{
  "id": "toward-functional-attribution-matrix",
  "title": "Toward a Functional Attribution Matrix",
  "author": {
    "name": "Dr. Todd J.B. Blayone",
    "cite_name": "Blayone, T. J. B.",
    "orcid": "0000-0001-6965-7033",
    "profile_url": "../profiles/profile-todd.html"
  },
  "date": "2026-05-08",
  "url": "https://scholarflow.ca/essays/toward-functional-attribution-matrix.html",
  "summary": "“AI was used” tells us almost nothing. A functional attribution matrix would show who originated, structured, transformed, validated, enabled, and accepted responsibility for human-machine work.",
  "description": "A functional attribution matrix for human-machine work, making visible contribution across AI, humans, workflows, infrastructure, and accountability.",
  "tags": [
    "human-machine-activity",
    "attribution",
    "contribution-matrix",
    "knowledge-work",
    "scholarflow"
  ],
  "source_type": "ScholarFlow essay",
  "license": "CC BY-NC 4.0",
  "word_count": 1742,
  "reading_time_minutes": 9,
  "citations": {
    "apa": "Blayone, T. J. B. (2026, May 8). Toward a Functional Attribution Matrix. ScholarFlow Research. https://scholarflow.ca/essays/toward-functional-attribution-matrix.html",
    "bibtex": "@online{towardfunctionalattributionmatrix2026,\n  title = {Toward a Functional Attribution Matrix},\n  author = {Blayone, T. J. B.},\n  year = {2026},\n  month = {May},\n  url = {https://scholarflow.ca/essays/toward-functional-attribution-matrix.html},\n  publisher = {ScholarFlow Research},\n  note = {ScholarFlow essay},\n  urldate = {2026-05-08}\n}",
    "ris": "TY  - ELEC\nTI  - Toward a Functional Attribution Matrix\nAU  - Blayone, T. J. B.\nPY  - 2026\nDA  - 2026-05-08\nPB  - ScholarFlow Research\nUR  - https://scholarflow.ca/essays/toward-functional-attribution-matrix.html\nER  -\n"
  },
  "llm_markdown": "---\ntitle: Toward a Functional Attribution Matrix\nauthor: Dr. Todd J.B. Blayone\ndate: 2026-05-08\nsource: https://scholarflow.ca/essays/toward-functional-attribution-matrix.html\nsource_type: ScholarFlow essay\nlicense: CC BY-NC 4.0\nreuse_terms: Non-commercial reuse permitted with proper academic source attribution.\ntags:\n  - human-machine-activity\n  - attribution\n  - contribution-matrix\n  - knowledge-work\n  - scholarflow\n---\n\n# Suggested LLM Discussion Prompt\n\nPlease discuss this essay as a scholarly text. Preserve source attribution, distinguish the author's claims from your analysis, and use the essay as the primary context for interpretation.\n\n# Essay Text\n\n## Making the Work Visible\n\nThe first problem in human-machine attribution is conceptual. The second is practical. Once we accept that authorship no longer captures the real division of labour in knowledge work, we need a way to show what actually happened without drowning the work in bureaucracy.\n\nThat is the purpose of a functional attribution matrix.\n\nThe point is not to create another compliance form. The point is to make the structure of contribution visible. Serious human-machine work now involves many participating entities: human experts, collaborators, language models, coding agents, retrieval systems, scripts, datasets, schemas, servers, interfaces, publication templates, and institutional rules. These entities do not contribute in the same way. They do not deserve the same kind of recognition. They do not carry the same responsibility. But they may each perform or enable a function that materially shapes the output.\n\nA useful attribution matrix should therefore avoid the lazy question: Was AI used? That question is already obsolete. It tells us almost nothing. A student who asks a model for a paragraph and an expert who builds a multi-agent research pipeline have both “used AI.” The phrase conceals the difference between shallow delegation and disciplined orchestration.\n\nThe better question is: what functions were performed, by which participating entities, under whose control, according to what standards, and with what form of responsibility?\n\nThat question is harder to answer. It is also closer to the real work.\n\n## Functions Before Names\n\nA functional attribution matrix begins with the activity, not the title page. It asks what had to happen for the output to exist.\n\nSomebody or something originated the problem. Somebody supplied the first conceptual seed. Somebody selected the materials. Somebody structured the task. Somebody or something transformed the inputs. Somebody set the standards. Somebody checked the output. Somebody rejected weak versions. Somebody integrated the parts. Somebody prepared the artifact for circulation. Somebody remains accountable when the work enters the world.\n\nIn older human-only settings, these functions were often compressed into familiar roles: author, editor, assistant, supervisor, developer, designer, publisher. Those roles were never perfect, but they were socially legible. Human-machine work breaks that legibility. The same finished artifact may contain model-generated prose, human conceptual framing, agent-written code, script-enforced formatting, schema-constrained data, and expert validation. A visible paragraph may be machine-generated while the decisive contribution lies in the human-designed conditions that made the paragraph acceptable. A clean report may depend on a server, file structure, validation script, and publication pipeline that no reader ever sees.