While crowdfunding and crowdsourcing has much potential for the decentralisation of architectural production, there are immense challenges in formulating mass consensus mechanisms. The urgency is to rethink the relationship between how data is controlled and communicated and the consequential value circularity within a common data environment. This investigation concerns the design of an architectural system that enables participatory processes and collective input. It frst unpacks the urgency as both a form of technical and socioeconomic engineering, introducing the idea of 21e8-decentralised indexing as system infrastructure. Then, it illustrates the idea through a series of strategic diagrams, comprising 1) an OSI model from physical to user layer, 2) a tech stack between blockchain, BIM, and AI, 3) and data fow from the indexing infrastructure to various modelling interfaces. Aferwards, it demonstrates how architectural information can be indexed and a voting mechanism that ranks contents, with all data transactions mined immutably with blockchain. Subsequently, it shows an ongoing experiment of urban remapping, where AI/BIM collaborate to increase and evaluate options within the system. Finally, this paper concludes by discussing how decentralised indexing may help to promote a peer-to-peer, data-fordata, compute-for-compute environment - a computational data market.
Within any distributed system, difculty in consensus increases as the size of the community scales
up [
In the attention economy, Google remains a dominant player; its centrality is both the reason
for its success and its downfall: it provides a single access point to digital information for
convenience; at the same time, its proprietary Page [
Indexing is the organisation of data according to a specifc schema to make information more
accessible; in a database, indexing helps to structure data in a way that improves retrieval
operations [
To contextualise this within our built environment discipline, individual designers should be able to directly contribute content to a Common Data Environment (CDE) and harness value in their work input via indexing operations. Such value can be realised through search and query: which results should come up frst according to a specifed parameter? Embedding participatory processes within indexing operation means users may consciously and directly infuence how information is ranked within a CDE, as opposed to being passively harvested as behavioural patterns. This highlights the key to value circularity in collective indexing, where blockchain provides prospects as a distributed ledger that records a list of information consensus.
Along these lines, the study investigates two interrelated questions: How can value be
transacted from p2p amongst a network of architectural content-contributors via decentralised
indexing? Consequently, how can value be captured and realised within such data transactions and
in what form?
2. Proof-of-Architectural-Work (PoAW)
Prior research demonstrates blockchain's technical potential for construction collaboration but
reveals critical governance gaps in value capture and stakeholder alignment. Tezel et al. [
Building on these eforts, this research hopes to advance the Proof-of-Work (PoW) mechanism
to a Proof-of-Architectural-Work (PoAW), where the system goal shifs from maximising fnancial
proft to improving built environment quality with the help of Building Information Modelling
(BIM) and Artifcial Intelligence (AI). BIM manages a comprehensive 3D model of a building’s
physical and functional components throughout its entire lifecycle, from design to construction;
while AI has the potential to diversify design through generative algorithms [
‘21e8’ is a blockchain concept that pioneers a “computational data market”, which are
competitive ecosystems that combine algorithmic content creation with distributed data exchange
to displace ad-based ranking and recommender systems [
Design production as sets of decision-making processes can be thought of as iterative
information feedback, where the realisation of the design is the reaching of an equilibrium in a
game, in which all players have no reason to deviate from their chosen strategy [
Once the contribution reaches a stability, an equilibrium, the output would be a design that
went through natural selection within the network and reached a maturity framed by user
preferences [
The architecture of a system that facilitates distributive design production should crowdsource intelligence of humans and machines alike. Today’s AI can be understood as rule/agent-based and Machine Learning (ML) systems. The latter achieves intelligence with machines that defne their own rules based on available data, transcending design from causation to correlations, from small to big data. This may help us to develop a statistical understanding towards our environment; also, increasing options in a system to crowdsource feedback for more democratic digital practices; also, to relieve and automate repetitive computational processes.
This research is interested in a particular type of ML, where Wiener’s [
If AI is to maximise choices in a system: increasing information entropy, then BIM is to
rationalise choices in a system: minimising information entropy. This stabilises a system where
entropy increases globally but decreases locally—Schrodinger's [
The sharing of intellectual labour and properties requires data to stay immutable and distributed so as to provide transparency, where blockchain can be useful. Together, they provide prospects in designing a system that is not only intelligent, but is able to aggregate diferent forms of intelligence to reproduce itself as a system. This requires beyond a mere stack integration of these technologies, to the design of their interconnection and communication.
