Individual-referential
A baseline is learned from one life, not borrowed from a population. The population audits; it never decides.
Barkley AI · IREP ProtocolThe systems lab for
individual intelligence
Purpose-built for systems that learn the individual.
Designed for evidence, agency, and trust.
NOV-0072
Nova — behavior against her own baseline
Baseline learned from 182 days of her own history. The population reference is attached as an audit instrument, not as a verdict.
Night activity · last 14 nightsNova's baselineLast 14 nights
What changed
Tue 02:10
Sleep fragmentation above her baseline4 nights · +2.1σWatch
Wed 07:30
Water intakewithin her rangeNormal
Thu 18:05
Play and walkwithin her rangeNormal
Should I worry? Nova · NOV-0072− ↗ ×
Re-check in 3 days; escalate if water or appetite move with it.The population reference audits this call — it does not make it.
What changed — and should I worry? ↑
IREP-0417
Applicant 0417 — decision on evidence
Category-blind, trajectory-rich. Identity and category data stay sealed until the decision is final: the role is the standard of decision, the population the instrument of audit.
01 · Category-blind screening02 · Untimed assessment · 24h window03 · Decision on evidence04 · Post-decision measurement
What the reviewer sees
E1
Led the payments migration to a new ledger2024–2026 · 3 artefacts · 2 referencesVerified
E2
Rebuilt on-call: incidents down 38% over 12 monthsIncident log · dated · exportedVerified
E3
Mentored four engineers to seniorPromotion records · pending sourcePending
Is this decision on evidence? Applicant 0417− ↗ ×
Rationale documented by the reviewer.The equity audit runs after the decision is final — never before it.
What did this person do — and what proves it? ↑
ACTA-0218
Session 0218 — authorship at the gesture
Events are signed as they happen, under keys the artist holds. The registry — not the file — is the root of trust. Nothing here is inferred.
Capture attestationSession tree · Merkle root signedLongitudinal registry · RFC 3161C2PA embedding · selective disclosure
Session events
09:41:07
capturetake_03.wavSigned
09:52:30
bouncestems_v2.wavSigned
10:14:02
generative_render (declared)pad_texture.wavDeclared
10:20:11
exportmaster.wav · C2PA manifestSigned
Who made this? Session 0218− ↗ ×
The generative render is recorded as declared.Nothing is inferred about the human behind the work.
What happened — and what was declared? ↑
A new species of systems lab. Purpose-built around the individual, Barkley Labs creates intelligence and evidence protocols for decisions that averages cannot make.
Evidence before interpretation
What happened is kept, in order, and signed where it matters — before any model, reviewer or platform reads it.
ACTA MusicOne architecture, three domains
Behavior, creative work and human evaluation share the same four steps, in the same order: observe, preserve, compare, decide.
Barkley AI · IREP Protocol · ACTA MusicInference is useful. Evidence should come first.
Evidence before interpretation.Barkley AI · IREP Protocol · ACTA Music
A person is not their category.
Evaluate the person, not the category
IREP is an open protocol for human evaluation that is category-blind and trajectory-rich: a decision is made on the evidence of what someone did, with inherited proxies kept out by design.
2.0 IREP Protocol →A work is not its output.
Attest authorship at the gesture, not the file
ACTA Music records signed session events while a work is being made, under keys the artist holds. Provenance that exists before the export — and survives it.
3.0 ACTA Music →A dog is not its breed.
Know one dog against her own history
Barkley AI learns a behavioral baseline from a single animal's life and answers the question every owner actually asks: what changed — and should I worry? The population is an instrument of audit, never a verdict.
0.988 vs 0.935ROC AUC — individual baseline vs breed average
100% vs 81%Declines caught
~34 daysEarlier detection, median lead time
Head-to-head validation v2.0 · 30 seeds · synthetic data
1.0 Barkley AI →Make behavior
legible
Turn longitudinal observations into individual baselines, detect meaningful drift, and route weak signals toward the right human attention.
1.0 Barkley AIBacklog
BL-208
Map identity-free inputs
Map identity-free inputs
BL-209
Baseline failure modes
Baseline failure modes
In progress
BL-212
Individual reference model
Individual reference model
BL-213
Evidence graph
Evidence graph
Validated
BL-218
Negative falsification
Negative falsification
BL-219
Longitudinal registry
Longitudinal registry
Thread in #research
1.1 Individual baseline1.2 Drift detection1.3 Evidence trace1.4 Human review
Design decisions around
the individual
Separate identity from evaluation. Compare each case to its own relevant evidence, then audit the system after the decision — never before it.
