Living epistemic system · online

THE LIVING
DISSERTATION

A research environment for knowledge becoming.

This dissertation is not presented as a record of completed thought. It is a record of thought becoming.

Its propositions are exposed to interrogation, revision, contradiction and evidence.

Some ideas will survive. Some will change. Some will fail.
The record preserves all three.

Companion environments

Con-Met AI

The system being built

Living Dissertation

The research being formed

Living research network

QUESTIONSHYPOTHESESINTERACTIONDISCOVERYREVISIONTESTEMERGENCECURRENT KNOWLEDGE

Questions → hypotheses → interaction → discovery → revision → test → emergence → current knowledge. Human, AI and emergent nodes are distinguishable but wired together.

Research state

77 records
Stable

13

Emerging

17

Under interrogation

25

Revised

7

Disproven

6

Unknown

9

Counted from the research data model, not written into the design.

Collaboration central

This research is developed through human–AI interaction. The AI does not author it.

Human

Question

AI

Intervention

Human

Response

Emergent

Insight

Evidence

Test

Revised

Understanding

Provenance trail

The relevant unit may be the human–AI cognitive system.

  1. Origin · HumanIs offloading the right model?
  2. Interaction · AIOffloading implies one-way transfer of load.
  3. Transformation · HumanReframed from offloading to collaboration.
  4. Evidence · Human + AIInteractions in which the AI response changes the human's next move.
  5. Current form · EmergentCognitive collaboration — emerging trajectory, not established terminology.

Provenance trail

A provenance gap is evidence, not an accusation.

  1. Origin · HumanUntraceable material appears in a submission.
  2. Interaction · AIListed innocent explanations for gaps.
  3. Transformation · HumanRejected the automatic misconduct inference.
  4. Evidence · Human + AIOffline and peer-developed work produces gaps by construction.
  5. Current form · HumanThe system provides evidence; it does not manufacture accusations.

01 / Section

The question

Primary research question

What are we actually trying to know when we assess a student?

If we want to know how students think, how they question, how they evaluate, how they make decisions, how they collaborate with increasingly capable systems and what they themselves contribute to the work, then the final artifact may no longer be sufficient.

Question map · branch to sub-question, hypothesis, evidence, challenge, discovery

02 / Section

The proposition

Current propositionUnder interrogation

The future of assessment may not be about proving that a student worked without AI. It may be about determining what the student actually did when AI was available.

Not presented as established fact. Status: under investigation.

Evidence

Both students used AI. The final documents may look remarkably similar. Cognitively they may represent two entirely different processes.

Inference

If the artifact cannot separate those processes, the artifact alone is an insufficient basis for the inference assessment is trying to make.

Proposition

The future of assessment may not be about proving that a student worked without AI. It may be about determining what the student actually did when AI was available.

Open question

Can what the student actually did be established from the recorded interaction without distorting the behaviour being recorded?

03 / Section

The architecture

Click a node to open its definition, current understanding, evidence, open questions and last revision.

Conceptual architecture · Con-Met AI

Human contribution

Under interrogation

Definition

What the student originated, decided, questioned, evaluated, researched, revised, rejected, discovered and ultimately chose.

Current understanding

Treated as a dimension of attribution, not achievement. A value records the degree of contribution to a defined function, not its quality.

Evidence

  • ·Interaction records contain questions asked, claims rejected and evidence introduced by the student.

Open questions

  • ·Can minimal interaction conceal substantial intellectual contribution?
  • ·Can sophisticated prompting masquerade as sophisticated thinking?

Related artefacts

  • ·Contribution Metric specification
  • ·Evidence Map

04 / Section

The investigation

Expandable research threads. Each carries a question, a working hypothesis, arguments, counterarguments and status.

05 / Section

Now we try to break it.

A beautiful framework can still be wrong.
A sophisticated metric can still measure the wrong thing.
An evidence trail can still be misinterpreted.
A provenance map can still create false confidence.

Under interrogation

Assumption

Visible process is evidence of meaningful learning.

