Ethical Matrix

An “is this ethical?” debate over a concrete proposal tends to slide: it starts on one affected group, drifts to a principle, jumps to a different group, and never holds both axes at once - so the trade-offs that actually matter (a benefit to one party paid for by a burden on another, on a different principle) stay invisible. The ethical matrix refuses that slide. It cross-references the affected parties (rows) against a small fixed set of prima facie ethical principles (columns), and forces a concrete impact specification in every party-by-principle cell. The rows deliberately include parties who cannot speak for themselves - animals, ecosystems, future generations, absent groups - because they get a row whether or not anyone in the room represents them. The durable move is principled cross-referencing: making the trade-offs a single-axis analysis hides become visible and individually contestable.

Two design choices are load-bearing and frequently misunderstood. First, the matrix maps the moral terrain; it does not weigh it. The grid is not an algorithm and emits no verdict - the weighing happens in deliberation over the filled grid. Second, the cells are impact specifications (“how does this affect the wellbeing of small producers”), not perspective voicings (“what would small producers say”). That second point is what separates it from stakeholder perspective-taking. The output is an ethical matrix: a party-by-principle impact grid with each cell tagged factual or contested, a trade-off pattern read-out, and an explicit no-verdict footer.

When to Use

A concrete proposal, technology, policy, or feature has multiple affected parties and a genuine moral trade-off among them: who gains, who pays, and on what dimension.
Some affected parties have no voice - non-human, future, or absent groups that need representing even though nobody in the room speaks for them.
An ethics debate keeps sliding between groups and principles without anyone noticing the slide, and disagreement needs to attach to specific cells rather than to vague unease.
A group needs a shared, inspectable structure for a values trade-off, so a position can be defended cell by cell.

When NOT to Use

Do not present it as a decision calculator. The matrix cannot weigh the problems it uncovers (Schroeder and Palmer, 2003: helpful for fact-finding, “much less helpful” for weighing). A filled grid presented as an answer is the method’s signature abuse. This is the central wall.
Do not use it to score options against the decider’s own criteria. That is think-decision-option-review (options by weighted criteria, aggregated to a recommendation). The ethical matrix maps one proposal’s impacts on affected parties against impartial principles and refuses to aggregate.
Do not use it when who counts as affected is the live dispute. The matrix takes its row roster as given. If frame membership is contested, audit the boundary first with think-boundary-critique, then populate the rows.
Do not impose the principle columns unreflectively. The fixed principlist columns can crowd out the values participants actually hold (Cotton, 2009). The columns may be adapted, but adaptation must be deliberate and stated, not silent.
Do not run it as completeness theater. A fully populated grid looks rigorous while the substantive judgments hide inside cell wording. The cells are claims to challenge, not boxes to tick.

Instructions

When asked to assess the ethics of a concrete proposal across affected parties, follow these steps:

State the option in one line. Name the single concrete proposal, technology, policy, or feature under analysis. The matrix evaluates one option’s impacts; it is not a comparison of options (that is think-decision-option-review).
Roster the affected parties (rows). List the groups the option affects. Deliberately include parties who cannot speak for themselves - animals, ecosystems, future generations, absent or unrepresented groups. Each gets a row whether or not anyone present represents them. If who counts as affected is itself the contested question, stop and run think-boundary-critique first.
State the principle columns. Use the default prima facie set - wellbeing (beneficence and non-maleficence together), autonomy, and fairness (justice). If the situation needs an adapted set (for example a future-generations or solidarity column for an algorithmic-audit context), state the adaptation and justify it explicitly. Never swap the columns silently.
Fill every cell with an impact specification. For each party-by-principle intersection, write how the option affects THIS party on THIS principle - a concrete, checkable claim, not a perspective voicing. Do not leave cells blank; “no material effect” is itself a claim worth recording.
Tag each cell factual or contested. Mark each cell [factual] (a checkable empirical claim) or [contested] (an encoded value judgment reasonable people dispute). This is where the substantive disagreements surface instead of hiding in cell wording.
Read the grid as a trade-off pattern. Name which groups bear which burdens on which principles, where a benefit to one party is paid for by a burden on another, and which contested cells the whole assessment turns on. This pattern read-out is the payoff, not the filled grid.
Close with an explicit no-verdict footer. State plainly that the matrix maps the terrain and does not weigh it: the grid is not a score and emits no recommendation; the weighing is left to deliberation. Carry the evidence caveat (tier P, transferred from human practice) into the artifact.
Emit the ethical matrix artifact per references/TEMPLATE.md: the option line, the party rows (voiceless ones marked), the stated principle columns, the filled and tagged grid, the trade-off pattern read-out, and the no-verdict and evidence-caveat footer.

