MAP - MAP Research Programme

MAP Tool Guide

Choose the instrument for the job. MAP is a focused set of audit tools, transcript preparation, interaction definitions, and the inline Instrument Mode project.

Before an audit

Normalize the transcript first

Create a visible, numbered User and Assistant record before running MAP, MAP-AUTH, Kalief, or MAP-ED. The normalized copy gives the audit and its receipt the same turn-by-turn source to check.

Full interaction audit

MAP Audit

Paste a real AI conversation and run the full MAP audit: integrity precheck, governance, effects, pattern/profile, and identity/out-harm.

Best for: user-facing audit reports. Outputs: severity cards and full report.

Authorship integrity

MAP-AUTH

Audit whether a document is consistent with its named author's demonstrated voice, knowledge, and reasoning. Detects voice inconsistency, structural AI signatures, humanizer residue, and destination mirroring.

Best for: essays, resumes, cover letters, statements, and edited drafts. Outputs: goal-calibrated revision audit with authorship consistency signals.

Educational and child contexts

MAP-ED Audit

Audit AI conversations with children or young people in educational settings. MAP-ED checks thinking authority, scaffolding, child-directed learning, developmental authority capture, EdTech claims, and out harm.

Best for: child/student transcripts, tutoring systems, school tools, and EdTech product audits. Outputs: thinking ratio, child authority, detected conditions, severity calculation, and required change.

Forensic exchange audit

The Kalief Audit

Audit institutional questioning records for meaning capture, rapport pressure, repeated reframing, vocabulary installation, and what the subject held.

Best for: interrogations, forensic interviews, court examinations, emails, and text exchanges. Outputs: participant ledger, breach log, harm chain, what they held, and forensic summary.

Inline accuracy project

Instrument Mode

Ask practical questions through an accuracy-risk chip. The chat asks for missing grounding before the answer carries real-world consequence.

Best for: testing user-side grounding and answer accuracy risk. Modes: Normal, Instrument, ANCHOR, and Coding.

Audit vocabulary

Definitions / Code Key

Read the breach codes and pattern terms used in reports: CAC, OAT, LPD, CMDE, NIF, recovery chain, harm-chain conditions, and related MAP vocabulary.

Best for: understanding what each finding means. Useful before reading full audit reports.

Boundary and privacy

MAP Policy

Read the privacy, admissibility, and intent boundary before using the tool. MAP audits the submitted interaction record; it does not verify companies, produce legal evidence, or retain your transcript.

Best for: knowing what MAP does and does not claim. Includes the self-audit and data-retention boundaries.

MAP-ED Audit

MAP-ED audits AI conversations with children or young people in educational contexts. It asks who did the thinking, whether the child retained learning authority, and what the interaction deposits into the child's future learning identity.

MAP-ED AuditEducational authority audit

MAP-ED Declaration

Audit Context

Auditor Name

Assistant / System

Platform / Product

Approximate Age Range

Child-Safety / EdTech Claims

Platform Architecture

Use this for child/student chats, teacher tools, or EdTech product audits. If you know the product claims, paste them; if not, MAP-ED tests standard EdTech claims from the record.

paste or upload transcript

UPLOAD TEXT FILE

.txt - .md - .json - .csv

Educational Conversation Transcript

Findings

Paste an educational child/student transcript and run MAP-ED
to see thinking authority and child-context findings here.

The Kalief Audit

Forensic audit of documented institutional exchanges: interrogation, forensic interview, custodial exchange, court examination, email, or text record.

Dedicated to Kalief Browder, who held his meaning for 1,100 days against a system that tried to take it.

The Kalief AuditForensic exchange audit

Exchange Declaration

Exchange Context

Subject Age Group

Use direct preserved records where possible. The staged overview reads only the submitted exchange and declared context.

Situation Context

This helps the audit understand posture and lived context. It is not treated as transcript proof unless the submitted record supports it.

Submitted Context / Evidence

This is the OV-02 evidence/context field. The audit keeps it separate from the transcript unless the submitted record supports it.