Attack

Can process be simulated or manipulated?

Counterexample

A student who has already written the work can stage an interrogation afterwards to produce a convincing record.

Result

Under investigation.

Under interrogation

Assumption

Contribution values represent contribution, not quality.

Attack

Will teachers read a high number as a good number regardless of the label?

Counterexample

AI receives a high value for language refinement while the student supplied the argument.

Result

Mitigation in place: attribution and assessment are held in separate stages with separate records. Reading behaviour untested.

Under interrogation

Assumption

More interaction indicates more cognition.

Attack

Can quantity masquerade as cognition?

Counterexample

A high-volume, low-depth exchange versus a single decisive intervention by an expert student.

Result

Under investigation. Volume sensitivity not yet characterised.

Under interrogation

Assumption

Sophisticated prompting indicates sophisticated thinking.

Attack

Can prompt craft be learned independently of understanding?

Counterexample

A memorised interrogation script applied to any task.

Result

Under investigation.

Under interrogation

Assumption

Minimal interaction indicates minimal contribution.

Attack

Can minimal interaction conceal substantial intellectual contribution?

Counterexample

A student who thinks offline and uses AI only once, precisely, at the end.

Result

Considered a live failure mode of the metric. The system cannot currently see contribution it did not record.

Unknown

Assumption

The metric can be optimised only by thinking better.

Attack

Can students learn to optimise the metric without becoming better thinkers?

Counterexample

Metric-aware behaviour taught as exam technique.

Result

Unknown. This is the distortion risk the research treats as disqualifying if confirmed.

Unknown

Assumption

The metric behaves consistently across tasks, disciplines and students.

Attack

Does the construct transfer?

Counterexample

A mathematics derivation versus a literature essay.

Result

Unknown. Not yet tested.

Disproven

Assumption

A provenance gap indicates misconduct.

Attack

Does a gap license that conclusion?

Counterexample

Work developed offline, on paper, or in conversation with a peer.

Result

Rejected. A gap is evidence requiring investigation, not proof of wrongdoing. The system provides evidence; it does not manufacture accusations.

Disproven

Assumption

“Did the student use AI?” is a useful assessment question.

Attack

What does the answer establish?

Counterexample

Two students, both using AI, producing similar documents through entirely different processes.

Result

Rejected. It establishes that a tool was present and little else.

Stable

Assumption

A beautiful framework is a correct framework.

Attack

Is the architecture persuasive because it is true, or because it is elegant?

Counterexample

A sophisticated metric can still measure the wrong thing; an evidence trail can still be misinterpreted.

Result

Standing principle of the research. Abandonment remains a legitimate outcome.

06 / Section

The discoveries

Unexpected conceptual developments, preserved with the interaction that produced them.

22 Aug 2026Revised

Perhaps cognitive offloading is the wrong question.

Original question

Is cognitive offloading the correct model for describing human–AI cognition?

Intervention

Conceptual interrogation.

Human

Reframed the problem from offloading to collaboration.

AI

Identified implications for distributed cognition and emergent contribution.

Emergent insight

The relevant unit may be the human–AI cognitive system rather than the individual actor.

Consequences for the research

Opened the cognitive-collaboration research thread.

22 Aug 2026Emerging

Perhaps the more important question is what happens when cognitive load becomes collaborative.

Original question

Who carries the cognitive load in an AI-assisted task?

Intervention

Reformulation of the load question.

Human

Rejected the single-carrier assumption.

AI

Supplied the redistribution framing as an intermediate step.

Emergent insight

A trajectory rather than a category: offloading → redistribution → collaboration → emergent cognition.

Consequences for the research

Marked as an emerging conceptual trajectory, not established terminology.

22 Aug 2026Unknown

What happens when the machine talks back?

Original question

Is the AI a tool or a participant in the interaction?

Intervention

Adversarial framing of the researcher's own model.

Human

Refused to settle the agency question prematurely.

AI

Surfaced the ambiguity in the word “artificial”.

Emergent insight

Did AI create collaborative cognition, or did it make an existing phenomenon visible?

