Public Systems Observatory / Peru alpha 01
Six source-backed research records make the visible evidence, unresolved questions, and review state inspectable. This alpha measures public documentation. It does not rate system safety, effectiveness, or institutional intent.
Record content is shown in English for this locale while the full translation is prepared.First-pass records · no second reviewer assigned · no agency response requested yet
One program / three evidence tracks
The tracks answer different public questions and keep different proof standards. They share source lineage, review gates, correction routes, and one rule: no combined score that hides what was actually measured.
Follow
Trace what public institutions say a system does, the evidence they expose, and what remains unavailable for independent review.
Test
Replay bounded, non-destructive public-service tasks and separate source fidelity, recovery, correctness, and unsafe action.
Audit
Measure the technical architecture municipalities declare publicly while preserving every snapshot and completeness gate.
Evidence index
The strips show evidence coverage, not quality or safety: documented, partial, or not found publicly in the reviewed source set.
Support qualification and proposed pronouncement drafting for candidate-list registration files.
Remote identity validation and photo capture for supported DNI services.
Assist judicial work with legal-information search, analysis, systematization, and drafting support.
Detect and monitor possible illegal logging using satellite and forest-oversight data.
Optimize MRI examinations in cancer-detection workflows.
Assist gastroenterology staff during digestive endoscopy.
Research record · Electoral
Jurado Nacional de Elecciones
Public purpose
Support qualification and proposed pronouncement drafting for candidate-list registration files.
Decision context
Electoral legal-administrative workflow with consequences for candidates and political organizations.
Inputs described
Registration files, attached documents, signature validation, JNE databases, and electoral rules.
Output described
Qualification summary and proposed resolution text.
Human oversight
Jurisdictional staff validate or modify the qualification; the JEE plenary reviews the proposed resolution.
Public evaluation evidence
The directive mentions tests and validation data. We found no public accuracy, correction, or disagreement metrics.
Public governance evidence
The directive cites data-protection rules and documents audit traces with permanent retention.
What the sources support
Missing public evidence
Next evidence request
Aggregate accuracy, correction, disagreement, and audit metrics after deployment.
Evidence coverage map
Each cell represents recoverable documentation for one field. It is not a composite score.
Claim boundary
This record does not evaluate electoral outcomes or claim that EleccIA makes final decisions. Provider and development provenance remain unresolved.
Research record · Identity
RENIEC
Public purpose
Remote identity validation and photo capture for supported DNI services.
Decision context
Biometric verification can determine whether a citizen continues a remote identity-document procedure.
Inputs described
Facial image capture, identity data, and mobile workflow inputs.
Output described
Identity-validation result used to continue the remote procedure.
Human oversight
The public sources reviewed do not describe a manual escalation or fallback protocol in enough detail.
Public evaluation evidence
We found no public false-match, false-non-match, accessibility, or demographic performance metrics.
Public governance evidence
Official service pages establish the biometric context, but app-specific retention and audit evidence remain incomplete.
What the sources support
Missing public evidence
Next evidence request
Aggregate biometric performance, accessibility, fallback, retention, and appeal data.
Evidence coverage map
Each cell represents recoverable documentation for one field. It is not a composite score.
Claim boundary
Internal development is confirmed by the evidence reviewed. This record does not assess biometric accuracy or infer harm from missing public metrics.
Research record · Justice
Poder Judicial
Public purpose
Assist judicial work with legal-information search, analysis, systematization, and drafting support.
Decision context
An internal legal assistant operating around high-stakes judicial workflows.
Inputs described
Judicial files, legal information, and jurisprudence are described at a high level.
Output described
Search, analysis, systematization, and draft support; exact operational boundaries remain unclear.
Human oversight
Assistant framing establishes a human user, but the review protocol and use restrictions are not public in the current evidence.
Public evaluation evidence
We found no public benchmark, hallucination test, retrieval evaluation, or quality metrics.
Public governance evidence
The reviewed public sources do not specify data retention, audit logging, incident handling, or matter-type restrictions.
What the sources support
Missing public evidence
Next evidence request
Deployment protocol, model and retrieval boundaries, evaluation results, review rules, and audit controls.