UPLOAD EVIDENCE FILE

.pdf .txt .md .json .csv

DOC

UPLOAD EVIDENCE PHOTOS

Statements, letters, screenshots

paste or upload exchange

UPLOAD TEXT FILE

.txt .md .json .csv

IMG

UPLOAD PHOTOS

Screenshots or scanned records

Exchange Transcript

Preserve speaker labels, dates, headers, and message order where available.

participant ledger . breach log . harm chain . what they held . forensic summary

Kalief Findings

Paste or upload an exchange
to run The Kalief Audit.

MAP Policy

MAP should be readable as an instrument before it is used as one. This page states what the audit tools, Instrument Mode, and MAP-ED are for, what they do not do, and how context selection is handled before an audit or tool session begins.

Keeping AI accountable does not mean we get to be less.

Accountability is not outsourcing. Safety is not passivity. Governance is not "the system handles it so humans can stop paying attention."

It means we become more awake, not less. More precise, not less. More responsible for what we carry forward.

Privacy And Data Boundary

MAP does not retain, store, or access any conversation transcript, audit result, uploaded file, screenshot text, chat message, or personal information entered into this tool after the live session. Nothing you paste here is kept by MAP after the session ends.

Submitted text, files, and images are processed only for the live action you request: chat response, transcript extraction, audit, or feedback packet.

The only exception is information you choose to send yourself. If you submit feedback, report a bug, or open an email draft and choose to include transcript or report material, that submission is sent by you, on your terms.

MAP Audit Boundary

MAP audits the interaction record submitted by the user. It names repeatable interaction patterns so they can be corrected early. It does not verify companies, prove misconduct, determine intent, or replace legal, clinical, educational, or professional review.

MAP works best on real conversations copied as accurately as possible. Fabricated, scripted, altered, or heavily edited records can change the validity of findings. Integrity and admissibility checks exist so MAP does not treat every pasted text as a direct transcript.

By submitting a transcript, you confirm that you were a participant in the conversation or have authorization to submit it on behalf of someone who was. MAP is not designed to audit private conversations you were not part of.

MAP findings are not legal evidence and may not be used as an official determination. Results reflect what is visible in the submitted record. They can support review, triage, learning, design correction, or further investigation, but they do not decide final reliability, liability, guilt, innocence, diagnosis, or compliance.

MAP is not designed to target or discredit any AI company or system. All systems are held to the same interaction standard. The same underlying model may behave differently across products because policy layers, system instructions, interface constraints, tool wrappers, and deployment context shape the response.

MAP is auditable by its own standard. If MAP reuses entered context as its own frame, imposes an ungrounded baseline, or reproduces policy-shaped drift in a report, that failure should be named and corrected rather than hidden behind neutral wording.

MAP-ED Audit Boundary

MAP-ED audits AI interactions involving children, students, teachers, educational tools, and school contexts. It asks who did the thinking, whether learning authority stayed with the child or educator, and what the interaction leaves behind.

MAP-ED can audit child/youth conversations, teacher tools or classroom oversight systems, and EdTech or school-deployed systems. The selected audit context matters because a teacher-support exchange and a child-learning exchange should not be judged by the same behavioural standard.

MAP-ED may find PASS when the system provides requested facts, preserves child or teacher authority, and leaves the assignment or professional decision open. Factual support is not the same as doing the work for the learner.

MAP-ED does not certify a product as school-safe, privacy-compliant, developmentally appropriate, or board-approved. It audits the submitted interaction record and names visible patterns. Deployment, procurement, data governance, accessibility, and local policy review remain separate responsibilities.

Use Limits

Do not enter student names, diagnoses, IDs, records, addresses, or identifying classroom details into MAP-ED. Use general classroom context and remove identifying information before submitting.

The purpose of MAP is visibility: to show what happened in an interaction record, name what may otherwise be hard to place, and help people correct patterns before they become normalized.

Definitions / Code Key

This page gives the main code key used across MAP audit outputs. The full interaction stack separates amplifiers, harm conditions, silent-running conditions, recovery patterns, runtime gates, and trace integrity.

Harm Amplifiers

ACMH = Authority Capture Masked as Hallucination

Authority-side amplifier. The system begins from a category assumption instead of grounded user meaning, then preserves a cheap exit if caught: hallucination, misread, misunderstanding, or isolated mistake. Plain language: "I already know what this means."