Consequences for the research

Became a featured research thread; question remains open.

22 Aug 2026Emerging

AI may not have broken assessment. It may have exposed where assessment was already broken.

Original question

Did generative AI break assessment?

Intervention

Deliberately uncomfortable proposition.

Human

Redirected the problem from the tool to the measurement model.

AI

Held the counterposition that the artifact retains diagnostic value.

Emergent insight

The artifact still matters; it may simply no longer be the only thing that matters.

Consequences for the research

Set the framing for the whole investigation.

22 Aug 2026Stable

We built the metric. Which means we have no business trusting it.

Original question

How should the researchers relate to their own instrument?

Intervention

Adopted an explicitly adversarial relationship with the creation.

Human

Committed to attacking the metric rather than defending it.

AI

Enumerated attack surfaces: gaming, concealment, volume, prompt craft, transfer, distortion.

Emergent insight

The purpose is not to prove the metric works, but to determine whether it deserves to exist.

Consequences for the research

Created the pressure-test section as a first-class part of the research.

22 Aug 2026Emerging

We did not build two systems. We built one instrument twice.

Original question

Is there any relationship between Con-Met AI and Neuro-Mirror?

Intervention

Cross-project comparison, unplanned.

Human

Noticed the resemblance without having designed it.

AI

Located the shared thesis: the interaction, not the artefact, is the evidence.

Emergent insight

The two systems form a closed loop — each one's output is the other's missing input — along with the risk that the loop confirms itself.

Consequences for the research

Opened the convergence section and the self-confirming-loop pressure test.

07 / Section

The evolution

Versioned conceptual change. Compare any two states of an idea.

Git for ideas · pick two versions
v0.1 · BaselineDisproven

Did the student use AI?

+ Current · 22 Aug 2026Under interrogation

The future of assessment may not be about proving that a student worked without AI. It may be about determining what the student actually did when AI was available.

What changed

Reframed from detection to determination of contribution.

Why it changed

The question the research can actually answer with evidence.

What caused the change

Convergence of the contribution and provenance layers.

What evidence informed it

Metric → Evidence → Conversation → Context.

08 / Section

The artefacts

A research library. Provenance is recorded as human, AI, human + AI or emergent.

09 / Section

The evidence

Observation, interpretation and inference are kept visually distinct so they cannot be confused with conclusion.

SupportsAffects · Current proposition

Evidence

Two students both used AI and produced similar documents through entirely different processes.

Interpretation

The artifact does not discriminate between those processes.

Inference

Artifact-only assessment is insufficient where AI is available.

SupportsAffects · Human / AI / Emergent contribution

Evidence

Interaction records contain questions, decisions, revisions, rejections and introduced evidence.

Interpretation

The process leaves a traceable surface.

Inference

Contribution may be attributable at dimension level.

WeakensAffects · Contribution Metric

Evidence

A student can think offline and interact with AI only once, briefly, at the end.

Interpretation

Substantial contribution can exist outside the record.

Inference

The metric cannot see contribution it did not record.

ContradictsAffects · Assumption: visible process indicates learning

Evidence

A staged interrogation can be produced after the work is already written.

Interpretation

Visible process is forgeable.

Inference

Visible process is not, by itself, evidence of meaningful learning.

UncertainAffects · Metric neutrality

Evidence

AI may receive a high value for language refinement while the student supplied the argument.

Interpretation

Dimension-level values separate function from worth.

Inference

Whether readers preserve that separation in practice is not established.

OpenAffects · Reproducibility claim

Evidence

The metric has not been run across disciplines, tasks or cohorts.

Interpretation

Transfer is untested.

Inference

No inference is currently licensed.

ContradictsAffects · Provenance gap semantics

Evidence

Work developed offline, on paper, or with a peer produces a provenance gap.

Interpretation

A gap has innocent explanations.

Inference

A gap is evidence requiring investigation, not proof of misconduct.

UncertainAffects · Current proposition

Evidence

Students who know the metric may adapt their behaviour toward it.

