Why Doesn't Anybody Else Know This?

Australia's stranded knowledge, and the walls that hide it

Jul 02, 2026

This article was written with the aid of AI and is sixth in The Salience Series

This is a companion to Why Don’t We Know This?. That article asked why Australians have forgotten their own history — the D-Day medicine, the industry built and lost. This one asks the opposite question: why can nobody outside find what Australia has built? Different stories, same cause. Both are the work of artificial walls we place around our own knowledge — walls that turn a national record into a private one, and leave the value stranded in the gap between the people who hold it and the people who could use it.

In 2022, a San Francisco company called Built Robotics raised $64 million to deploy autonomous excavators on construction sites across the United States.^[1] In 2025, a company called Bedrock Robotics — founded by veterans of Waymo’s self-driving vehicle program — emerged from stealth with $80 million, followed by a $270 million Series B in early 2026, backed by Alphabet’s growth fund CapitalG, NVIDIA’s venture arm, and Tishman Speyer.^[2] Both companies are commercialising retrofit autonomy kits for excavators: LiDAR, GPS, inertial measurement units, cameras, onboard compute, installable in a day on any existing machine.^[3]

Neither company came out of the mining industry. Neither appears to have been aware that the Australian Coal Industry’s Research Program had been funding research on exactly this problem — autonomous excavation, shovel automation, dragline control, digital terrain mapping, real-time machine guidance — since 1994.

There are at least 58 reports in the ACARP archive on overburden removal and excavation automation alone. The research thread runs from C3007 (Dragline Automation, 1996) to C27063 (Automated Mission Planning, 2023).^[4] Nearly three decades of progressive, industry-backed research on the precise problem that Built and Bedrock have just raised $350 million to solve.

The ACARP archive is behind a paywall. Search engines index it poorly, AI systems are unlikely to surface it, and a Waymo engineer designing an autonomous excavator would never think to search a coal industry research portal.

This is not a story about bad research. It is a story about Australia’s stranded knowledge problem — and ACARP is its clearest illustration.

Australia’s stranded knowledge problem

Australia generates between 1,300 and 1,700 new publicly funded research projects every year. Over thirty years, that is somewhere between 30,000 and 45,000 projects — across the ARC, the CRC program, CSIRO, Defence, state co-funded schemes, and the Rural R&D Corporations. The investment is substantial. The outputs are real. The knowledge exists.

But it cannot be found.

There is no central registry of Australian research projects. No national record of what was funded, by whom, with what objectives, involving which researchers and institutions, producing what results. The EU has operated CORDIS — the Community Research and Development Information Service — since 1990, currently documenting more than 133,000 projects across successive framework programs. Each entry records objectives, funding, partners, key personnel, and results. It is searchable, linkable, and publicly accessible.^[5]

Australia has nothing equivalent. The records of decades of publicly funded research exist in fragments — scattered across agency reporting systems, university research offices, acquittal documents in filing cabinets, and the memories of researchers who did the work. When programs end, the records effectively cease to exist. When researchers retire, the tacit knowledge they carry goes with them.

This is Australia’s stranded knowledge problem: not knowledge that is legally locked — though that occurs too — but knowledge that is practically inaccessible. Knowledge created with public or quasi-public funding, of genuine commercial and social value, that cannot be found by the people most likely to use it.

This matters because knowledge diffusion is not a peripheral concern. It is now widely recognised as a critical determinant of a nation’s absorptive capacity — the ability of firms to recognise, absorb, and apply new knowledge — which the innovation literature repeatedly identifies as the binding constraint on research translation, and which I have argued elsewhere is the single most important and most persistently ignored weakness in the Australian innovation system.^[6] A paywalled, sector-siloed, unindexed archive is not a minor inefficiency. It is a direct assault on the one capability that determines whether research ever becomes a product.

The consequences are invisible in any single budget or agency report. The research duplicated by a well-funded overseas startup does not appear as a loss. The commercial translation that never happened because nobody found the prior work leaves no trace. These are costs measured in absence, and absence is the hardest thing for institutions to see.

What stranded knowledge actually means

Stranded knowledge is not a single phenomenon. It takes different forms depending on how the knowledge was created, where it sits, and what would be needed to release it.

The most common form is architectural stranding — knowledge that exists in a structure designed for a different purpose than diffusion. An archive built for funder accountability. A database organised by sector rather than by technology. A portal whose URL signals to the wrong audience that there is nothing here for them.

The second form is access stranding — knowledge behind paywalls, login walls, or technical barriers that make it invisible to search engines, inaccessible to AI tools, and undiscoverable by anyone who does not already know where to look. In the age of search and large language models, a paywalled archive is a dark archive. The content exists but does not participate in the knowledge ecosystem.

The third form is program stranding — knowledge produced by programs that have ended, where the outputs were never transferred to a persistent, accessible repository. When a Cooperative Research Centre closes, its project records typically disappear with it. When a grant program is restructured, the acquittal documents from previous rounds are rarely made publicly accessible. The knowledge existed. The program ended. The knowledge ceased to exist in any practical sense.

Underlying all three forms is a common failure: the absence of any institution or incentive structure whose job is to ensure that publicly-funded knowledge remains findable, usable, and alive after the program that created it has moved on.

What the reports contain that papers and patents do not

A persistent assumption in Australian innovation policy is that publications and patents constitute an adequate public record of research outputs. They do not — and in some respects they actively mislead.

A patent is deliberately obscure. It is written to maximise claim breadth while minimising useful disclosure. A patent tells you that someone solved a problem and where the approximate boundary of their solution sits. It does not tell you how they solved it, what failed before the solution worked, what the operating conditions were, or how to reproduce it. That is by design.

