing. The platform administers research instruments for organizational selection and descriptive profiling. It is explicitly not a medical device, not a diagnostic tool, not a substitute for professional advice. Scores are probabilistic descriptions of response patterns. This framing is enforced across consent, results, and marketing.

6. Integrity: built for the AI-cheating era

A human can copy a test item into a chatbot and paste the answer back. World ID cannot prevent that. So the platform layers defenses, honestly framed, because no single layer is a silver bullet:

Per-candidate perturbation. Every candidate sees a structurally identical but surface-different version of each item (different names, magnitudes, contexts). Memorizing the canonical item is useless. This is the strongest defense and is load-bearing for the v2 openness transition.
Forced-choice format. Moral and cognitive-style items present two parties with legitimate claims; there is no "right" answer to endorse. Substantially harder to fake than Likert self-report.
Stylometric + behavioral detection. Free-text responses are screened for the statistical signatures of AI-generated prose; behavioral signals (response timing, paste events, focus loss) feed a post-hoc classifier. Raw behavioral signals never leave the candidate's browser.
One human, one test. World ID binds a single result to a single unique human, once, forever. Bot-pipe farming is impossible at the account layer.

The output is an integrity band (High / Medium / Low) surfaced to the reviewing organization. It is not a blunt auto-pass/fail, and it never exposes the raw classifier internals (which would help adversaries tune against it).

7. Privacy & identity

Nullifier-keyed. A result is bound to a World ID nullifier, a pseudonymous, unlinkable credential. Never the wallet, never the name, never an email.
Identity separation. The platform never holds the candidate's real name. Where verification requires identity (creative-achievement evidence), the candidate supplies pointers to artifacts; the candidate's name appears on the artifact only when a consented reviewer opens it, within audited disclosure scope.
Consent-driven portability. The candidate grants each organization access and revokes it. There is no data-ownership edge between candidate and org, only a consent edge.
v2: zero-knowledge selective disclosure. The schema is pre-shaped so a future ZK layer can reveal a single subscore to a verifier without revealing the rest of the profile.

8. Distribution & decentralization

v1: operated, verifiable. A single steward runs delivery, scoring, and the integrity classifier. Canonical instruments are content-committed on-chain (integrity is public); plaintext stays gated (validity is preserved). Organizations subscribe to use the service.

v2: decentralized. The canonical bank publishes openly; any registered operator can deliver the test from the public bank and open scoring code; scoring remains deterministic and verifiable; EAS score-attestations and ZK selective disclosure land. Governance of the on-chain contracts is upgradeable but timelocked and multisig-controlled, with no single-party rewrite key. EAS upgrades happen by registering new schema versions rather than mutating old ones, so past attestations stay verifiable.

The openness gate is honest. The bank does not flip to fully-open until perturbation is validated as load-bearing across every instrument where secrecy currently defends validity. Until then, the protocol ships verifiable-integrity-only rather than shipping broken instruments.

9. Sustainability and economics: an open question

(Not resolved. This section states the options and the decision criteria rather than a chosen model.)

A decentralized v2 protocol needs some economic mechanism for operators and recurring costs (LLM scoring, integrity APIs, hosting, World ID verification are all real per-candidate costs, and all off-chain). The candidate models:

Subscription + grant. v1 funded by org subscriptions (USD/USDC) and grant funding; v2 operators charge for the hosted experience; the raw protocol is free. Cleanest for a public-good ethos; avoids token-regulatory complexity.
Token + treasury. Operators stake and are slashable for misbehavior; fees flow in-token; a governance token coordinates protocol upgrades. More standard for a decentralized protocol; adds regulatory and design weight.

Decision criteria (to be resolved before v2): who are the operators, what skin do they have in the game, and what's the minimum viable economic security for an integrity-critical credential? The team's lean is toward the subscription + grant model unless decentralization demands staked operators. This is explicitly open and will be the subject of a separate design pass.

