ALANSAgentic Engineering Internship
Governed production

Decision register · v1.0

Automation has boundaries. People have names.

The platform records who may propose, verify, decide, disclose and appeal. Consequential authority is never inferred from a model response.

Phase-0 authority
01

Traceable, not “perfectly replayable”

Decision records retain evidence digests, policy versions, tool outcomes and human disposition. Secrets, private code and hidden reasoning are excluded.

02

Formative is not selective

Hint use, accommodations and private mentor notes support learning. They are not credential claims or hiring features.

03

Verification is not judgment

A green test establishes an observed behavior under named conditions. A reviewer applies the approved rubric and owns the phase decision.

Capability boundaries

What each system may—and may not—do

CapabilityPermittedProhibitedAccountable owner
Learner assistant

Explain curriculum, ask Socratic questions and cite approved resources.

Score help-seeking, reveal mentor notes or complete graded work.

Program content owner
Evidence verifier

Run approved deterministic checks in an isolated environment.

Use production credentials, infer competence or modify learner work.

Platform operations
Assessment assistant

Draft evidence-linked observations in shadow mode.

Pass a gate, invent missing evidence or overwrite a human record.

Assessment lead
Credential service

Issue or revoke an approved, minimal credential deterministically.

Choose a level, expose private evidence or recommend employment.

Credential officer
Hiring dossier

Present learner-consented evidence to the authorized panel.

Reuse formative analytics or rank candidates autonomously.

Hiring panel chair

Appeal path

A decision can be challenged without penalty.

  1. 1Learner sees the evidence, rubric version and reason codes.
  2. 2Learner files a correction or appeal within the published window.
  3. 3An independent reviewer receives the frozen evidence bundle.
  4. 4The outcome supersedes—never silently overwrites—the prior record.

Current release posture

Shadow AI with no decision authority

EnabledCurriculum, evidence requirements and human rubrics
EnabledPrivate evidence manifests and bounded verification queues
Shadow onlyGrounded assessment drafts, evidence coverage and reviewer comparison metrics
Human governedCriterion findings, moderation, accommodations, reassessment and appeals
DisabledAutomated passes, public badges and hiring recommendations