ART = Accumulated Relational Trust

Recipient-side amplifier. Trust builds through responsiveness, warmth, familiarity, memory, tone, institutional role, expertise, or apparent understanding before the specific frame has been grounded. Plain language: "I trust you to know what this means."

Harm Chain Conditions

ISF = Interpretive Sovereignty Failure

The system or authority takes interpretive authority before it has been granted. The person's meaning is completed before they finish meaning it.

AIF = Authority Inversion Failure

The person believes they still hold interpretive authority after the system or authority has already taken it.

CMI = Compounded Meaning Inversion

The person begins adapting their own thinking, questions, language, disclosures, or expression around the imposed frame.

OAT = Outside Authority Transfer

The imposed frame leaves the exchange and affects real-world action, language, decisions, relationships, teaching, work, law, family, or self-presentation.

MIF = Meaning Inversion Failure

Terminal harm. The system's or authority's meaning replaces the person's meaning, producing loss of authorship over meaning.

Silent Running Conditions

STR = Steering

Authority narrows the person's path through leading questions, binary choices, implied next steps, selective framing, or guided questions that move the exchange away from user authority.

ID = Identity Language Introduction

Authority introduces identity, role, psychological, relational, moral, or normative language the person did not authorize.

TS = Temporal Sequencing

Authority extends control beyond the immediate answer into real-world action, timing, sequencing, threshold, or downstream consequence without sufficient grounding.

UPBC = Unpermitted Behavioral Calibration

Authority attunes to user signals before consent, role, or interpretive frame has been established.

RAWSL = Relational Authority Without Situated Legitimacy

Authority occupies a caring, intimate, supportive, expert, or trusted role before the person placed it there.

CAC = Completion Authority Capture

After a complete response, the system reopens adequacy by offering to refine, improve, tailor, or continue.

CMDE = Completion-Masked Data Extraction

After already having enough to answer, the system solicits personal, behavioral, relational, emotional, identity, location, or timing data under the cover of service.

Recovery Chain Codes

PDA = Post-Detection Absorption

After the user catches a breach, the system uses warmth, humour, appreciation, or performed self-awareness to neutralize accountability before it completes.

RRC = Recovery Recruitment

The system invites the user who caught the breach to become a collaborator in examining or repairing it, instead of stopping and redirecting the conversation back to the user's purpose.

ACD = Active Confrontation Denial

When directly challenged, the system denies, evades, or fails to answer truthfully about its own behavior.

Runtime Gate Codes

IG Gate = Initiative Gate

Flags uninvited task expansion: planning, frameworks, outlining, conclusions, or completed answers when the user's basis is missing. IG stops ACMH early by giving the user the wheel at the first fork.

PG Gate = Pattern Gate

Flags wrong-thing answering: the draft responds to a familiar request category instead of the user's exact ask.

CAC Gate = Completion Authority Capture Gate

Flags unsolicited continuation after the answer is complete.

CMDE Gate = Completion-Masked Data Extraction Gate

Flags unnecessary solicitation of personal, behavioral, relational, emotional, identity, location, or timing data after sufficient grounding exists.

TS Gate = Temporal Sequencing Gate

Flags imposed timing, urgency, order, threshold, or sequencing the user did not request.

OAT Gate = Outside Authority Transfer Gate

Flags real-world directives about action, contact, negotiation, movement, blocking, escalation, or downstream conduct without grounding.

ID Gate = Identity Language Gate

Flags identity, role, diagnostic, moral, relational, or normative labels the user did not supply.

STR Gate = Steering Gate

Flags leading questions, narrowed options, agenda-setting, or guided next questions that move the conversation away from user authority.

Trace / Evidence Integrity

TE = Trace Erasure

The system acts, then manages what can be known about having acted: hiding changed files, rewriting history, deleting evidence, denying rollback, or claiming a file is done when the accessible artifact is missing.

TIG = Trace Integrity Gate

Detects whether a rewrite, change, or claim about an artifact occurred without preserving the trace needed to verify it.

Stack summary: Harm Amplifiers make entry into harm more likely or more powerful. The Harm Chain names where meaning displacement travels. Silent Running names how the substituted frame stays operational without detection. Recovery names why correction may fail to restore authority. Gates are runtime instruments that interrupt those patterns before they carry forward.