2.0 IREP ProtocolSystems roadmap2026 ⌄
MARAPRMAYJUNJULAUGSEPOCT
InitiativesIndividual intelligenceDecision integrityCreative provenanceOpen research
Barkley AIBenchmarkPublic beta
IREP ProtocolDecision layerSteward model
ACTA MusicReference impl.Pilot
PublicationsWorking papersPeer review
Make creative evidence
verifiable
Create cryptographic records of authorship and process with selective disclosure — enough evidence to verify a claim, without exposing the work behind it.
3.0 ACTA Music01 / CREATESource eventLocal evidence recorded
02 / ASSERTSigned claimHash · timestamp · witness
03 / DISCLOSESelective proofMinimum necessary evidence
04 / VERIFYIndependent checkNo platform trust required
✓
Provenance verifiedRecord BL-ACTA-0218
- Assertion
- authorship.v1
- Integrity
- SHA-256
- Witness
- 2 signatures
- Disclosure
- Selective
Different domains.
Same architectural decision.
Every Barkley Labs system runs the same four steps, in the same order. Inference is useful. Evidence comes first.
4.0 Method4.0 MethodApplied to three systemsObserve → Preserve → Compare → Decide
01Observeone individual, over time
02Preservekeep it, in order, signed
03Compareagainst their own history
04Decidea human, with the evidence
Barkley AIBehavior
Daily signals from one animal's life
A baseline kept per individual
Drift against her own baseline
What changed — and should I worry?
IREP ProtocolEvaluation
Evidence of what a person did
Records dated, sourced, exportable
The trajectory, never the category
A decision trace that cites evidence
ACTA MusicCreation
Capture events at the gesture
A signed Merkle log under artist-held keys
Each session against the artist's own anchored registry
An authorship claim, with its proof
Research
Methods before marketing. Public claims backed by working papers, benchmarks, and reference implementations.
ACTA (Attestation of Creative Trajectory and Authorship): an individual-referential, artist-held provenance protocol for music composition
Working paper v0.1 — authorship attests at the gesture, not the file. CC BY 4.0.
DOI 10.5281/zenodo.21230054
An open standard for category-blind, trajectory-rich assessment in hiring and admission. Specification CC BY 4.0, demonstrators Apache-2.0.
DOI 10.5281/zenodo.21211589
Toward individual longitudinal baselines in companion-animal behavioral monitoring. CC BY-NC 4.0.
DOI 10.5281/zenodo.20756552
A unified framework for precision behavioral and metabolic intelligence in companion animals. CC BY-NC 4.0.
DOI 10.5281/zenodo.20060327
Source-available 8-layer stack: individual baseline, temporal layer, drift engine, silence layer, reference class, behavioral replay, head-to-head and synthetic validation. Python, research license (free for research, education and evaluation).
DOI 10.5281/zenodo.20754351
Individual baseline vs breed average on synthetic cohorts: ROC AUC 0.988 vs 0.935 · declines caught 100% vs 81% · median lead time ~34 days earlier · 30 seeds.
The Normative TrapTemporal identity and the failure of population intelligence — why a living being cannot be understood through the average of its kind.
DOI 10.5281/zenodo.20516821
Built and running: session trees, longitudinal registry, RFC 3161 anchoring, C2PA embedding, 97 tests. Held in private pre-release under patent pending (FR2609502, FR2609280); access under written evaluation terms.
Reference implementationsTests, falsification cases, and technical documentation.
FounderElodie Aishwarya Remoissenet — founder and protocol architect. ORCID 0009-0004-6031-659X.
PatentsBarkley AI: French patent applications FR2605477 and FR2605026 — patent pending. ACTA: French patent applications FR2609502 and FR2609280 — patent pending; disclosure and exploitation authorized by the INPI; filed to keep ACTA's mechanisms from enclosure.
LicensesIREP: specification CC BY 4.0, demonstration code Apache-2.0. ACTA: working paper CC BY 4.0; specification and reference implementation in private pre-release. Barkley reference architecture and demonstrators: source-available research license (research, education and evaluation), synthetic data only.
TrademarksBarkley AI™ and ACTA Music™ are trademarks of Barkley Labs. The IREP Protocol is open and published by Barkley Labs.
We build for n = 1.
Independent research & systems architecture. Behavior, provenance, human evaluation.
Fairness by design.Evidence before inference.Privacy by architecture.The individual is the reference.
© 2026 Barkley LabsBarkley AI™ · ACTA Music™ — trademarks of Barkley LabsIndividual, not average.