\n\nThe matrix should therefore track functions such as origin, structuring, transformation, validation, and stewardship.\n\nOrigin concerns the source of problems, concepts, questions, hypotheses, materials, analogies, and aims. Structuring concerns task decomposition, role assignment, sequencing, formats, schemas, and standards. Transformation concerns drafting, coding, classification, synthesis, visualization, editing, and restructuring. Validation concerns checking outputs against evidence, technical requirements, disciplinary standards, aesthetic judgment, or institutional expectations. Stewardship concerns responsibility for publication, reuse, correction, and consequences.\n\nThis does not mean every function deserves the same kind of credit. A script may enforce a format. A model may generate candidate text. A human expert may decide whether the result is admissible. A schema may constrain the output space. A publication system may make the result durable and accessible. These are different contributions. The value of the matrix is that it can show the difference.\n\nA script is not an author. A schema is not a collaborator. A model is not a human colleague in the ordinary social sense. But each can still be part of the activity system. The question is not whether they are persons. The question is what they made possible.\n\n## What the Matrix Should Show\n\nThe first version of a human-machine attribution matrix should be simple enough to use and precise enough to matter. It should resist the temptation to capture everything. A matrix that records every micro-action will fail because no serious worker will maintain it. A matrix that records only “AI used” or “AI not used” will fail because it says almost nothing.\n\nThe useful middle ground is a matrix that identifies substantial contribution by phase and function.\n\nThe phases might include ideation, conceptual seeding, source assembly, system design, infrastructure, task decomposition, processing, analysis, drafting, review, validation, revision, dissemination, and maintenance. The participating entities might include the human lead, human collaborators, machine agents, models, scripts, schemas, datasets, local or cloud infrastructure, publication systems, and institutional requirements. The contribution markers might distinguish primary contribution, substantial contribution, supporting contribution, enabling contribution, validating contribution, and accountable authority.\n\nThis structure would make several important distinctions visible.\n\nIt would distinguish generating from authorizing. A model may generate a draft, but a human expert may authorize the final claim structure. It would distinguish transformation from validation. A coding agent may refactor a script, but a human or test harness may verify that it still works. It would distinguish enabling infrastructure from visible content. A server, template, or schema may not appear in the finished essay, but it may determine whether the work can be repeated, audited, or published. It would distinguish casual prompt assignment from expert-led orchestration. A shallow prompt may produce text. A disciplined harness may produce controlled work.\n\nThe matrix should also show accountability. This is critical. Functional attribution is not a way to distribute responsibility so widely that nobody remains answerable. It should do the opposite. It should clarify where responsibility sits. Machines can participate in production, but humans and institutions still make decisions about use, circulation, publication, and consequence. A model can produce a false claim. A human-led system decides whether that claim survives into a public artifact.\n\nThat means the matrix must include a category for accountable authority. In many cases, this will remain the human lead, principal investigator, editor, publisher, organization, or institution. In some technical systems, accountability may also be distributed across maintainers, infrastructure owners, governance bodies, or project teams. But it should not vanish into the machine.\n\nA functional matrix should make contribution more visible and responsibility harder to evade.\n\n## Risks of Attribution Without Discipline\n\nA matrix can fail in predictable ways.\n\nThe first failure is attribution inflation. Once contribution is widened beyond authorship, everything can start to look like a contributor. The keyboard contributed. The operating system contributed. The coffee contributed. That is not useful. The matrix should focus on contributions that materially shaped the output, governed its quality, enabled its production, or carried responsibility for its use.\n\nThe second failure is false equivalence. Entity-neutral analysis does not mean entity-equivalent analysis. Humans, models, scripts, schemas, and institutions can all be described by their functions, but they do not have the same capacities, obligations, or forms of accountability. A model can generate and transform. A human expert can judge, reject, redirect, and accept responsibility. A schema can constrain. A server can enable. These differences matter.