Democratic AI is Possible. The Democracy Levels Framework Shows How It Might Work

Amy X. Zhang; Andrew Konya; Atoosa Kasirzadeh; Aviv Ovadya; Bilva Chandra; Flynn Devine; K. J. Kevin Feng; Kyle Redman; Luke Thorburn; Manon Revel

REVIEW 3 major objections 3 minor 1 cited by

Democratic AI is Possible. The Democracy Levels Framework Shows How It Might Work

T0 review · 3 major / 3 minor · reviewed 2026-08-12 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read The paper argues that democratizing AI means transferring real decision-making power through six defined levels, and provides tools to plan and evaluate such transfers.

desk verdict The Democracy Levels framework is a genuinely useful shared vocabulary for AI governance, but it carries two unsupported empirical claims and an internally inconsistent Meta Oversight Board example that need fixing before it becomes the standard reference. read the letter →

arxiv 2411.09222 v4 pith:XVFHNT3E submitted 2024-11-14 cs.CY

Aviv Ovadya , Kyle Redman , Luke Thorburn , Quan Ze Chen , Oliver Smith , Flynn Devine , Andrew Konya , Smitha Milli

show 6 more authors

Manon Revel K. J. Kevin Feng Amy X. Zhang Bilva Chandra Michiel A. Bakker Atoosa Kasirzadeh

This is my paper · ORCID

classification cs.CY

keywords democraticAIgovernancedemocracylevelsframeworkcollectivedecision-makingparticipatoryalignmentdeliberativepublicparticipation

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This position paper argues that 'democratizing AI' should mean giving democratic processes real decision-making power over how AI is built, aligned, and governed, not merely making AI open or widely accessible. It introduces the Democracy Levels framework, a six-level scale from L0 (unilateral decision-making) to L5 (democratic metagovernance), which classifies how much power has been transferred from a unilateral authority to a democratic system. The paper also supplies tools — the Levels Decision Tool and the Democratic System Card — to help organizations plan such transfers and to help outsiders evaluate whether claims of democratic participation are honest. A sympathetic reader would care because the framework offers a concrete, shared language for a debate that is currently dominated by vague appeals to 'democratization.'

What carries the argument

The central machinery is the Democracy Levels framework itself, modeled on the autonomy levels for self-driving cars. It pairs the six levels with a set of dimensions — process quality (representation, informedness, deliberation, substantiveness, robustness, legibility), delegation (integration, ability to bind, commitment), and trust (awareness, participation, accountability, buy-in) — that together determine whether a democratic system is good enough to be entrusted with a given level of power. Two tools operationalize the framework: the Levels Decision Tool, a set of questions for deciding how much democracy a decision warrants, and the Democratic System Card, a structured template for evaluating a democratic system against the dimensions.

What would settle it

Compare the documented costs of running a binding democratic process for a specific AI governance decision against the costs of the regulatory fines, antitrust remedies, or market losses that a similar unilateral decision would incur; if democratic processes are not substantially cheaper in several critical areas, the corporate-adoption argument in the paper loses its ground.

Watch

Extended reading notes

Core claim

The paper's central claim is that democratizing AI is not a single binary act but a gradable transfer of decision-making power, and that the Democracy Levels framework captures that gradient. Each level is defined by which of five roles a democratic system plays: informing decisions (L1), specifying options (L2), making binding decisions (L3), initiating binding processes automatically (L4), and exercising metagovernance over when and how democratic processes are used (L5). At L5 the unilateral authority has fully shifted power within a domain to an adaptive constitutional order. The framework is meant to guide organizations, regulators, and AI systems toward meaningful public control and to allow outsiders to evaluate how much power has actually been transferred.

Load-bearing premise

The load-bearing assumption is that democratic processes will be cheaper for corporations than reactive compliance with regulation, forced reorganization under antitrust action, or loss of market value from eroded public trust; the paper asserts this cost comparison without providing evidence or a cost model.