The Open Systems Interconnection (OSI) model is ‘a conceptual model that characterises and
standardises communication functions of a telecommunication or computing system without
regard to its underlying internal structure and technology’ [
Ng [
Figure 2 shows how diferent sofwares may form a collaborative content-creation environment via the 21e8 index, recording input from multiple sources. As the proposed system is technically distributed, it contains no fles centrally, but a ‘yellow page’ that records all content contributions. Contents can be pulled to a user’s local device upon request for p2p transactions.
The index records users’ query or vote through downloading a JSON fle on the local device,
forming a network of distributed databases. JSON is a programming language, chosen for its ‘open
standard fle and data interchange format that uses human-readable text to store and transmit data
objects’ [
Architectural information can be broadcasted, voted, and mined on 21e8, enabling user rating on information objects for indexing, ranking and searching purposes.
First, a search function within CDEs. Links are ranked according to the amount of votes. Figure 3 is a working demo illustrating a query for ‘ark.page’, the frst link received the most votes—6 votes represented as 6 green boxes next to the link, thus being assigned with the highest value amongst all searches. The vote action puts users in a conscious position of their data contribution. In comparison to Google’s proprietary search index, users have no access to the reason behind certain searches being ranked as the best results. For instance, large platforms like Twitter naturally have the highest score for content linkage, it will always occupy the top search results. Whereas 21e8 applies PoW to create a custom bias over the content through voting mechanics— users can invest a few clicks to infuence the index structure by which the content is ranked. This is how an information object is turned into a digital asset, which is a database record of all transactions and users who have invested computing power, creating network efects.
With each vote, a JSON fle is being downloaded on the user’s local computer to contribute storage and computing power to the network. This secures one’s vote without exposing one’s search history, forming a network of immutable distributed ledgers and decentralising the overall system. The option of creating a shared index among multiple parties or publishing transactions on a public or consortium blockchain can be enabled on request.
Figure 4 contains an example JSON recording the 1) search query ‘provides Ng’, 2) retrieved url, 3) cryptographic hash code (the identity of the data), 4) assigned index value, and 5) target index ‘21e8’. This helps to build diferent ways of investing compute power in a computational data market, relieving symptoms of data licensing.
Second, a content broadcast function. Imagine BIM’s CDE working like a social media platform where designers as content-contributors are able to broadcast their work in the hope of harnessing network effects. The hashtag function helps users to place their content under a certain query or topic so as to enable search optimization. In Facebook, content-contributors have the option of purchasing ads to get their content ranked up on the platform; whereas the 21e8 index provides a more straightforward means for content broadcast without having to go through third party intermediaries. Figure 3 illustrates how one can navigate the topic ‘omniverse demo chair’ and broadcast one’s content underneath using the mine button and anchor it to the index. If one wishes to rank the content up, one would only have to invest a few clicks and computing power to vote it up.
This creates a microeconomy around indexing that transacts computational value, which is realised when the compute is being put to use. Currently, this demo version only mines URLs onto the index. Further tests have to be carried out to expand data types, as illustrated in Figure 3, including executable files (.exe) for building compute services, application files (.app), image files (.jpeg) and so on, are currently undergoing preliminary tests.
In order to demonstrate the larger social implication of decentralised indexing, here tabulates a sketch of potential participatory urban planning processes that can happen on 21e8, taking Hung Shui Kiu (HSK) New Development Planning (NDA) in Hong Kong (HK) as an example.
Currently, the NDA planning process takes place in a centralised manner that has difculties in comprehending and crowdsourcing opinions from citizens. The master plan is directly formulated by the government which becomes statutory afer a 3 weeks period of public consultation; the communication is unidirectional. Figure 5 provides a high-level system overview and visualises user interactions.
The overall system goal is for planners to crowdsource ideas and comprehend views from both experts and general users. Designers and citizens can contribute graphs and urban data (e.g. street maps, etc.) for the AI to output urban planning options, which will be evaluated using BIM and anchored on 21e8. Users can then vote on these options to rank them on the index. Content contributors would be rewarded with data / compute / tokens according to received votes. All information would be mined and hashed via blockchain. Planners can then rationalise and consolidate options into a conceptual plan and feedback on the index for citizens to go through a second round of voting.