\n\nThe third failure is accountability laundering. In weak systems, machine participation can become a convenient fog. People can blame the model for poor work, cite automation as if it were validation, or hide behind “AI assistance” when they have not performed serious review. A good attribution matrix should make this harder. It should expose whether standards were specified, whether outputs were checked, and who authorized the result.\n\nThe fourth failure is institutional flattening. Universities, funders, journals, cultural organizations, and employers may try to reduce this problem to disclosure language. Disclosure is necessary, but it is not enough. “AI was used to assist with drafting” is a crude label. It does not tell us whether the model supplied minor phrasing, generated major sections, classified sources, wrote code, designed a workflow, or operated inside a validated pipeline. A matrix can offer a more accurate record without turning every project into a legal proceeding.\n\nThe fifth failure is invisibility of infrastructure. This may be the most familiar error. Knowledge work has always hidden infrastructure: libraries, databases, software, templates, assistants, administrative systems, funding mechanisms, and technical support. Human-machine work intensifies the problem because infrastructure now shapes cognition more directly. The tools do not merely store the work. They structure the possible actions. They define what can be retrieved, transformed, validated, remembered, versioned, and published.\n\nA serious attribution system should not treat infrastructure as scenery.\n\n## Attribution as Governance\n\nThe deeper value of a functional attribution matrix is not symbolic fairness. It is governance.\n\nA matrix can help collaborators see where work is actually happening. It can help teams identify brittle dependencies. It can show whether a system relies too heavily on unvalidated machine transformation. It can expose whether the human contribution is conceptual, procedural, evaluative, infrastructural, or merely nominal. It can help funders understand why infrastructure work matters. It can help reviewers distinguish disciplined human-machine research from careless automation. It can help workers describe new forms of skill that old job categories do not yet recognize.\n\nIt can also support better project memory. Human-machine workflows are often fragile because important decisions are scattered across chats, prompts, temporary files, model responses, scripts, and undocumented revisions. A matrix does not solve that problem by itself, but it encourages a better habit: record the functions that mattered. Show where the work came from. Show how it was transformed. Show who or what checked it. Show who accepted responsibility.\n\nThis is especially important for research automation and multi-agent knowledge work. As systems become more capable, outputs will become easier to generate and harder to interpret. The central question will shift from whether something can be produced to whether the production process can be trusted. Trust will depend less on the charm of the final artifact and more on the clarity of the activity system behind it.\n\nFunctional attribution is one way to make that system visible.\n\nThe future attribution question is not who touched the artifact last. It is not whether a machine was somewhere in the loop. It is not whether the final prose sounds human. The better question is: how was this work brought into being, through which functions, by which entities, under what constraints, and under whose authority?\n\nThat question will not answer every dispute about credit, authorship, ownership, or responsibility. But it gives us a better starting point. It follows the work rather than the inherited categories. It recognizes that knowledge production is becoming a coordinated activity of humans, machines, infrastructures, and institutions. It refuses both human-only romanticism and machine mystification.\n\nA functional attribution matrix is therefore more than a record. It is a way of seeing the new division of labour.\n",
  "body_text": "## Making the Work Visible\n\nThe first problem in human-machine attribution is conceptual. The second is practical. Once we accept that authorship no longer captures the real division of labour in knowledge work, we need a way to show what actually happened without drowning the work in bureaucracy.\n\nThat is the purpose of a functional attribution matrix.\n\nThe point is not to create another compliance form. The point is to make the structure of contribution visible. Serious human-machine work now involves many participating entities: human experts, collaborators, language models, coding agents, retrieval systems, scripts, datasets, schemas, servers, interfaces, publication templates, and institutional rules. These entities do not contribute in the same way. They do not deserve the same kind of recognition. They do not carry the same responsibility. But they may each perform or enable a function that materially shapes the output.\n\nA useful attribution matrix should therefore avoid the lazy question: Was AI used? That question is already obsolete. It tells us almost nothing. A student who asks a model for a paragraph and an expert who builds a multi-agent research pipeline have both “used AI.” The phrase conceals the difference between shallow delegation and disciplined orchestration.\n\nThe better question is: what functions were performed, by which participating entities, under whose control, according to what standards, and with what form of responsibility?