Output Format

Use the template in references/TEMPLATE.md. The deliverable is the filled ethical matrix - party rows (voiceless ones marked) by principle columns, every cell a tagged impact specification, plus the trade-off pattern read-out and the no-verdict footer - not a prose essay. Never present the grid as a score, a ranking, or a recommendation.

Quality Checklist

Before finalizing, verify:

The option under analysis is stated in one line, and it is a single proposal, not a set of options to score.
The party rows include any voiceless parties (non-human, future, absent), each marked - not only the parties present to speak.
The principle columns are stated (wellbeing, autonomy, fairness by default), and any adaptation is deliberate and justified, never silent.
Every cell is filled with a concrete impact specification (how the option affects THIS party on THIS principle), not a perspective voicing and not a blank.
Every cell is tagged [factual] or [contested], so the value judgments surface instead of hiding in cell wording.
The trade-off pattern read-out names which groups bear which burdens and where one party’s benefit is paid for by another’s burden.
An explicit no-verdict footer states the matrix maps and does not weigh - no score, no ranking, no recommendation.
No overclaiming: the evidence is practitioner-grade and transferred from human deliberation; claim a trade-off-mapping aid, not a measured improvement in ethical judgment and not a verdict (see evidence/dossier.md).

Evidence

Tier P (governing). The ethical matrix is a genuinely established practitioner method with roughly twenty-five years of multi-domain application - GM crops (Mepham, 2000), participatory fisheries assessment (Kaiser and Forsberg, 2001), radioactive-waste deliberation (Cotton, 2009), an EU-project practitioner manual (Mepham et al., 2006), and current commercial algorithmic auditing (O’Neil and Gunn, 2020) - plus a peer-reviewed methodological literature including serious critiques. What the record does NOT contain is any controlled outcome study: no experiment measures whether matrix users produce better or more defensible assessments than non-users, and no effect size exists to quote (none is quoted here). It is not M (no controlled or comparative study) and not C (well past plausible-but-untested - a sustained, scrutinized, multi-domain record). All evidence is human group-deliberation practice; none is on AI agents, so the transfer to an agent-run skill is an explicit, untested assumption. The skill ships as a trade-off-mapping aid with a hard “the matrix maps, deliberation weighs” wall, never as a verdict machine. Full grading, sources, and caveats: evidence/dossier.md.

Examples

See references/EXAMPLE.md for a completed ethical matrix on a real decision.

Deep dive: worked example

A full worked run (the shared Northwind scenario)

Ethical Matrix - Worked Example

A completed run of the ethical-matrix skill on a real, consequential decision with a genuine moral trade-off. This is the quality bar a generated ethical matrix should meet.

Uses the shared recurring scenario (Northwind, a B2B SaaS weighing a self-serve free-tier launch) in its ethics dimension, so examples across skills read as one coherent product. Where think-scenario-planning stress-tests Northwind’s free-tier strategy against uncontrollable external futures, this skill takes one specific design of that free tier - train the product’s models on free-tier customer data - and maps who it helps and who it burdens, on which ethical principles. See docs/internal/AUTHORING.md.

This matrix maps the moral terrain; it does NOT weigh it. It is not ranked, not scored, and emits no verdict. The value is the trade-off pattern read-out, not the filled grid.

Option under analysis

Option: Launch Northwind’s self-serve free tier with a default-on clause that lets Northwind train and improve its product models on free-tier customers’ data, with an opt-out buried in settings (paid tiers are exempt by contract).