10. Risks & limitations (honest)

Small-cohort reliability. Several instruments (notably divergent-thinking originality, moral-dilemma parameter estimation) are cohort-relative; at the small cohort sizes realistic for v1 (5–20 candidates), percentiles are noisy. The platform surfaces this caveat, refuses percentile claims below n = 5, and defers cross-organization norming to v2.
The Kirton verification gate. One instrument (cognitive style) is a replica of a commercially licensed measure. Until its citation and licensing are verified directly with the publisher, it ships as a 3-item pilot subset rather than the full 16.
Adverse impact. Any selection instrument can produce demographic skew. The platform builds in 4/5ths-rule monitoring on opt-in demographic data, treats LLM-detector false positives on non-native English as a real adverse-impact concern, and keeps a human reviewer in the loop on every integrity flag.
No silver bullet on cheating. The integrity stack raises the cost and detectability of LLM-assisted gaming; it does not eliminate it. The platform says so in candidate and reviewer documentation.
Openness vs. validity. Publishing item banks openly degrades validity for secrecy-protected instruments. The platform accepts this trade-off only for the one instrument (creative achievement) whose defense is evidence verification rather than secrecy, and gates openness for the rest on perturbation being load-bearing.
Why not fully LLM-generate the items. It would collapse construct validity (a generated item is a new, unvalidated instrument sharing only a label), break the scoring-key determinism that the trust-minimized core depends on, and likely make items easier for LLMs to answer. LLMs are used only where they genuinely help (surface perturbation, pool expansion under psychometric gating, per-org values items), not as the sole author of the canonical battery.

11. Roadmap

v1 (operated, verifiable integrity). Six-instrument battery; World ID gating; on-chain content commitments + signal-binding; consent-driven portability; non-clinical framing enforced.
Augmentation track (parallel). Generalized perturbation engine; LLM-drafted pool expansion under mandatory pilot gating; LLM-eval harness; post-administration item analysis + result nullification (re-score, not re-take; World ID enforces one test per human).
v2 (decentralized). Canonical bank published openly (gated on §8); multi-operator delivery; EAS score-attestations; ZK selective disclosure; governed, upgradeable contracts; optional TEE-attested verification.

12. References & further reading

Primary sources

Atari, M., Haidt, J., Graham, J., Koleva, S., Stevens, S. T., & Dehghani, M. (2023). Morality Beyond the WEIRD. JPSP 125(5). DOI 10.1037/pspp0000470.
Beketayev, K., & Runco, M. A. (2016). Scoring Divergent Thinking Tests by Computer With a Semantics-Based Algorithm. Europe's Journal of Psychology 12(2):210–220. DOI 10.5964/ejop.v12i2.1127. (CC-BY 3.0)
Carson, S. H., Peterson, J. B., & Higgins, D. M. (2005). Reliability, Validity, and Factor Structure of the Creative Achievement Questionnaire. Creativity Research Journal 17(1):37–50.
Gawronski, B., Armstrong, J., Conway, P., Friesdorf, R., & Hütter, M. (2017). Consequences, Norms, and Generalized Inaction in Moral Dilemmas: The CNI Model. JPSP 113(4).
Kaufman, J. C., & Beghetto, R. A. (2009). Beyond Big and Little: The Four C Model of Creativity. Review of General Psychology 13(1):1–12.
Kirton, M. J. (1976). Adaptors and Innovators. JAP 61(5). (verification-gated)

Companion documents (published on this site)

Morality research bibliography: annotated sources behind the moral-reasoning instruments (MVP, MTRS).
Creativity research bibliography: annotated sources behind the creativity instruments (CAI, AUT, CPSS, ECI).
ZK feasibility memo: private, verifiable score credentials for v2.

The implementation plan and instrument-level specification remain internal at this stage; §4 and §5 of this litepaper summarize their load-bearing decisions.

This litepaper describes a research-instrument platform for organizational selection and descriptive profiling. It is not a medical device, does not provide medical or mental health diagnoses, and is not a substitute for professional advice. Selection decisions are made solely by participating organizations. © 2026 Peira. Draft for public review. Feedback welcome.