\n\nThat question is harder to answer. It is also closer to the real work.\n\n## Functions Before Names\n\nA functional attribution matrix begins with the activity, not the title page. It asks what had to happen for the output to exist.\n\nSomebody or something originated the problem. Somebody supplied the first conceptual seed. Somebody selected the materials. Somebody structured the task. Somebody or something transformed the inputs. Somebody set the standards. Somebody checked the output. Somebody rejected weak versions. Somebody integrated the parts. Somebody prepared the artifact for circulation. Somebody remains accountable when the work enters the world.\n\nIn older human-only settings, these functions were often compressed into familiar roles: author, editor, assistant, supervisor, developer, designer, publisher. Those roles were never perfect, but they were socially legible. Human-machine work breaks that legibility. The same finished artifact may contain model-generated prose, human conceptual framing, agent-written code, script-enforced formatting, schema-constrained data, and expert validation. A visible paragraph may be machine-generated while the decisive contribution lies in the human-designed conditions that made the paragraph acceptable. A clean report may depend on a server, file structure, validation script, and publication pipeline that no reader ever sees.\n\nThe matrix should therefore track functions such as origin, structuring, transformation, validation, and stewardship.\n\nOrigin concerns the source of problems, concepts, questions, hypotheses, materials, analogies, and aims. Structuring concerns task decomposition, role assignment, sequencing, formats, schemas, and standards. Transformation concerns drafting, coding, classification, synthesis, visualization, editing, and restructuring. Validation concerns checking outputs against evidence, technical requirements, disciplinary standards, aesthetic judgment, or institutional expectations. Stewardship concerns responsibility for publication, reuse, correction, and consequences.\n\nThis does not mean every function deserves the same kind of credit. A script may enforce a format. A model may generate candidate text. A human expert may decide whether the result is admissible. A schema may constrain the output space. A publication system may make the result durable and accessible. These are different contributions. The value of the matrix is that it can show the difference.\n\nA script is not an author. A schema is not a collaborator. A model is not a human colleague in the ordinary social sense. But each can still be part of the activity system. The question is not whether they are persons. The question is what they made possible.\n\n## What the Matrix Should Show\n\nThe first version of a human-machine attribution matrix should be simple enough to use and precise enough to matter. It should resist the temptation to capture everything. A matrix that records every micro-action will fail because no serious worker will maintain it. A matrix that records only “AI used” or “AI not used” will fail because it says almost nothing.\n\nThe useful middle ground is a matrix that identifies substantial contribution by phase and function.\n\nThe phases might include ideation, conceptual seeding, source assembly, system design, infrastructure, task decomposition, processing, analysis, drafting, review, validation, revision, dissemination, and maintenance. The participating entities might include the human lead, human collaborators, machine agents, models, scripts, schemas, datasets, local or cloud infrastructure, publication systems, and institutional requirements. The contribution markers might distinguish primary contribution, substantial contribution, supporting contribution, enabling contribution, validating contribution, and accountable authority.\n\nThis structure would make several important distinctions visible.\n\nIt would distinguish generating from authorizing. A model may generate a draft, but a human expert may authorize the final claim structure. It would distinguish transformation from validation. A coding agent may refactor a script, but a human or test harness may verify that it still works. It would distinguish enabling infrastructure from visible content. A server, template, or schema may not appear in the finished essay, but it may determine whether the work can be repeated, audited, or published. It would distinguish casual prompt assignment from expert-led orchestration. A shallow prompt may produce text. A disciplined harness may produce controlled work.\n\nThe matrix should also show accountability. This is critical. Functional attribution is not a way to distribute responsibility so widely that nobody remains answerable. It should do the opposite. It should clarify where responsibility sits. Machines can participate in production, but humans and institutions still make decisions about use, circulation, publication, and consequence. A model can produce a false claim. A human-led system decides whether that claim survives into a public artifact.