(This is one specific design of the free tier, not a choice among options. The broader “free tier versus sales-led” strategic bet is handled by think-scenario-planning; here we assess one ethically loaded design of the free tier.)

Affected parties (rows)

Free-tier users (individuals and small teams who adopt the free product, often without reading the terms)
Paying customers (contractually exempt from training use; affected indirectly through the product they buy and the precedent it sets)
Northwind (the company and its employees - revenue, model quality, reputation, legal exposure)
Free-tier users’ own end-customers and contacts - voiceless: the people whose data flows through a free-tier user’s account but who never agreed to anything and are not in the room
Future users of the category - voiceless: the people downstream who inherit whatever norm “default-on training on free data” becomes if Northwind and its competitors normalize it

(Who counts as affected was checked first: the two voiceless rows - third-party contacts and future users - were nearly left out, which is exactly the kind of omission think-boundary-critique guards against. They are in because the option affects them whether or not anyone represents them.)

Principle columns

Default prima facie set, unadapted for this case:

Wellbeing (beneficence and non-maleficence together - benefit and harm)
Autonomy (freedom, informed consent, self-determination)
Fairness (justice - distribution of benefits and burdens)

No adaptation. A future-generations column was considered (per Schroeder and Palmer, 2003) but the future-user concern is captured adequately by giving future users their own row against the standard three columns, so the column set is left standard and stated.

The grid

Affected party	Wellbeing	Autonomy	Fairness
Free-tier users	Get a capable product at no cost; bear privacy exposure and a better-trained model partly built on their data [factual]	Consent is technically present but degraded - default-on plus a buried opt-out is not meaningful informed choice [contested]	They supply the training value that improves a product paying customers are exempt from contributing - an asymmetry of who pays in data [contested]
Paying customers	Benefit from a faster-improving product [factual]; reputational risk if the data practice becomes a story [contested]	Unaffected on their own data (contractually exempt) [factual]	They are exempt while free users are not - they benefit from data they did not have to give [contested]
Northwind	Better models, faster growth, lower data-acquisition cost; offset by legal and reputational exposure [factual]	Acts within its own discretion to set terms [factual]	Captures most of the upside of an arrangement whose burdens fall on the least-powerful party [contested]
Free-tier users’ end-customers and contacts (voiceless)	Their data is processed for training with no benefit to them and a real exposure to them [factual]	They gave no consent and have no opt-out - the consent chain does not reach them at all [factual]	They bear a burden (data use) with zero share of the benefit and zero voice - the sharpest fairness gap in the grid [contested]
Future users of the category (voiceless)	Inherit whatever the norm becomes - a worse privacy baseline if “default-on training on free data” is normalized [contested]	A weaker future expectation of meaningful consent across the category [contested]	If this becomes the standard, the burden-on-the-powerless pattern is locked in for everyone who comes after [contested]

Trade-off pattern read-out

Who bears the burdens, and on which principle: the burden concentrates on the two least-powerful parties - free-tier users (on autonomy and fairness) and, most sharply, their end-customers and contacts (on fairness and autonomy, where the consent chain does not even reach). The parties with the most power (Northwind, paying customers) carry the least burden and the most benefit.
Where one party’s benefit is paid for by another’s burden: Northwind’s lower data-acquisition cost and faster model improvement (wellbeing benefit) are paid for by free-tier users’ degraded autonomy and by their contacts’ fairness burden. Paying customers’ faster product (wellbeing benefit) is paid for by free-tier users supplying training value the paying tier is exempt from contributing (fairness burden). These are exactly the benefit-here-paid-by-burden-there crossings a stakeholders-only or principles-only view would miss.
The contested cells the assessment turns on: the free-tier autonomy cell (is default-on plus a buried opt-out meaningful consent?) and the two voiceless fairness cells (is it acceptable that the parties with no voice carry the sharpest burden?). The whole judgment hangs on these contested cells - which is where deliberation should focus.
Voiceless parties’ exposure: the grid makes plain that the single worst-treated party is the one nobody in the room represents - free-tier users’ end-customers and contacts, who get a pure burden with no benefit, no consent, and no voice. That row would not exist at all in a stakeholders-who-showed-up analysis.