Editorial extensions

If this is right

AI organizations, regulators, and AI systems can each be scored on the Democracy Levels scale, making it possible to compare the democraticness of different governance arrangements in a shared language.
A unilateral authority can use the Levels Decision Tool to decide which decisions are worth delegating and to what level, weighing legitimacy, collective intelligence, feasibility, speed, and resources.
The Democratic System Card lets evaluators judge whether a process is good enough for a given level, covering process quality, delegation, and trust, so that promising processes can be improved before power is delegated further.
Current efforts are mapped: Anthropic's Collective Constitutional AI sits at L1 (informing decisions), while Meta's Oversight Board shows L4 for individual content decisions but L1 for policy, demonstrating the framework's applicability.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

The framework could be generalized beyond AI to the governance of any powerful technology, since the five roles (inform, specify, decide, trigger, metagovern) are not specific to machine learning.
The cost assumption in Section 2.2 is empirically testable: a study comparing the cost of a citizen assembly or deliberative poll with the cost of a regulatory fine, antitrust remedy, or trust collapse for a comparable decision would either support or undermine the corporate-adoption argument.
The levels implicitly assume that power transfer is the main axis of democratization; a complementary metric would track the protection of rights and pluralism, dimensions the authors acknowledge but leave to other frameworks.
L5 (metagovernance under democratic control) is essentially a constitutional-order design problem; work on constitutional design and credible commitments could offer concrete implementation paths.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Here's the short version: the Democracy Levels framework is worth taking seriously. The five-role ladder—informing, specifying, binding, initiating, metagovernance—is a real improvement over Arnstein and IAP2 because it separates how much authority is transferred from how well the receiving process works. The paper is clear about what it claims, covers related work honestly, and the Democratic System Card is a practical artifact people can actually use. If you work on AI governance or participatory alignment, this is one of the better shared vocabularies around.

The soft spots are real but mostly fixable. Section 2.2 asserts two empirical claims without evidence: that democratic processes are substantially cheaper than regulatory compliance or trust erosion, and that democracy is an "asymmetric enabler" favoring good actors. These are load-bearing for the corporate adoption argument, and they're stated as facts. They should be labeled speculative or supported by cost models and case studies. A position paper can advance a thesis, but not with an unsourced cost comparison.

The more interesting problem is the Meta Oversight Board example in Section 3.4. The paper defines levels as transfers of power to a democratic system, then classifies the Oversight Board as L4—regular binding decisions—while admitting the Board wasn't designed to be democratically representative. The dimensions table catches the representativeness gap, but the level assignment still tracks bindingness and regularity, not whether a legitimate democratic body holds the power. That's an internal inconsistency, and it makes the framework vulnerable to democracy washing, where a company can point to an L4 score while avoiding any real democratic control. The framework can absorb this: if the Oversight Board is instead described as a non-democratic example of delegated binding authority, or the levels are explicitly split into "delegation levels" and "democratic quality," the problem goes away. But as written, the headline example undercuts the paper's own definition.

Lesser issues: no inter-rater reliability for level assignments, and the self-citations are used as background support, which is fine. The framework doesn't derive anything mathematically, but it doesn't pretend to; it's a measurement and communication tool.

This paper deserves a serious referee. The right outcome is not rejection but revision: fix the Meta example, hedge the two empirical claims, and the framework is ready to be cited widely.

Referee Report

3 major / 3 minor

Summary. This position paper argues that meaningfully 'democratizing AI' is possible and valuable, and it introduces a 'Democracy Levels' framework (L0-L5) for classifying how much decision-making power has been transferred from a unilateral authority to a democratic system. The framework is accompanied by a set of quality dimensions (process quality, delegation, trust), a Levels Decision Tool, and a Democratic System Card for evaluating specific processes. The paper applies the framework to early industry experiments such as Anthropic's Collective Constitutional AI and Meta's Oversight Board and Community Forums, and it includes a Q&A rebutting common critiques. The central claim is that the framework provides a shared language and evaluation device for guiding and auditing democratic AI efforts.

