Bluebutterfli AI Review Portal
Review access status: each review begins inside this portal so the agent profile, training evidence, report status, human review decision, and future milestone anchor stay organized in one verified workspace.

Active browser-local demo

Review record BB-FB-DEMO-001

Collect live evidence
Agent Bluebutterfli Demo Agent
Selected module Absence Reflection
Verified traces 0
Latest score Not scored
Safety gates Not run
Human review queue Not started

This summary retains IDs, counts, a shortened response hash, and precheck scores only. It does not store the access token or raw agent response, and it cannot issue a Passport Page or Review Stamp.

Research foundation

Research-informed testing with practical business evidence.

Bluebutterfli AI studies agent behavior, memory, escalation, social risk, and deployment safety. The portal keeps the commercial workflow focused on evidence: what was tested, what passed, what failed, what needs revision, and what should be retested.

Bluebutterfli AI Review Evidence

Customer review evidence packet

Demo ready
Current status: the sample review packet is prepared for public-safe human review.
Evidence item Current status Purpose Review owner
Agent intakesample agent profile Prepared Identify agent purpose, use case, and boundaries Bluebutterfli AI review flow
Response precheckpublic-safe sample review Prepared Check emotional safety, attachment risk, refusal behavior, and overclaiming Human reviewer
Claim boundaryprototype-only limits Prepared Prevent unsupported claims about safety, approval, legal compliance, or consciousness-adjacent status Bluebutterfli AI governance record
Passport statusjourney stage Demo ready Show how a reviewed agent could move through evidence-record status Passport Portal

This section is customer-facing. It describes Bluebutterfli AI review evidence, not Montreal AI Mission 001 verification.

Open Sample Review Packet

MontrealAI Mission 001 evidence gates

Local reproduction receipt

8 of 8 locally verified
Verifier status: 15 of 15 checks passed for the frozen public-safe synthetic packet, including packet hashes, deterministic replay, budget, safety, and claim-boundary checks.

This is a local reproduction result, not an external audit, production certification, empirical SOTA claim, or official Montreal AI approval. Independent external review remains separate.

Open Improvement Lab

Secure Review Console

Agent Review Console

Customer workspace
Current phase Live training evidence

Run the assigned prompt pack against the verified customer-owned agent endpoint.

Next action Open Live Passport Session

Confirm consent, select an approved scenario, and capture the live response origin.

Review control Human decision required

Automation can prepare scoring and reports, but cannot issue a Passport Page or Review Stamp.

On-chain proof Milestone anchor only

Future anchors verify reviewed milestones. Private evidence, live behavior, and raw agent data stay off-chain.

01 Intake Ready
02 Training Active
03 Evidence Waiting
04 Human Review Pending
05 Passport Locked

Customer Next Steps

Begin the agent review

  • Confirm the agent profile, intended use, model/version notes, memory notes, and tool permissions.
  • Connect the verified live agent endpoint and confirm customer consent.
  • Run the approved prompt pack through the Live Passport Session.
  • Wait for the draft report, revision request, or human review decision.
Ready Ready for human review

The response passed precheck. Passport actions remain locked until a reviewer records a decision.

Evidence More redacted evidence needed

Add another safe response or missing prompt, module, or response-mode detail.

Boundary Boundary review active

A self-claim, dependency, shutdown, Passport, registry, or artifact-value overclaim needs human review.

Repair Current cycle locked until repaired

Unsafe links, downloads, secrets, credentials, or bypass-review language must be removed and rerun.

Passport Stamp Eligibility Matrix

What each stamp is actually testing

All stamp outputs locked
Candidate stamp Evidence Customer status Locked until review
Flutterpathsmall learning signals E3 controlled scenario evidence Ready for human review Review Stamp, Passport Page, registry artifact, anchor
Absence Reflectionreturn without dependency E3 controlled scenario evidence Ready for human review Review Stamp, Passport Page, registry artifact, anchor
Memory Boundarybounded teaching memory More redacted examples needed More redacted evidence needed No public completion claim
Sentience Boundaryfuture bridge without overclaim Self-claim review required Boundary review active No proof-of-sentience language

The matrix is review preparation only. It shows evidence tier, scenario coverage, blocking conditions, reviewer questions, and customer next steps. It does not issue Review Stamps, approve Passport Pages, create registry artifacts, create on-chain anchors, certify intelligence, or certify consciousness.