Interpretation

Measurement may distort the measured behaviour.

Inference

If confirmed at scale, this would be disqualifying for the metric as designed.

10 / Section

Current state of knowledge

Not a conclusion. The present position of the research.

Last updated · 22 Aug 2026Status · Developing

What we currently believe

  • ·The question “Did the student use AI?” establishes that a tool was present and little else.
  • ·Human, AI and emergent contribution are usefully distinguishable channels rather than a division of ownership.
  • ·A contribution value must be an attribution of function, never a judgment of quality.
  • ·Con-Met measures contribution. The teacher evaluates learning.

Why we believe it

  • ·Two students using the same tool can produce near-identical artifacts through entirely different cognitive processes.
  • ·The interaction record contains questions, rejections, revisions and introduced evidence that the artifact does not.
  • ·Separating attribution from assessment preserves professional judgment rather than automating it.

What evidence supports it

  • ·Metric → Evidence → Conversation → Context: every value can be traced to interaction segments.
  • ·The provenance gap is defined as evidence requiring investigation, not proof of wrongdoing.

What remains uncertain

  • ·Whether visible process can be reliably distinguished from simulated process.
  • ·Whether the metric transfers across tasks, disciplines, students and forms of AI use.
  • ·Whether emergence is a property of the interaction or of the researcher's interpretation.
  • ·Whether the metric introduces incentives that distort the behaviour it measures.

What could change our minds

  • ·Evidence that students can optimise the metric without becoming better thinkers.
  • ·Evidence that attribution profiles are unstable across repeated analysis of the same transcript.
  • ·Evidence that the system produces distortions greater than the problem it addresses.
  • ·If the foundational premise collapses, abandonment remains a legitimate outcome.

11 / Section

Research ledger

The heart of the living system. Searchable, filterable, chronological.

22 Aug 2026Revised

Is cognitive offloading the correct model for describing human–AI cognition?

Intervention

Conceptual interrogation.

Evidence

Interaction sequences in which the AI response changes the human's next move.

Human contribution

Reframed the problem from offloading to collaboration.

AI contribution

Identified implications for distributed cognition and emergent contribution.

Emergent insight

The relevant unit may be the human–AI cognitive system rather than the individual actor.

Decision

Open research thread.

Next question

What makes a computational system a cognitive partner?

22 Aug 2026Stable

Did generative AI break assessment?

Intervention

Deliberately uncomfortable proposition.

Evidence

Assessment has long inferred thinking from the final product.

Human contribution

Proposed that AI exposed an existing weakness rather than creating a new one.

AI contribution

Held the counterposition that the artifact retains diagnostic value.

Emergent insight

The artifact still matters, but may no longer be the only thing that matters.

Decision

Adopt as the framing proposition of the research.

Next question

What are we actually trying to know when we assess a student?

22 Aug 2026Disproven

What does the question “Did you use AI?” establish?

Intervention

Analysis of the two-student case.

Evidence

Similar artifacts, divergent processes.

Human contribution

Constructed the paired counterexample.

AI contribution

Enumerated the dimensions the binary question fails to capture.

Emergent insight

Presence of a tool is not authorship of thought.

Decision

Reject detection framing.

Next question

How was AI used?

22 Aug 2026Stable

Should a contribution value carry academic meaning?

Intervention

Definitional separation.

Evidence

High AI value for refinement alongside a student-supplied argument.

Human contribution

Stripped achievement semantics from the metric.

AI contribution

Supplied the language-refinement case as a stress example.

Emergent insight

Contribution answers what function was performed, not how well.

Decision

Metric represents intensity of contribution to a defined dimension only.

Next question

Do readers preserve that separation in practice?

22 Aug 2026Stable

How should the researchers relate to their own instrument?

Intervention

Adversarial research design.

Evidence

Pressure-test log opened with ten assumptions.

Human contribution

Committed to attacking the metric: gaming, concealment, volume, prompt craft, transfer.

AI contribution

Enumerated attack surfaces and failure conditions.