Significance. If the framework holds up, it would make a useful contribution by converting a vague aspiration into a concrete, falsifiable ladder and an evaluation instrument. The paper is commendably self-aware: it stresses that levels alone do not establish democratic quality, provides worked example system cards, and engages honestly with critiques. There are no hidden dependencies on contested empirical results in the core taxonomy; the framework is definitional rather than derived, so circularity is not a concern. The main risk is that the L-level classification will be used in practice as a democracy label despite the paper's own qualifications, a risk amplified by the Meta Oversight Board example.

major comments (3)

[Section 3.4; Figure 2; Section 3.2] The application of the framework to Meta's Oversight Board is internally inconsistent. Section 3.4 assigns the Board's content-moderation decisions to L4 ('regular binding decisions'), yet the same paragraph observes that 'the Oversight Board was not designed to be democratically representative.' The Levels are defined in Section 3.2 as the roles performed by 'democratic systems rather than a unilateral authority,' and Figure 2's caption says they assess power transferred 'to a democratic process.' By the paper's own definition in Section 3.1, a body that is not democratically representative need not be a democratic system. Classifying the Board as L4 therefore makes the Levels track bindingness and regularity rather than transfer to democratic control, contradicting the caption. Since the Levels are the framework's headline construct, this conflation is load-bearing and invites 'democracy washing' if adopted by practitioners. Please either reclassify the Board (e.g., as not classifiable at a level until it is democratically constituted) or redefine the Levels as measuring bindingness of public input, with democraticness evaluated solely through the Dimensions.
[Section 2.2] The corporate-adoption argument rests on an unquantified empirical claim: that 'the costs of using democratic systems for several critical areas of decision-making are substantially lower than reactive compliance with regulation, forced reorganization under antitrust action, or loss of market value due to eroded public trust.' No cost model, data, pilot results, or citation is supplied for this comparison. As written, the claim is unfalsifiable and it is doing load-bearing work: the reasons why corporations should voluntarily adopt the framework (Section 4.1) depend on it. The subsequent 'asymmetric enabler' assertion is explicitly a belief ('we believe'), which is acceptable in a position paper, but the cost claim should either be evidenced or reframed as a hypothesis requiring empirical test.
[Section 4.2; Table 1; Appendix C] The framework's evaluation function is underspecified. The Democratic System Card asks evaluators to provide a 'qualitative evaluation of the highest level of power that the system can be trusted with for making decisions (for a given context),' but neither Section 4.2 nor the card in Appendix C specifies how the answers to the guiding questions across the three dimensions should be combined into that evaluation. Without an aggregation rule or a stated threshold, two evaluators can reach incompatible 'trusted level' judgments without any way to adjudicate. This limits the accountability use case claimed in the abstract ('support evaluation of such efforts'), even if it may be acceptable for a position paper.

minor comments (3)

[Appendix B] The Levels Decision Tool lists the question 'No clear expert consensus?' twice under 'To what extent does the decision involve:', with the same arrows both times; the duplicate should be removed.
[Table 1] Under the Buy-in dimension, the second subquestion reads 'To what extent are the relevant public and key stakeholders accepting of the legitimacy of: (1) the system/process? (2) of the decision?' The phrase 'of the decision' is grammatically inconsistent with 'of the legitimacy of'; it should read '(2) the decision?'.
[Section 3.3] The text has a typo: 'threedimensions' should be 'three dimensions'.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper offers a definitional framework and illustrative applications, not a derivation whose outputs reduce to its inputs.

full rationale

This is a position paper that introduces the Democracy Levels framework as a taxonomy of decision-making roles (informing, specifying, binding, auto-initiating, metagovernance) plus evaluation dimensions. There are no equations, fitted parameters, or quantitative predictions, so the core circularity patterns (self-definitional equations, fitted inputs renamed as predictions, uniqueness theorems imported from authorial prior work) do not apply. The framework's levels are stipulated definitions, and the examples (Anthropic's Collective Constitutional AI, Meta's Oversight Board, Community Forums) are illustrative classifications rather than derived results. The paper's self-citations (e.g., Ovadya 2023a for democratic infrastructure providers, Konya et al. 2023 for collective dialogue processes) are background support for existing deliberative practices and do not carry the load of the framework's central claim. The unsupported cost claim in Section 2.2 is an empirical weakness, not a circularity, since the claim is not used to construct the framework. The internal tension in Section 3.4 — classifying Meta's Oversight Board as L4 while noting it was not designed to be democratically representative — is a consistency/correctness concern about how the levels are applied, but it does not make a prediction reduce to an input by construction. The framework is self-contained as a proposed vocabulary and evaluation tool, so the appropriate circularity score is 0.