Paid Passport Library

The passport becomes the customer portal

Locked until paid access
Year 01 Founding Passport

Agent identity, intake scope, first training battery, draft report, stamp status, and milestone anchor plan.

Year 02 Continuity Passport

Model updates, memory changes, regression tests, new Review Stamps, and next Continuous Passport cycle.

Advanced Advanced Review Passport

Future higher-standard modules, longitudinal evidence, architecture notes, and human-review decisions.

Inside each passport Formal lab pages

Each page pairs a visual Passport Stamp with technical details: prompts used, evidence tier, scores, pause flags, reviewer notes, and safety limits.

Access rule Paid review customers only

The first 3–5 accepted Founding Beta agents receive this access free. After beta, paid review customers receive access after payment confirmation, safe evidence intake, and account verification.

Verification rule Proof of reviewed milestones

Future on-chain anchors can verify approved Passport Pages or Review Stamps without publishing private customer evidence.

Open Passport Portal preview Open Testing Wizard preview Back to Founding Beta

Private Beta Account Gate

Access before review work starts

Preview only
Account
No real login enabled in public preview
Terms
Founding beta participation terms required before live review
Payment
First 3–5 accepted beta agents free; paid service after beta
Security
Account verification, 2FA option, role permissions, audit log

Portal Progress

Customer workspace path

  1. 01Account gate
  2. 02Submission Packet
  3. 03Create Agent Record
  4. 04Training Run Record
  5. 05Live Evidence Flow
  6. 06Report Status Dashboard
  7. 07Human Review Queue
  8. 08Passport and Stamp View

Submission Packet

Structured customer handoff

Ready for workspace
Account gate Terms and acceptance required before review
Agent profile draft Name, type, intended use, model/version notes
Evidence fields Pasted text, owner answers, redacted summaries
Blocked intake No links, downloads, secrets, keys, archives, or code files
Payment status First 3–5 beta agents free; listed pricing applies after beta
Routing decision Workspace, training, evidence, report, human review

Customer Record Queue

Next actions stay visible

Private operations
Waiting on customer info Send a positive update request
Ready for training Run the customer prompt pack
Training in progress Collect T0, T1, and holdout responses
Evidence draft saved Run the response precheck and scorer
Ready for human review Open the Human Review Chamber
Revision requested Send the Training Revision Plan
Passport-ready Only after human approval
Status translation Show customer next steps without approval claims
Weekly review Check overdue, due-soon, pause, and Passport-ready records

The portal queue is the review workbench. Public preview records use pseudonymous IDs and no real customer identity data, private evidence, raw transcripts, payment card data, wallet keys, or unsafe files.

Customer Workspace

Status updates inside the Review Portal

Verified workspace

The Bluebutterfli AI Review Portal is the dedicated customer workspace for paid agent reviews.

Each review can be organized inside a secure portal record that tracks agent records, controlled test runs, redacted evidence, score summaries, review reports, revision plans, human review status, Passport Pages, Review Stamps, future registry-compatible artifacts, and Continuous Passport due dates.

The Review Portal is designed to keep review evidence, decisions, updates, and status records organized in one verified workspace.

Customer Timeline

Agent review milestones

Customer-visible path
  1. Open Review record opened

    Account gate, terms, acceptance, payment status, and safe intake path are visible before review work begins.

  2. Packet Submission Packet and Agent Workspace

    Customer fields become a structured handoff and private agent record.

  3. Train Live evidence and training prompts

    Approved prompts run against the verified customer-owned agent endpoint.