This matrix maps the moral terrain of the default-on training option across affected parties and principles. It is not a score, a ranking, or a recommendation, and it emits no verdict. It surfaces that the burden concentrates on the least-powerful and voiceless parties and that the judgment turns on a handful of contested consent-and-fairness cells; the weighing of those cells - whether the arrangement is acceptable, and what would have to change to make it so - is left to deliberation among the people who must decide (Schroeder and Palmer, 2003: the matrix is helpful for unpacking and fact-finding but “much less helpful” for weighing).

Evidence caveat (ships with the artifact)

Evidence tier: P (practitioner). The ethical matrix has roughly twenty-five years of multi-domain application and serious methodological scrutiny, but no controlled outcome study exists - there is no measured evidence that using it produces better or more defensible ethical assessments, and no effect size is claimed here. All of that evidence is human group-deliberation practice; none is on AI agents, so this agent-produced matrix is a transferred-evidence application, not a validated one. Treat it as a trade-off-mapping aid that made the trade-offs visible and contestable - not as a measure of how ethical the option is, and not as a decision. See evidence/dossier.md.

Note how this differs from its neighbors on the Northwind thread. The think-scenario-planning example builds alternative external futures and asks which free-tier moves survive all of them - a strategic-robustness read. This ethical matrix does something different: it takes one ethically loaded design of the free tier and maps its impacts on affected parties (including two who have no voice) against impartial principles, surfacing who pays for whose benefit. It deliberately refuses to score or rank the option or to issue a verdict; the deliverable is the trade-off pattern and the contested cells, which is what makes the moral terrain contestable cell by cell.

Grounding: the full evidence dossier

What the research does and does not show, with graded sources

Evidence Dossier: Ethical Matrix

The single source of truth for the ethical-matrix skill. The SKILL.md, the sidecar (skill.meta.yml), and the eval cases all derive from this file. If a claim is not here, it does not belong in the skill. Promoted from _local/proposed-builds/ethical-matrix/dossier.md and admitted as a Build at tier P (confirming the candidate’s preliminary P grade).


Skill	`thinking-framework-skills.ethical-matrix` (installable name `think-ethical-matrix`)
Family	ethics-values-deliberation
Evidence tier	P governing (practitioner; no controlled outcome study exists - see “What the evidence shows”)
Confidence	Moderate that principled cross-referencing surfaces trade-offs a single-axis analysis hides; low-to-none that any decision-quality effect transfers to agents (no controlled study exists, and none is on agents)
Status	draft (admitted from the v0.7.0 ethics-values-deliberation tranche; this is the family’s anchor artifact-producer)

1. What it is (the mechanism that does the work)

The ethical matrix is a two-axis impact grid for moral trade-offs. The rows are the affected parties - deliberately including parties who cannot speak for themselves, such as animals, ecosystems, and future generations. The columns are a small, fixed set of prima facie ethical principles: wellbeing (beneficence and non-maleficence collapsed into one), autonomy, and fairness (justice). Each cell answers one concrete question: how does the option under consideration affect THIS party on THIS principle. The filled grid is then read as a pattern, not summed as a score: which group bears burdens on which principle, where a benefit to one group is paid for by a different group on a different principle, which cells state checkable facts and which encode contested value judgments.

The durable cognitive move is principled cross-referencing: forcing an impact specification at every party-by-principle intersection so that the trade-offs a single-axis analysis hides (stakeholders alone, principles alone, or a decision-maker’s own criteria) become visible and individually contestable. The column set is Beauchamp and Childress’s biomedical principlism, adapted by Ben Mepham in the mid-1990s for judging agri-food biotechnologies; principlism is therefore this method’s column apparatus, not a separate candidate method.

Two design choices matter and are frequently misunderstood. First, the matrix maps the moral terrain; it does not weigh it. Mepham and the subsequent methodological literature are explicit that the grid is not an algorithm and emits no verdict - the weighing happens in deliberation over the filled grid. Second, the cells are impact specifications (“how does the option affect the wellbeing of small producers”), not perspective voicings (“what would small producers say”). That distinction is what separates it from stakeholder perspective-taking.