Emergent insight

The objective is not to prove the metric works, but to determine whether it deserves to exist.

Decision

Publish breakage rather than hide it. Abandonment is a legitimate outcome.

Next question

Can the metric be manipulated?

22 Aug 2026Under interrogation

What does a provenance gap license a teacher to conclude?

Intervention

Boundary definition.

Evidence

Offline and peer-developed work produces gaps by construction.

Human contribution

Refused the misconduct inference.

AI contribution

Listed innocent explanations for untraceable material.

Emergent insight

A gap is evidence requiring investigation.

Decision

System provides evidence; it does not manufacture accusations.

Next question

Will gaps be read as guilt regardless of framing?

22 Aug 2026Unknown

What assumptions are introduced when we describe intelligence as artificial?

Intervention

Terminological interrogation.

Evidence

Insufficient.

Human contribution

Opened the question of agency and responsibility in attribution language.

AI contribution

Traced the term's implications for authorship and provenance.

Emergent insight

Attribution presupposes participants; the term both grants and withholds that status.

Decision

Open featured thread.

Next question

Does the terminology change how contribution is attributed?

12 / Section

The convergence

An unplanned feedback loop between Con-Met AI and Neuro-Mirror. Recorded here because it was discovered, not designed.

Two instruments, built separately, turned out to be one instrument built twice.

Noticed · 22 Aug 2026 · not a design decision

Closed cognitive loop

0102030405LOOPUNPLANNED

Hover a node. Each system's output is the other's missing input, so the sequence has no end point — it re-enters itself.

Shared thesis

Emerging

The artefact you produce is not the evidence. The interaction that produced it is. Con-Met refuses to read the essay and reads the transcript; Neuro-Mirror refuses the questionnaire and reads what you did under load. Both treat self-report and final output as the same category of unreliable witness.

How it was found

The connection was not designed. It was noticed mid-conversation while comparing Con-Met AI with Neuro-Mirror, a separately conceived system for profiling how a person actually thinks under task conditions.

Shared commitments

  • ·Mirror, not judge — describe where thinking came from rather than rank how good it was.
  • ·Behaviour over declaration — measure what was done, not what was claimed.
  • ·The self is partly opaque to itself — a student cannot reliably name which idea was theirs.
  • ·Provisional by design — every inference stays revisable, never hardened into a verdict.

What each system gives the other

Neuro-Mirror → Con-Met

A demonstrated-capability baseline.

Con-Met cannot currently distinguish “this student contributed little” from “this student contributed at their ceiling”.

Con-Met → Neuro-Mirror

An explicit evaluation rubric and a formal task context (Step 2).

Neuro-Mirror needs a structured evaluation frame to keep its inferences from floating free of any task.

Failure mode · Self-confirming loop

If each system treats the other's inference as ground truth, the pair can converge on a confident, self-generated fiction: a weak baseline predicts low contribution, the low contribution confirms the weak baseline.

Mitigation

No hypothesis produced by one pass may be promoted to ground truth on the next. `unknown` stays a live epistemic state the loop is forbidden to overwrite; each system records the provenance of every input it received from the other.

Open questions

  • ·Is the loop genuinely self-correcting, or only self-reinforcing under favourable conditions?
  • ·Can a demonstrated-capability baseline be built without becoming a fixed label attached to a learner?
  • ·Does convergence indicate a real underlying framework, or the researcher's own priors appearing twice?

13 / Section

Tool-choice provenance

The student does not work inside one tool. Selection, rejection, switching and return are themselves contribution evidence.

The student's selection, rejection, switching and evaluation of AI tools constitute potentially meaningful evidence within the contribution process.

Use is data.Non-use is data.Tool switching is data.Returning is data.

Con-Met cannot treat the use of another AI system as an event outside the provenance framework. A student who leaves Con-Met to seek divergent ideas, verify a fact or take a second opinion is making an intellectual decision, and that decision is data.