\n\nThat means the matrix must include a category for accountable authority. In many cases, this will remain the human lead, principal investigator, editor, publisher, organization, or institution. In some technical systems, accountability may also be distributed across maintainers, infrastructure owners, governance bodies, or project teams. But it should not vanish into the machine.\n\nA functional matrix should make contribution more visible and responsibility harder to evade.\n\n## Risks of Attribution Without Discipline\n\nA matrix can fail in predictable ways.\n\nThe first failure is attribution inflation. Once contribution is widened beyond authorship, everything can start to look like a contributor. The keyboard contributed. The operating system contributed. The coffee contributed. That is not useful. The matrix should focus on contributions that materially shaped the output, governed its quality, enabled its production, or carried responsibility for its use.\n\nThe second failure is false equivalence. Entity-neutral analysis does not mean entity-equivalent analysis. Humans, models, scripts, schemas, and institutions can all be described by their functions, but they do not have the same capacities, obligations, or forms of accountability. A model can generate and transform. A human expert can judge, reject, redirect, and accept responsibility. A schema can constrain. A server can enable. These differences matter.\n\nThe third failure is accountability laundering. In weak systems, machine participation can become a convenient fog. People can blame the model for poor work, cite automation as if it were validation, or hide behind “AI assistance” when they have not performed serious review. A good attribution matrix should make this harder. It should expose whether standards were specified, whether outputs were checked, and who authorized the result.\n\nThe fourth failure is institutional flattening. Universities, funders, journals, cultural organizations, and employers may try to reduce this problem to disclosure language. Disclosure is necessary, but it is not enough. “AI was used to assist with drafting” is a crude label. It does not tell us whether the model supplied minor phrasing, generated major sections, classified sources, wrote code, designed a workflow, or operated inside a validated pipeline. A matrix can offer a more accurate record without turning every project into a legal proceeding.\n\nThe fifth failure is invisibility of infrastructure. This may be the most familiar error. Knowledge work has always hidden infrastructure: libraries, databases, software, templates, assistants, administrative systems, funding mechanisms, and technical support. Human-machine work intensifies the problem because infrastructure now shapes cognition more directly. The tools do not merely store the work. They structure the possible actions. They define what can be retrieved, transformed, validated, remembered, versioned, and published.\n\nA serious attribution system should not treat infrastructure as scenery.\n\n## Attribution as Governance\n\nThe deeper value of a functional attribution matrix is not symbolic fairness. It is governance.\n\nA matrix can help collaborators see where work is actually happening. It can help teams identify brittle dependencies. It can show whether a system relies too heavily on unvalidated machine transformation. It can expose whether the human contribution is conceptual, procedural, evaluative, infrastructural, or merely nominal. It can help funders understand why infrastructure work matters. It can help reviewers distinguish disciplined human-machine research from careless automation. It can help workers describe new forms of skill that old job categories do not yet recognize.\n\nIt can also support better project memory. Human-machine workflows are often fragile because important decisions are scattered across chats, prompts, temporary files, model responses, scripts, and undocumented revisions. A matrix does not solve that problem by itself, but it encourages a better habit: record the functions that mattered. Show where the work came from. Show how it was transformed. Show who or what checked it. Show who accepted responsibility.\n\nThis is especially important for research automation and multi-agent knowledge work. As systems become more capable, outputs will become easier to generate and harder to interpret. The central question will shift from whether something can be produced to whether the production process can be trusted. Trust will depend less on the charm of the final artifact and more on the clarity of the activity system behind it.\n\nFunctional attribution is one way to make that system visible.\n\nThe future attribution question is not who touched the artifact last. It is not whether a machine was somewhere in the loop. It is not whether the final prose sounds human. The better question is: how was this work brought into being, through which functions, by which entities, under what constraints, and under whose authority?\n\nThat question will not answer every dispute about credit, authorship, ownership, or responsibility. But it gives us a better starting point. It follows the work rather than the inherited categories. It recognizes that knowledge production is becoming a coordinated activity of humans, machines, infrastructures, and institutions. It refuses both human-only romanticism and machine mystification.\n\nA functional attribution matrix is therefore more than a record. It is a way of seeing the new division of labour."
}