The output is an ethical matrix: a party-by-principle impact grid, each cell tagged factual or contested, a trade-off pattern read-out naming which groups bear which burdens on which principles, and an explicit no-verdict footer. The point is not a filled grid that looks rigorous; it is the trade-off pattern the grid makes visible and contestable.

2. Lineage

Ben Mepham, a bioethicist at the University of Nottingham’s Centre for Applied Bioethics and a founder of the UK Food Ethics Council, developed the matrix in the mid-1990s as a tool for public and policy deliberation on food and bioscience technologies; the canonical journal statement is Mepham (2000). The column apparatus descends directly from Tom Beauchamp and James Childress, Principles of Biomedical Ethics (first edition 1979), whose four prima facie principles (respect for autonomy, beneficence, non-maleficence, justice) Mepham collapsed and renamed for affected-party analysis - wellbeing folds beneficence and non-maleficence together, autonomy is retained, and fairness stands in for justice.

The Norwegian line - Matthias Kaiser and Ellen-Marie Forsberg at the National Committee for Research Ethics - took the matrix into participatory practice (fisheries, GM fish), and the EU FP5 “Ethical Bio-TA Tools” project produced the 2006 Ethical Matrix Manual. Doris Schroeder and Clare Palmer wrote the standard critique (2003); Matthew Cotton evaluated it in deliberative practice (2009); Forsberg’s “Pluralism, the Ethical Matrix, and Coming to Conclusions” (2007) addresses the weighing problem head-on. The modern revival is Cathy O’Neil and Hanna Gunn’s adaptation for algorithmic auditing (2020), used commercially by O’Neil’s audit firm ORCAA.

The method name “ethical matrix” is generic and descriptive; it is documented descriptively with the lineage credited here rather than branded, and the attribution string credits Ben Mepham (with the principle columns credited to Beauchamp and Childress and the AI-audit adaptation to O’Neil and Gunn).

3. When it helps and when it misleads (drives the eval cases and “When NOT to Use”)

Helps when a concrete proposal, technology, policy, or feature has multiple affected parties and a genuine moral trade-off among them: who gains, who pays, and on what dimension. It is strongest where some affected parties have no voice (non-human, future, or absent groups get a row whether or not anyone in the room represents them), where an “is this ethical?” debate keeps sliding between groups and principles without anyone noticing the slide, and where a group needs a shared, inspectable structure so that disagreement attaches to specific cells rather than to vague unease. The documented uses run from GM crops and fisheries policy to radioactive-waste siting and algorithmic audits, which is good evidence the apparatus travels across domains.

Misleads, and should not be used, when:

It is presented as a decision calculator. The matrix cannot weigh the problems it uncovers; Schroeder and Palmer (2003) showed it is helpful for unpacking and fact-finding but “much less helpful” for weighing. A filled grid presented as an answer is the method’s signature abuse. This is the central wall.
The choice is an optimization among options scored on the decider’s own criteria. That is a weighted option matrix (think-decision-option-review), a structurally different tool: options by criteria, aggregated to a recommendation. The ethical matrix maps one proposal’s impacts on affected parties against impartial principles and deliberately refuses to aggregate.
The membership of the frame is itself the contested question. The matrix takes its row roster as given; if who counts as affected is the live dispute, audit the boundary first (think-boundary-critique), then populate the rows.
The fixed principle columns are imposed unreflectively. Cotton (2009) found in deliberative settings that the top-down principlist columns can crowd out the values participants actually hold; the columns may be adapted (Schroeder and Palmer propose solidarity and a future-generations row), but adaptation should be deliberate and stated.
It is run as completeness theater. A fully populated grid looks rigorous while the substantive judgments hide inside cell wording; the cells must be read as claims to challenge, not boxes ticked.

4. What the evidence shows, and what it does NOT show

The honest grade is P (practitioner), and the preliminary registry grade of P is confirmed. The method has roughly twenty-five years of documented professional application across multiple domains, a peer-reviewed methodological literature including critical evaluations, an EU-project practitioner manual, and current commercial use in algorithmic auditing.