The output will then be iterated until it reaches an equilibrium in the voting process into a master plan that can then become statutory. Implementation of the plan can be traced-and-tracked and monitored through the BIM system to coordinate with a supply-chain of contractors, all processes indexed and published on 21e8 to enhance efciency and transparency.
The following test demonstrates how AI may help in diversifying urban planning options, using various graphs to analyse the relationship of interacting nodes in HSK NDA, which is undergoing revitalisation with Transit-Oriented Development (TOD) strategies. The network analysis visualises existing urban density using StyleTransfer AI. Compared with HKSAR’s plan, which shows a clear functional division of the NDA into land fragments to tender out to developers, this study’s tests demonstrate how diferent plots may form various complex webs between existing low-rise and village houses, bridging them into central transit and other function hubs, synthesising urban fabrics and their communities.
The network analysis rationalises the urban fabric with topologies extracted from 1) e8 graph, 2) random graph, 3) small world graph and 4) preferential attachment graph, showing how transit hubs can be embedded in the area while giving consideration to the existing complexity of the fabric. The blue dots show the centrality of the network; while the e8 root system ofers a way to label functionality into each node with various colour labelling. The colour can be used as a semantic label by assigning centrality and functionality values to be plugged into BIM for analysis, and act as a datum to evaluate NDA plans in terms of how the area would network itself, which is the next step of the project.
From these tests, a brief summary on how 21e8 can help to improve the pipeline based on crowdsourcing and crowd-indexing. First, upscaling resolution, which was largely infuenced by the lack of open source data in the area. Second, crowdsourcing graphs can facilitate a larger range of options and variety in decision making and forming better evaluation metrics. Third, decentralised indexing can help to expand the labelling beyond functionality to other urban utilities, including power grid, sewage system, public facilities, etc., which are currently unavailable data. Fourth, crowdsourcing opinions from local inhabitants and stakeholders for participatory planning (e.g. infrastructural support, functional hubs, community services, etc.).
Refecting on the experiment, the core function of the decentralised index replaces proprietary search algorithms with a community-driven, blockchain-anchored method for ranking content in a CDE. Contextualising it in urban design, the index couples with AI-BIM Interaction, where AI diversifes design options while BIM evaluates feasibility. Together, they calibrate decision-making in participatory systems. The key components are as follows.
Content Contribution & Anchoring: Creators submit architectural data (BIM components, graphs, codes, etc.) to the CDE. Content is cryptographically hashed and "mined" onto the blockchain index via JSON fles, creating an immutable record.
Voting & Ranking: Users vote on content relevance/quality (e.g., design options), each vote is a JSON fle record on the voter’s local device (decentralized storage), consumes minimal compute resources (e.g., storage, processing), and increase the content’s index value, determining its search ranking.
Value Circularity: Votes act as microtransactions of value (data, compute, or tokens). Topranked creators earn rewards (compute/tokens) to incentivise quality contributions. Users “pay” with compute resources with each vote to prevent free-riding and “bad votes”.
Consensus & Security: Blockchain immutably records all transactions. No central server: Index is built from distributed JSON fles across users’ devices. Content hashes ensure data integrity; votes are cryptographically signed.
◦ ◦ Input: Crowdsourced urban data (e.g., ideas, graphs, votes).
Process: Uses ML (e.g., StyleTransfer, GANs) to
Generate diverse design variants (e.g., urban layouts).
Analyze complex relationships (e.g., density, connectivity via graph topologies). Output: Multiple design options with semantic labels (e.g., functional zoning, centrality scores). • • • • • •
Input: AI-generated options + physics-based constraints (e.g., structural, environmental). Process: Simulates real-world performance (e.g., energy use, spatial conficts); provides quantifable metrics for user evaluation (e.g., cost, carbon footprint).
Output: Rationalized options ready for voting on 21e8.
Calibration Loop: AI Diversifes → BIM Rationalizes → Users Vote → Iterative Feedback Loop. Together, BIM’s physics-based simulations ground ML’s statistical predictions, reducing speculative and low-quality input/output.