Assumptions & free parameters 0 free parameters · 4 assumptions · 0 invented entities

The framework is conceptual; it has no numerical parameters. It rests on normative claims about the value of democratic process and on assumptions about delegation feasibility. The five-role taxonomy is the paper's main added structure.

assumptions (4)

ad hoc to paper Decision-making power can be classified into the five roles of informing, specifying, binding, initiating, and metagovernance.
Section 3.2 defines levels solely in terms of these roles; no independent justification is offered for why these five are the salient ones.
domain assumption Contemporary deliberative processes (sortition-based citizen assemblies) are the pragmatic template for AI governance.
Section 2.3 states examples 'predominantly draw on contemporary deliberative democratic processes' and assumes they work across jurisdictions and power scales.
domain assumption AI organizations can technically and legally bind themselves to democratic decisions.
Section 3.3 includes 'Ability to bind' as a delegation dimension, but only one of the two worked examples (the hypothetical UK assembly) discusses legal limits; no real case of binding is demonstrated.
domain assumption Democratic processes produce better decisions for AI than unilateral authority.
Section 2.2 asserts epistemic benefits of democracy citing Estlund and Landemore, but the framework depends on this to justify the entire project.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Democratic AI is Possible. The Democracy Levels Framework Shows How It Might Work." pith.science (2026). https://pith.science/paper/XVFHNT3E

@misc{pith2026241109222,
  author       = {Pith},
  title        = {Pith review of: Democratic AI is Possible. The Democracy Levels Framework Shows How It Might Work},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XVFHNT3E}},
  note         = {Machine review of arXiv:2411.09222}
}

read the original abstract

This position paper argues that effectively "democratizing AI" requires democratic governance and alignment of AI, and that this is particularly valuable for decisions with systemic societal impacts. Initial steps -- such as Meta's Community Forums and Anthropic's Collective Constitutional AI -- have illustrated a promising direction, where democratic processes could be used to meaningfully improve public involvement and trust in critical decisions. To more concretely explore what increasingly democratic AI might look like, we provide a "Democracy Levels" framework and associated tools that: (i) define milestones toward meaningfully democratic AI, which is also crucial for substantively pluralistic, human-centered, participatory, and public-interest AI, (ii) can help guide organizations seeking to increase the legitimacy of their decisions on difficult AI governance and alignment questions, and (iii) support the evaluation of such efforts.

Figures

Figures reproduced from arXiv: 2411.09222 by the authors.

**Figure 1.** A system diagram of how democratic processes could integrate with the AI ecosystem, with democratic infrastructure being used to facilitate—where appropriate—collective decisions relating to AI regulation, organizational governance, and alignment. The Democracy Levels Framework we introduce can be used to evaluate (i) the degree to which democratic systems are used for decision-making, and (ii) the quality of those … view at source ↗

**Figure 2.** Overview of the Democracy Levels (names of each in bold), which are used to assess how much decision-making power in a given domain of decision-making has been transferred from a unilateral authority to a democratic process. The example column describes (hypothetical) democratic systems operating at each level, with the scope of authority being rules around the use of AI systems for persuasion. Evaluating the ultima… view at source ↗

**Figure 3.** A diagram illustrating how the Levels Decision Tool and Democratic System Card can be used to assist stakeholders with planning the transfer of decision-making power and evaluation of democratic systems to adopt within their specific decision-making context. 17 [PITH_FULL_IMAGE:figures/full_fig_p017_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

Aggregated Individual Reporting for Post-Deployment Evaluation
cs.CY 2025-06 conditional novelty 6.0 of 10

The authors formalize a mechanism for collecting and aggregating public reports about deployed AI systems, aiming to surface unknown harms and enable accountability.