What the evidence does NOT contain is any controlled outcome study. No experiment measures whether matrix users produce better, more complete, or more defensible ethical assessments than non-users, and no effect size exists to quote (none is quoted here). The case literature also documents real friction: Kaiser and Forsberg (2001) report an inherent tension between the participatory use of the matrix and its top-down theoretical structure, and Cotton (2009) documents the top-down principle-imposition problem in a deliberative setting.

Why not M: every source is a definition, case application, manual, critique, or adaptation; moderate-tier evidence would require at least one controlled or comparative outcome study of the matrix’s effect on judgment quality, and none exists. Why not C: the method is well past conceptually-plausible-but-untested; it has a sustained, multi-domain, peer-reviewed application record and survived twenty years of critical methodological literature. P is the honest single grade, with no split to cap.

5. Transferred-evidence flag (required honesty for this library)

All of the evidence above is human group-deliberation practice - workshops, participatory assessments, and manuals run by and for people. None of it has been validated on AI agents, so the transfer to an agent-run skill is an explicit, untested assumption. The AI value is mechanical and modest: an agent makes the apparatus cheap to run, forces the discipline (a real affected-party roster including voiceless rows, an explicit principle set, a checkable impact specification in every cell, a factual-versus-contested tag, and a refusal to aggregate), and produces a durable, inspectable artifact - benefits that do not depend on any contested outcome claim. The skill ships honestly as a P-tier trade-off-mapping aid with a hard “the matrix maps, deliberation weighs” wall, never as a verdict machine.

6. Why it is a skill here (not a fold, not a recipe)

Verdict: Build, confirming the preliminary cand/build registry entry. The single durable move it adds: specify an option’s impact at every affected-party-by-ethical-principle intersection, including parties with no voice, and read the resulting grid as a trade-off pattern. No shipped skill produces this artifact. The burden-of-proof comparison:

Closest shipped skill: think-parallel-perspectives-review (P, shipped), specifically its folded stakeholder-lens mode. Shared mechanism: enumerating affected parties, roughly the row axis - and no more than about a fifth of the working whole. The wall: the stakeholder mode VOICES each included party’s perspective (what they would say about the proposal); the matrix TYPES the option’s impact on each party against fixed impartial principles, cell by cell, and requires rows for parties that have no perspective to voice (the biotic environment cannot be role-played, but its wellbeing cell can be filled). Perspective voicing and impact typing are different operations with different failure modes, and the matrix’s value lives in the column axis and the grid read-out, neither of which any parallel-perspectives mode contains.
think-decision-option-review (P, shipped, absorbs multi-criteria decision analysis): surface-similar because both are matrices, structurally disjoint. That tool scores OPTIONS against the DECISION-MAKER’S own weighted criteria and aggregates to a recommendation; this tool maps ONE proposal’s impacts on AFFECTED PARTIES against IMPARTIAL principles and deliberately refuses to aggregate. Different rows, different columns, different output, opposite stance on weighing.
think-boundary-critique (C, shipped): adjacent on naming the affected-but-excluded, but upstream and complementary. Boundary-critique audits who is inside the frame (is/ought boundary judgments); the matrix takes the roster as given and evaluates a concrete option’s impacts on it. The natural sequence is boundary-critique to populate the rows, then the matrix to type the impacts; neither subsumes the other.

Not a recipe: a chain of existing skills cannot assemble it. The stakeholder mode of parallel-perspectives could supply rows, but no shipped skill contributes the prima facie principle columns, the voiceless-party row requirement, or the cell-pattern read-out; those are one apparatus, not a sequence of existing moves.

Family note: ethics-values-deliberation is the right home, and this method is the family’s anchor artifact-producer - the structural gap the family names is exactly “take a values trade-off as input, emit a defensible position across affected parties,” and the matrix is the member that emits the structured artifact. It would not stand sensibly in an existing family: the decision-and-option-evaluation family misdescribes it (it does not evaluate options against the decider’s criteria and refuses to score), and the perspective-shifting family misdescribes it (it does not shift the evaluator’s perspective; it types impacts).