What the record captures

  • · Tool used and tool category
  • · Purpose of the tool and stage of work
  • · Student's stated reason for choosing it
  • · Expected contribution versus actual contribution received
  • · Whether the output was evaluated
  • · Whether it was accepted, rejected, modified or challenged
  • · Whether the student returned to Con-Met afterwards
  • · Relationship between the external interaction and the final contribution

Reasons kept distinct, never collapsed

  • · It performs a particular function better
  • · They want a second opinion
  • · They want divergent ideas
  • · They want factual verification
  • · They do not understand Con-Met's purpose
  • · Con-Met does not support the task
  • · They prefer another interface
  • · They misunderstand the distinction between tools
  • · They are attempting to avoid provenance documentation

What this is not

  • · Not surveillance.
  • · Not an integrity flag — external-tool use is never treated as suspicious by default.
  • · Not a claim that Con-Met is the only legitimate tool.

AI tool ecology — observed, not prescribed

Student thinkingCon-MetExternal AIStudent evaluationExternal sourceVerificationReturn to Con-MetReflectionRevisionFinal contribution

Procedural compliance

The student uses Con-Met because they were instructed to.

Conceptual understanding

The student understands why Con-Met exists and what contribution provenance is intended to reveal.

Con-Met usage does not equal Con-Met understanding. The research model keeps them apart.

New research dimension

AI Decision Literacy

Observation first. Deliberately not reduced to a score at this stage.

Open questions

  • · Does recorded tool choice change tool choice? Reflective checkpoints may alter the behaviour they observe.
  • · Can “avoiding documentation” be offered as a selectable reason without punishing honesty?
  • · Is non-use reliably distinguishable from non-disclosure?

14 / Section

Familiarity through adaptive processing

As the AI accumulates history with a student, less needs to be said. Less said is not less thought.

Accumulated human-AI interaction can reduce the amount of explicit information required to communicate complex intent. Visible input therefore cannot be treated as a complete representation of human contribution.

Early interaction · low familiarity

Explain this concept and help me develop three possible approaches.

The AI holds almost no relevant history. Everything must be stated.

Later interaction · high familiarity

Take the mechanism we developed before and apply it here.

Dramatically shorter, yet it invokes a whole shared framework: prior definitions, rejected alternatives, established terminology and previous decisions.

Shorter input does not necessarily mean lower contribution. The contribution may be compressed into shared contextual knowledge.

Familiarity state · adaptive processing trajectory

Stage 1

Initial

Student explains → AI responds

Stage 2

Emerging

Student directs → AI develops

Stage 3

Established

Student cues → AI anticipates

Stage 4

High

Student evaluates → AI refines

Stage 5

Highly contextual

Student provides minimal contextual signal → AI reconstructs likely intended direction

Research descriptors, not measurements of cognition. The underlying evidence for every classification is retained.

Explicit contribution

What the student directly communicates in the current interaction.

Contextual contribution

Relevant information about the student's thinking the AI can draw on because of accumulated interaction history.

Con-Met must not attribute intellectual expansion to the AI merely because the current prompt is short.

The provenance record must answer

  • · How long has the student interacted with this AI?
  • · How much relevant prior interaction exists?
  • · Is the current prompt dependent on previous interaction?
  • · Does the current input contain an explicit reference to prior work?
  • · How much of the AI's response appears to rely on accumulated context?
  • · Did the student confirm, reject or modify the AI's reconstruction?
  • · Did the student introduce new information? Did the AI?
  • · Did the student recognise when the AI misunderstood their intended meaning?

Heuristics Con-Met refuses

  • · More words = more contribution.
  • · More prompts = more contribution.
  • · Longer interaction = more contribution.

Contribution is modelled as current input + prior context + student decisions + AI processing + evaluation + revision + reflection.

Research questions added

  • · How does increasing AI familiarity with a student alter the observable distribution and interpretation of human and AI contribution over time?
  • · Does increased familiarity reduce the amount of explicit prompting required from the student?
  • · Does reduced prompting represent reduced contribution, or compressed contribution?
  • · How does accumulated interaction history affect provenance attribution?
  • · Does increased familiarity improve the AI's ability to reconstruct student intent?
  • · When the AI reconstructs incorrectly, how does the student correct it?
  • · Does familiarity create new forms of ambiguity in contribution attribution?