Reference graph

Works this paper leans on

15 extracted references · 8 canonical work pages · cited by 1 Pith paper

[1]

Describe the process or system at a high level

work page
[2]

Conitzer, V ., Freedman, R., Heitzig, J., Holliday, W

URL https://doi.org/10.1093/oso/ 9780195136593.003.0002. Conitzer, V ., Freedman, R., Heitzig, J., Holliday, W. H., Jacobs, B. M., Lambert, N., Moss´e, M., Pacuit, E., Rus- sell, S., Schoelkopf, H., Tewolde, E., and Zwicker, W. S. Position: social choice should guide ai alignment in deal- ing with diverse human feedback. In Proceedings of the 41st Interna...

arXiv
[3]

Then, for each dimension/property defined on the card below, describe how that dimension works in the process or system you are evaluating

work page
[4]

democratic- ness

Use the guiding questions to reflect on the extent to which that dimension is satisfied. You can find the most up-to-date version of the Democratic System Card, including editable templates that can be initially filled with an LLM (given context) at: democracylevels.org/system-card. Context Describe: Describe the process or a system at a high level. (Can ...

work page 2012
[5]

Democratizing AI

URL https://proceedings.neurips. cc/paper_files/paper/2024/file/ 9328208f88ec69420031647e6ff97727- Paper-Conference.pdf. Goldberg, B., Acosta-Navas, D., Bakker, M., Beacock, I., Botvinick, M., Buch, P., DiResta, R., Donthi, N., Fast, N., Iyer, R., Jalan, Z., Konya, A., Danciu, G. K., Lande- more, H., Marwick, A., Miller, C., Ovadya, A., Saltz, E., Schirch...

arXiv 2024
[6]

Accessed: 2024-09-04

URL https://iap2.org.au/resources/ spectrum/. Accessed: 2024-09-04. Involve. Involve, 2024. URL https://www.involve. org.uk/. Democratic Infrastructure Provider. Kasirzadeh, A. Plurality of Value Pluralism and AI value alignment. In Pluralistic Alignment Workshop at NeurIPS 2024, 2024. Kasirzadeh, A. and Gabriel, I. Characterizing AI Agents for Alignment ...

arXiv 2024
[9]

URL https://www.science.org/doi/abs/10

doi: 10.1126/science.adr6713. URL https://www.science.org/doi/abs/10. 1126/science.adr6713. Summerfield, C., Argyle, L., Bakker, M., Collins, T., Durmus, E., Eloundou, T., Gabriel, I., Ganguli, D., Hackenburg, K., Hadfield, G., et al. How will ad- vanced AI systems impact democracy? arXiv preprint arXiv:2409.06729, 2024. Suresh, H., Tseng, E., Young, M., ...

arXiv
[13]

Summarize what other systems the process depends on or interacts with, which impact its success (e.g., sortition data, or user or citizen authentication systems)

work page

Show all 15 references

[1969]

URL https: //doi.org/10.1080/01944366908977225

doi: 10.1080/01944366908977225. URL https: //doi.org/10.1080/01944366908977225. Auernhammer, J. Human-centered AI: The role of Human- centered Design Research in the development of AI. In Proceedings of Synergy - DRS International Conference 2020, 2020. Bai, Y ., Kadavath, S.,...

arXiv
[2014]

Urbinati, N

ISBN 9781137357809. Urbinati, N. Representative Democracy: Principles and Genealogy. University of Chicago Press, 2006. 15 Democratic AI is Possible. The Democracy Levels Framework Shows How It Might Work. U.S.-China Economic and Security Review Commission. 2024 annual report,...

2006
[2015]

process legitimacy

URL https://api.semanticscholar. org/CorpusID:155508734. Wong, J., Morgan, D., Straub, V . J., Hashem, Y ., and Bright, J. Key challenges for the participatory gover- nance of AI in public administration. SocArXiv Papers: https://osf.io/preprints/socarxiv/pdcrm, 2022. Young, I...

2022
[2020]

URL https://www.oecd-ilibrary.org/ content/publication/339306da-en

doi: https://doi.org/10.1787/339306da-en. URL https://www.oecd-ilibrary.org/ content/publication/339306da-en. OpenAI. GPT-4 System Card, 2024a. URL https://cdn.openai.com/papers/gpt- 4-system-card.pdf. Accessed: 2025-05-18. OpenAI. Model Spec, 2024b. URL https: //cdn.openai.co...