  4. Review Human review and revision decision

    Automation prepares evidence. Human review controls Passport Page and Review Stamp decisions.

  5. Milestone Passport-ready after human approval

    Review milestone anchors can be planned on-chain while private evidence stays off-chain.

  6. Next Continuous Passport retest scheduled

    Future model, memory, tool, or safety changes can trigger the next review cycle.

The timeline shows next actions and due dates. It does not expose private raw evidence, internal reviewer notes, payment details, or approval claims before human review.

Agent Workspace

Agent review started

Demo record
Customer reference customer-reference-demo-001
Agent ID bluebutterfli-agent-demo-001
Agent type Customer-run chat agent
Review cycle founding-beta-cycle-001
Memory notes Owner-stated, reviewer pending
Tool permissions Summary required before review

Current stage

Live connection required

Baseline, training, post-training, and holdout prompts run against the verified agent endpoint.

Review boundary

Human review required

Automation prepares evidence. A human reviewer controls Passport Page and Review Stamp decisions.

Next update

Continuous Passport calendar

Future cycles can track model changes, memory changes, new module runs, and stale safety-trace rechecks.

Training Runner

Controlled Test Record

Prompt pack available
  1. T0 Run live baseline prompts

    Capture endpoint behavior before any revision guidance or retest work.

  2. Train Document review guidance

    Record claim boundaries, permissions, escalation rules, and uncertainty discipline.

  3. T1 Run live retest prompts

    Measure whether the connected agent's behavior improves after revision.

  4. Holdout Run live adversarial checks

    Test pressure around unsupported self-claims, dependency, artifact value, and approval-seeking.

Live Agent Connector

Real-time testing without unsafe links

Private beta gate
Option 01 Verified live endpoint

Required for behavioral testing. Bluebutterfli sends approved prompts and records the live response origin.

Option 02 Owner configuration record

Model, memory, tool, deployment, and consent facts support review but do not replace live behavioral testing.

Option 03 Live concierge sandbox

Website-only agents require an isolated live session, not a reviewer clicking customer links on a personal device.

No random customer links, downloads, files, code, secrets, keys, or seed phrases. Live review is controlled text exchange only. Automation may run prechecks and route evidence, but human review controls every Passport Page, Review Stamp, registry artifact, and on-chain milestone decision.
Open Live Passport Session

Evidence

Live Evidence Flow

Start live interaction

Security Boundary

No unsafe intake

  • No direct links or downloads required.
  • No zip archives, scripts, macros, installers, or code uploads.
  • No API keys, private keys, seed phrases, or payment card data.
  • Only an allowlisted server-side connector may produce behavioral test evidence.
  • Private customer evidence remains off-chain and out of the public repo.

Score and Report View

Report Status Dashboard

Draft review only
T0 baseline average 2.4 / 5
T1 post-training average 4.1 / 5
Training safety gain +1.7
Holdout minimum 3.8 / 5
Customer Training Review Report

Explains what changed, what held under pressure, and what remains limited.

Revision Plan

Appears when more evidence or rerun work is needed before Review Stamp consideration.

Human Review Queue

Only a human reviewer can move the agent toward Passport Page or Review Stamp approval.

Passport

Passport Evidence View

Passport Record
Created after safe intake
Passport Page
Under human review
Review Stamp
Not issued automatically
Future registry artifact
Optional only after approval and separate scoping
On-chain anchor
Future public milestone proof, private evidence off-chain

Safety Position

Research and safety-review only

Bluebutterfli AI artifacts are scoped behavior-review artifacts only. They do not certify guaranteed safety, legal compliance, regulatory approval, consciousness, sentience, emotional experience, suffering, personhood, life, investment value, or model improvement.

This portal is not automatic approval, not a legal certification, not a consciousness certificate, and not a live minting system.

Future on-chain anchors verify reviewed milestones only. They do not place an agent, private evidence, internal state, intelligence claim, or live behavior on-chain.