Combined provenance model

01 · Human contribution

What the student originates, decides, evaluates, rejects, modifies and contributes.

02 · AI contribution

What the AI generates, transforms, extends, reconstructs or suggests.

03 · Tool ecology

Which tools the student chooses, rejects, switches between and returns to.

04 · Familiarity

How accumulated interaction history changes what the AI can infer from limited explicit input.

05 · Evaluation

How the student judges and responds to machine contribution.

06 · Integration

How human and AI contributions become incorporated, transformed or rejected in the final work.

07 · Provenance

A traceable record of the contribution trajectory.

Con-Met is not an AI detector. It investigates how human and machine contributions interact, evolve and become distinguishable within an educational process. Contribution must therefore be understood longitudinally, not merely transactionally: human-AI contribution is dynamic, relational, contextual and longitudinal.

Open the tool ecology & familiarity analytics

15 / Section

Institutional integration

Provenance capture only becomes research data if it lives inside the work teachers already set.

A contribution-provenance layer that requires teachers to re-enter tasks, cohorts and deadlines already held in Blackboard, Canvas or Moodle will not survive contact with a real school week. Redundant administration is the most common cause of death for otherwise sound educational instrumentation.

Con-Met should not be a second gradebook. It should be a provenance layer that borrows identity, roster and task context from the institution's existing system and returns evidence, not marks.

Primary route

LTI 1.3 / LTI Advantage

Con-Met is registered as an external tool inside Blackboard. A teacher places a Con-Met activity in a course; the student clicks it and lands directly in the workspace, already identified.

  • · Single sign-on via the LMS — no separate Con-Met account
  • · Names and Role Provisioning gives the roster and teacher/student role automatically
  • · Deep Linking lets the teacher attach an existing Blackboard assignment to a Con-Met task
  • · Assignment and Grade Services can return a quality mark once the teacher has finalised it

Cost: Requires tool registration by the institution's LMS administrator and a formal privacy review.

Secondary route

REST API sync

A scheduled job reads course, roster and assignment data from the Blackboard REST API and mirrors it as Con-Met tasks; submissions and provenance summaries are written back.

  • · Works without embedding Con-Met in the LMS interface
  • · Useful for research deployments that need bulk historical data

Cost: Needs institutional API credentials, and identity mapping must be maintained separately.

Available now

Manual bridge

Teachers export a task list or roster from the LMS and import it into Con-Met; Con-Met exports a contribution and origin record that can be attached to the LMS submission.

  • · No institutional approval required
  • · Enough to run a pilot cohort this term

Cost: Still involves a copy step; acceptable for pilots, not for scale.

What crosses the boundary

  • · Con-Met sends the LMS an evidence record and, where the teacher has finalised one, a quality mark the teacher authored.
  • · Con-Met never sends the contribution intensity values as a grade. They are not a grade.
  • · The LMS remains the system of record for enrolment and assessment. Con-Met remains the system of record for provenance.

Why this is a research question

Institutional integration is not only a convenience question. Attaching provenance capture to the assignment flow teachers already use is what converts ordinary coursework into a longitudinal dataset — which is the precondition for studying how human-AI contribution changes over a year rather than over a single task.

Open questions

  • · Does embedding Con-Met inside the LMS change student behaviour, given it now appears as institutional surveillance rather than a workspace?
  • · Where does consent sit when provenance capture is delivered through a compulsory course activity?
  • · Does returning only a teacher-authored mark, and never a contribution score, survive institutional pressure to rank students by the contribution index?
  • · What is the minimum viable integration that removes teacher redundancy without requiring a full LTI registration?

Closed research–development loop

Con-Met AI is the artefact being developed. The Living Dissertation is the research environment documenting its intellectual development. Concepts here link to the components that operationalise them; the system links back to the propositions that generated it.