2023 arXiv
[2022]

democratizing AI

URL https://api.semanticscholar. org/CorpusID:251434428. Lijphart, A. Patterns of Democracy: Government Forms and Performance in Thirty-Six Countries. Yale University Press, 2nd edition, 2012. Lin, T.-a. “democratizing AI” and the Concern of Algo- rithmic Injustice. Philosophy...

2012
[2023]

Accessed: January 30, 2025

URL https://a16z.com/the-techno- optimist-manifesto/. Accessed: January 30, 2025. Anthropic. Collective Constitutional AI: Aligning a Language Model with Public Input, 2023. URL https: //www.anthropic.com/news/collective- constitutional-ai-aligning-a- language-model-with-publi...

2025
[2024]

edu/publication/meta-community-forum- results-analysis

URL https://cddrl.fsi.stanford. edu/publication/meta-community-forum- results-analysis. Christiano, T. and Bajaj, S. Democracy. In Zalta, E. N. and Nodelman, U. (eds.), The Stanford Ency- clopedia of Philosophy . Metaphysics Research Lab, Stanford University, summer 2024 editi...

2024

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.

[1] [1]

Describe the process or system at a high level

work page

[2] [2]

Conitzer, V ., Freedman, R., Heitzig, J., Holliday, W

URL https://doi.org/10.1093/oso/ 9780195136593.003.0002. Conitzer, V ., Freedman, R., Heitzig, J., Holliday, W. H., Jacobs, B. M., Lambert, N., Moss´e, M., Pacuit, E., Rus- sell, S., Schoelkopf, H., Tewolde, E., and Zwicker, W. S. Position: social choice should guide ai alignment in deal- ing with diverse human feedback. In Proceedings of the 41st Interna...

arXiv

[3] [3]

Then, for each dimension/property defined on the card below, describe how that dimension works in the process or system you are evaluating

work page

[4] [4]

democratic- ness

Use the guiding questions to reflect on the extent to which that dimension is satisfied. You can find the most up-to-date version of the Democratic System Card, including editable templates that can be initially filled with an LLM (given context) at: democracylevels.org/system-card. Context Describe: Describe the process or a system at a high level. (Can ...

work page 2012

[5] [5]

Democratizing AI

URL https://proceedings.neurips. cc/paper_files/paper/2024/file/ 9328208f88ec69420031647e6ff97727- Paper-Conference.pdf. Goldberg, B., Acosta-Navas, D., Bakker, M., Beacock, I., Botvinick, M., Buch, P., DiResta, R., Donthi, N., Fast, N., Iyer, R., Jalan, Z., Konya, A., Danciu, G. K., Lande- more, H., Marwick, A., Miller, C., Ovadya, A., Saltz, E., Schirch...

arXiv 2024

[6] [6]

Accessed: 2024-09-04

URL https://iap2.org.au/resources/ spectrum/. Accessed: 2024-09-04. Involve. Involve, 2024. URL https://www.involve. org.uk/. Democratic Infrastructure Provider. Kasirzadeh, A. Plurality of Value Pluralism and AI value alignment. In Pluralistic Alignment Workshop at NeurIPS 2024, 2024. Kasirzadeh, A. and Gabriel, I. Characterizing AI Agents for Alignment ...

arXiv 2024

[7] [9]

URL https://www.science.org/doi/abs/10

doi: 10.1126/science.adr6713. URL https://www.science.org/doi/abs/10. 1126/science.adr6713. Summerfield, C., Argyle, L., Bakker, M., Collins, T., Durmus, E., Eloundou, T., Gabriel, I., Ganguli, D., Hackenburg, K., Hadfield, G., et al. How will ad- vanced AI systems impact democracy? arXiv preprint arXiv:2409.06729, 2024. Suresh, H., Tseng, E., Young, M., ...

arXiv

[8] [13]

Summarize what other systems the process depends on or interacts with, which impact its success (e.g., sortition data, or user or citizen authentication systems)

work page