Future technical integration should test two components. First, the data bridge between AI/BIM tools connected via APIs and Omniverse using USD format. This would enable real-time 3D data synchronisation from AI-generated graphs to the BIM simulation environment. Second, with 21e8 as Mediator, this experiment only tested hosts ranked options for voting. Next step should include rewards for creators of high-voted AI/BIM content with compute/resources.
The following points should be considered before further implementation. First, Decentralized Indexing does not equal Traditional Databases as there would be no central database; instead, an Index is a distributed network of JSON fles on the communities’ local devices. Also, it is not a system for fle storage: 21e8 stores content metadata (hashes, votes), while actual data remains peer-to-peer.
Second, AI-BIM users are not Competitors in such a system, instead, They Complement each other. AI users’ role is to expand solution space (e.g., 10 urban layout variants from a single idea seed contributed by creators). While BIM users are validators who narrow down options via simulated constraints. This bridges big data patterns with Newtonian physics to diversify balanced options.
Third, it is important to note that this system, while decentralised, possesses a certain level of centrality based on such evidence-driven mechanisms to create informed decisions.
Finally, and most importantly, PoAW is not a Proof-of-Stake. There will be no fnancial speculation as "Work" is a piece of contribution in the form of design content/compute. Also, this decentralised system has the goal to enhance built environment quality, not proft maximization. This create a computational market that circulate values beyond monetary: • •
Data-for-data: Contributors trade datasets (e.g., open street maps for BIM analytics). Compute-for-compute: Each voting action contributes minute local resources (e.g., CPU for ML training).
The use case clarifed these key conceptual ambiguities through processes of citizens crowdsourcing, BIM evaluation, expert voting, to iterating options until consensus (Nash equilibrium). This synergy enables scalable, transparent participatory design while mitigating centralization risks and "bad voting"—a core issue in many decentralised autonomous organisations.
The proposed decentralized index transforms Common Data Environments (CDE) into a selforganizing, p2p computational market where Ranking is democratized via blockchain-anchored voting and value fows as data/compute (not just money). This Eliminates central gatekeepers (e.g., Google, Facebook) to create content ranks that truly refect collective preferences beyond corporate/advertising interests. Further, the AI-BIM interaction creates a calibrated feedback loop, where AI users explore possibilities and BIM users ground decisions in reality considerations. Rewarded via 21e8’s incentive layer for improving built-environment outcomes, This enables users to retain control over their data/compute resources while creating validated options and mitigate “bad voting”.
Through considering the integration between blockchain, BIM, and AI from both technical and socioeconomic perspectives, this study discussed how such forms of system design might prompt changes around architectural production. It elaborated on the idea of a computational data market and how it may aggregate the intelligence of both humans and machines, and continuously reproduce itself as a collective system. This tackled an urgent problem in large scale information systems—the ability to index and rank information, a consensus mechanism to validate work and content contribution. From these arguments, the study proposes the application of the 21e8 infrastructure.
Overall, the investigation exemplifed how the system might work at three levels: a system design of the technologies and their communication, a user interaction design, and a presentation of data output from ML. On the technical side, it positioned blockchain as the shared data layer to be integrated with BIM’s CDE, using JSON to store transactions on local desktops. On a conceptual level, it innovates creator/diversifer/validator roles: BIM evaluates content as a frame of reference and interfaces communication between a network of scattered actors, while AI algorithms are compute services that learn from crowd contributions to diversify design options. Along these lines, the study ofered means to large-scale human-machine interactions, bridges proprietary and crowdsource eforts, automating decentralised indexing from real-time data streams.
This project emerged from a partnership with Mark Wilcox, a great friend who has truly inspired me throughout my journey to decentralisation. Also, I must thank Prof. Mario Carpo for his mentorship during my postgraduate study, who provided great support and guidance in bridging digital theory with design applications. Finally, many of these ideas frst came to my mind when I was with the Strelka Institute, a stimulating arena under the direction of Prof. Benjamin Bratton.
During the preparation of this work, the author used ChatGPT for grammar and spelling checks. Afer using this tool, the author reviewed and edited the content as necessary and takes full responsibility for the content of the publication.