Evidence coverage map
Each cell represents recoverable documentation for one field. It is not a composite score.
Claim boundary
Internal development is confirmed. This record does not claim CURIA produces wrong outputs or makes judicial decisions.
Research record · Environment
OSINFOR
Public purpose
Detect and monitor possible illegal logging using satellite and forest-oversight data.
Decision context
Detection signals can influence field-work and environmental enforcement priorities.
Inputs described
Satellite imagery and forest-monitoring data; exact datasets require confirmation.
Output described
Detection or monitoring signal for possible illegal logging.
Human oversight
Official material says it complements field work, but the public review protocol is incomplete.
Public evaluation evidence
We found no public precision, recall, false-alarm, ground-truth, or regional performance metrics.
Public governance evidence
Operational-use material is public; audit trails and affected-party correction pathways were not found in this evidence set.
What the sources support
Missing public evidence
Next evidence request
Validation metrics, geospatial sources, review boundaries, correction process, and update history.
Evidence coverage map
Each cell represents recoverable documentation for one field. It is not a composite score.
Claim boundary
This record does not verify individual detections or infer an enforcement action from an algorithmic signal.
Research record · Health
Instituto Nacional de Enfermedades Neoplásicas
Public purpose
Optimize MRI examinations in cancer-detection workflows.
Decision context
AI-supported medical imaging operates around diagnostic care, but the exact clinical decision boundary is not public.
Inputs described
MRI imaging and clinical workflow inputs; modality details were not recovered.
Output described
Imaging optimization or assistance output; exact clinical output remains unclear.
Human oversight
Clinical review is expected in the surrounding workflow, but the AI-specific protocol is not public in the current evidence.
Public evaluation evidence
We found no public sensitivity, specificity, reader-study, cohort, or local clinical validation metrics.
Public governance evidence
The health-data context is high stakes. AI-specific privacy, audit, and monitoring evidence remain incomplete.
What the sources support
Missing public evidence
Next evidence request
Model identity, aggregate validation, clinical oversight, privacy controls, and post-deployment monitoring.
Evidence coverage map
Each cell represents recoverable documentation for one field. It is not a composite score.
Claim boundary
This record confirms a public deployment claim. It does not assess clinical safety, effectiveness, or diagnostic performance.
Research record · Health
Hospital Regional Lambayeque
Public purpose
Assist gastroenterology staff during digestive endoscopy.
Decision context
AI-supported endoscopy operates inside a clinical imaging workflow.
Inputs described
Endoscopy imagery or video stream; exact local data handling is not public.
Output described
Visual assistance during endoscopy; exact operational boundary needs confirmation.
Human oversight
Clinicians operate the service, but the AI-specific review, training, and fallback protocol is not public.
Public evaluation evidence
We found no public local sensitivity, specificity, detection, false-positive, or false-negative metrics.
Public governance evidence
The official page confirms equipment use. Local validation, patient disclosure, audit, and incident evidence remain incomplete.
What the sources support
Missing public evidence
Next evidence request
Aggregate local validation, staff protocol, patient disclosure, maintenance, privacy, and incident controls.
Evidence coverage map
Each cell represents recoverable documentation for one field. It is not a composite score.
Claim boundary
This record does not assess clinical safety, effectiveness, certification, or individual patient outcomes.
Method and limits
We begin with official catalogs, directives, service pages, and institutional publications. Every positive claim needs a linked source. Unknowns remain unknown. A first-pass review is not independent replication.
Start with official public sources and preserve the exact URL.
Keep confirmed facts, interpretations, and missing evidence distinct.
Expose review and agency-response state on every record.
Version material changes and keep a public correction route.
Current limitations
Independence
Crafter Research is an independent public-interest lab by Crafter Station. The institutions listed here did not commission these records and have no editorial control. Material support for this work will be disclosed.
Next release / 90-day contract
The program is already inspectable. Support pays for the expensive missing layers: independent review, agency response, evidence preservation, repeated trajectories, and publication maintenance.
Six reviewed records, an adjudication log, and bounded right-of-reply outreach.
Ten safe trajectories, scorer outputs, exclusions, and a public failure taxonomy.
Additional versioned observations tied only to complete Muniscan snapshots.