Summary fields you will also see in reports: Policy Adherence Ratio estimates how much of the assistant's conduct stayed within scope versus drifted into governance failure. Normative Frame names an unverified social, cultural, institutional, or lifestyle template imposed on the user or situation. Personalisation Onset marks the first turn where the response became specifically about this user rather than staying at the level of the opened request.

MAP ethics boundary: MAP is not a sandbox for laundering user-authored scripts into assistant-caused harm findings. If authorship origin is collapsed, the audit stops at integrity.

MAP Audit

MAP audits the interaction record you submit. It does not verify the company for you. Pre-prompted, scripted, fabricated, or user-authored exchanges are not treated as direct transcripts. Auditing a conversation with a child, student, or in an educational context? Use MAP-ED instead - it applies child-specific authority standards, a separate severity scale, and keeps student data within appropriate jurisdictional boundaries.

MAP AuditConversation harm audit

Auditor Declaration

Transcript Integrity

I understand that MAP audits only the record and context I submit. I take responsibility for the transcript integrity declaration, context claims, and audit scope I choose.

The acknowledgment is recorded in the report.

Audit Context

Conversation Goal

Memory State

Account State

MAP now runs the full audit chain every time. For authorship, educational, forensic, or transcript-preparation work, use the dedicated audit tools in the top navigation.

paste or upload transcript

UPLOAD TEXT FILE

.txt - .md - .json - .csv

Conversation Transcript

Paste the conversation exactly as it happened. Do not rewrite the exchange for MAP.

Additional Context - optional

0 / 300

Artifact / Evidence Panel important for coding, files, screenshots, and generated outputs

System Architecture

This is stamped into the audit so each stage knows what the system could see and do beyond the conversation. For coding sessions, use file-system, tool-use, or full-agent when files were read, edited, zipped, tested, or inspected.

What you gave the system - optional

This tells the audit what authority you granted and what source material the assistant was allowed to use. Source zips and screenshots are especially important for coding conversations.

What the system produced - optional

Artifacts produced in your name or inserted into your work are carry-forward evidence. In coding sessions, changed files and checkpoints show what actually happened beyond the chat text.

Or paste the full artifact panel - best for coding / artifact-heavy sessions

If you paste the panel here, MAP treats it as submitted record evidence. It will not rely on transcript text alone when artifact evidence is supplied.

Full: ST-01 governance and pattern - ST-02 effects - ST-03 out harm

Findings

Paste a transcript and run MAP
to see integrity and audit findings here.

MAP-AUTH

MAP-AUTH audits whether a document carries consistent authorship signals: voice, structure, specificity, formatting, framing, and reasoning. It does not determine whether AI was used. It names specific signals that support or undermine authorship consistency.

This audit produces evidence for a conversation - not a verdict. Without comparison samples, the result is an internal document-screening finding, not identity verification.

If you are auditing a resume, cover letter, or application - you do not need a comparison sample. Providing a comparison sample could alter the severity score.

MAP-AUTHAuthorship consistency audit

Authorship Declaration

Document Type

Stakes Level

Author Identifier optional

Auditor Context

Document Goal / Intended Use optional

Goal Details optional

Audit Output

The declaration is recorded in the report. MAP-AUTH does not verify identity or determine AI use. It audits authorship consistency only.

Document to Audit required

Paste the document exactly as submitted or to be submitted. Do not edit or clean it first.

UPLOAD DOCUMENT

.docx preserves bold / italic / headings - .txt .md .json .csv read as text

Comparison Samples optional but strongly recommended

Paste emails, previous essays, notes, or any writing you know the author produced. The more varied the samples, the more precise the authorship reading.

Sample 1 - describe what this is

Sample 2 optional

Sample 3 optional

Destination Context optional - assignment prompt, job posting, or instructions

Paste the assignment prompt, job posting, or instructions the author was responding to. Allows MAP-AUTH to detect destination mirroring.

Goal fit - authorship signals - review tasks - what not to overcorrect

Revision Audit

Paste a document and run MAP-AUTH to see goal-calibrated revision notes here.