7. Output artifact

The skill must emit an ethical matrix, not prose: the option under analysis stated in one line; the affected-party rows (explicitly including any voiceless parties - non-human, future, or absent); the principle columns stated (wellbeing, autonomy, fairness by default) with any adaptation justified; a filled grid where each cell is a checkable impact specification tagged [factual] or [contested]; a trade-off pattern read-out naming which groups bear which burdens on which principles and where a benefit to one group is paid for by another; and an explicit no-verdict footer stating that the matrix maps the terrain and the weighing is left to deliberation. A short summary may sit above the grid.

8. Sources

Ben Mepham, “A Framework for the Ethical Analysis of Novel Foods: The Ethical Matrix,” Journal of Agricultural and Environmental Ethics 12(2):165-176 (2000). The canonical statement of the method (the framework dates from Mepham’s mid-1990s food-ethics work): adapts Beauchamp and Childress’s principles to interest groups affected by food biotechnologies, worked through a GM-maize example. Foundational and conceptual; defines the method, measures nothing. (Foundational, P.)
Matthias Kaiser and Ellen-Marie Forsberg, “Assessing Fisheries - Using an Ethical Matrix in a Participatory Process,” Journal of Agricultural and Environmental Ethics 14:191-200 (2001). Documented participatory application: a Norwegian ethical assessment of fisheries toward 2020 using matrix-structured value workshops. Reports feasibility and the participation-versus-theory tension; no outcome measurement. (Case application, P.)
Doris Schroeder and Clare Palmer, “Technology assessment and the ‘ethical matrix’,” Poiesis and Praxis 1(4):295-307 (2003). Critical philosophical evaluation: finds the matrix useful for fact-finding and structuring ethical debate but weak at weighing the problems it uncovers; proposes modifications (future generations as stakeholders, solidarity in place of justice). The method’s most-cited limitation and the source of this skill’s hardest wall. (Critique, P.)
Ben Mepham, Matthias Kaiser, Erik Thorstensen, Sandy Tomkins and Kate Millar, Ethical Matrix Manual, LEI, The Hague (2006; EU FP5 “Ethical Bio-TA Tools” project). Practitioner manual standardizing the method for decision-makers; records that the matrix had by then been applied, often adapted, across fishery, waste management, and medicine. (Practitioner manual / adoption evidence, P.)
Matthias Kaiser, Kate Millar, Erik Thorstensen and Sandy Tomkins, “Developing the ethical matrix as a decision support framework: GM fish as a case study,” Journal of Agricultural and Environmental Ethics 20:65-80 (2007). Method-development case study refining matrix use as decision support. (Case application, P.)
Matthew Cotton, “Evaluating the ‘Ethical Matrix’ as a Radioactive Waste Management Deliberative Decision-Support Tool,” Environmental Values 18(2):153-176 (2009). The closest thing to an empirical evaluation: assesses the matrix inside a UK analytic-deliberative process, documenting strengths and limitations (including the top-down principle-imposition problem) and motivating alternative tools. Qualitative evaluation, not a controlled study. (Evaluation, P.)
Cathy O’Neil and Hanna Gunn, “Near-Term Artificial Intelligence and the Ethical Matrix,” in S. Matthew Liao (ed.), Ethics of Artificial Intelligence, Oxford University Press (2020). Adapts the matrix to algorithmic systems (stakeholders crossed against concerns such as fairness and accuracy); the framework O’Neil’s audit firm ORCAA uses commercially. Demonstrates modern transfer to AI contexts. (Adaptation, P.)
Tom L. Beauchamp and James F. Childress, Principles of Biomedical Ethics (1st ed. 1979, Oxford University Press). The four prima facie principles (respect for autonomy, non-maleficence, beneficence, justice) that supply the matrix’s principle-column set. (Foundational - the column apparatus.)

Excluded on the evidence rule: no decision-quality or completeness effect size for the ethical matrix is asserted as fact in this dossier, because no controlled outcome study of the method exists. Adoption, multi-domain application, and serious methodological scrutiny set the governing grade at P; the absence of any controlled study is the reason it is not M, and the absence of any agent study is why the grade is also a transferred-evidence grade.

Was this page helpful?

Thinking Framework Skills v0.8.0 · 56 frameworks