A research paper is selectively honest. It reports what worked, under the conditions that produced a publishable result. Failed approaches, boundary conditions, the contextual knowledge that makes a result reproducible in the field — almost none of that survives the journal. In applied research the gap between what a paper reports and what a project actually learned can be enormous.

A project report is the real record. The problem as it was actually framed. The approaches that failed before the approach that worked. The constraints that shaped the solution. The operating conditions that determined the result. The tacit knowledge of the team, encoded in the decisions made across the life of the project. The adjacent possibilities that were glimpsed but fell outside scope.

In my experience across 36 years as a researcher, no company has ever approached me asking for access to a patent or a paper. They have seen documentation of a project — a description of what was built, for what purpose, under what conditions — and asked whether it could be applied somewhere else. The conversation begins with the project report. It cannot begin with the abstract of a journal article.

Built Robotics and Bedrock Robotics could have found papers from ACARP-funded research. The papers would not have told them enough. The project reports would have. Those they could not find.

ACARP: the case study

ACARP deserves genuine recognition before it receives any criticism. Since 1992 it has administered research funding from the Australian coal industry through a five cents per tonne levy on saleable coal, and maintained an archive of approximately 30,000 project reports.^[7] That is a serious institutional achievement. Most industry research programs produce outputs that disappear into filing cabinets, hard drives, and the memories of researchers who eventually retire. ACARP built something that persists.

But the archive illustrates every form of stranded knowledge identified above — and adds one more layer of its own.

The pricing model that penalises translation. ACARP’s access model works like this: if you are an Australian coal miner, the reports are free. If you are a researcher — including, presumably, the researcher who wrote the report — the cost is $44 per document. If you are a multinational company, perhaps an OEM looking to understand whether a technology is worth commercialising, the cost is $88.^[8]

The people who already know what is in the reports pay nothing. The people most likely to translate that research into commercial products pay the most. The $44 and $88 price points are not high enough to generate meaningful revenue — ACARP is not funding its operations on report sales — but they are high enough to create fatal friction for anyone who does not already know exactly what they are looking for.

There is a further irony. In an innovation system already struggling with attribution — where researchers outside academia have no recognised way to establish their contribution to a project — the paywall charges the researcher who wrote the report to access their own work. I am named as author on several reports in the ACARP archive, including C14043, Accurate Guidance and Measurement for Excavators using a Laser Scanner — a project that demonstrated many of the sensing and guidance concepts now being commercialised by companies such as Bedrock Robotics.^[9] That report costs me $44 to access.

Designed for the wrong audience. The ACARP website was built for the Australian coal industry — its legitimate purpose. The site is organised by mining category: Underground, Open Cut, Coal Preparation. The language is the language of the industry. A construction engineer looking for autonomous excavation technology would search industry terms, OEM catalogues, robotics conference proceedings. They would find Built Robotics and Bedrock Robotics. It would never occur to them to search a coal portal, and so they would not find ACARP.

The underlying technology is entirely industry-agnostic. The perception systems, path planning algorithms, machine control architectures, and terrain mapping methods developed for autonomous draglines are directly transferable to construction excavators, agricultural machines, forestry equipment, and disaster response vehicles. But the knowledge is labelled as coal mining research and housed on a coal industry website. So the construction industry never finds it. The agricultural sector never finds it. The startups that could commercialise it never find it.

The dark archive problem. In 2026, we find knowledge through search engines, citation networks, and AI tools. A paywalled archive does not participate in any of these. Search engines index it poorly, and AI systems are far less likely to discover, read, or incorporate it into their understanding of the field. Citation networks cannot link to it. The 30,000 reports that ACARP has carefully preserved are, from the perspective of the machines that now mediate knowledge discovery, invisible.

This is not hypothetical. I have spoken with researchers at several mining OEMs — companies whose core business is building the very equipment this research addresses — who had never heard of ACARP at all. If the diffusion failure reaches the audience closest to the work, the one that funds, buys, and builds mining technology for a living, it is not a matter of friction at the margins. The knowledge is not merely hard to find. For the people who most need it, it is not on the map.

The operational and technology research distinction

Not all of this is ACARP’s fault, and the distinction matters for what a solution would look like.

A large proportion of ACARP’s portfolio addresses operational problems: ventilation, gas management, ground control, dust suppression, mine rehabilitation. This research improves what mines do. The beneficiary is the mine operator, and the knowledge needs to reach mine managers, safety officers, and engineers — within which community it largely does. Australian coal mines are safer and more productive as a result. For this category, the access model is broadly fit for purpose. The friction is real but not fatal.

Technology research is different in kind. Autonomous systems, machine guidance, sensor fusion, terrain mapping — this research produces knowledge that requires an OEM or technology company to turn into a product. The mine cannot build the autonomous excavator itself. It needs someone outside the fence to do that.

For technology research, the closed model does not merely create friction — it defeats the purpose. The research exists to produce technology that someone will manufacture and sell back to the industry. But the people who would manufacture it cannot find it. So the coal industry ends up paying twice: once through the levy to fund the research, and again to buy the commercial product from a company that solved the problem from scratch.

ACARP was not designed with this distinction in mind. The operational model — closed, low friction within the industry — was applied uniformly to both categories. For operational research it works. For technology research it is a structural blind spot, invisible to a system that was only ever measuring operational outcomes.

The broader pattern

ACARP is the best-documented case, but it is not an isolated one. The same structural failure recurs across Australia’s research infrastructure.

Australia has fifteen Rural R&D Corporations, each administering levy-funded research for a specific agricultural sector: grains, meat, dairy, horticulture, fisheries, and others. Each maintains its own project repository. None of those repositories is cross-searchable. A technology developed for grain production that has direct application in horticulture or viticulture is invisible to anyone searching outside the silo in which it was created. The pattern is identical to ACARP: operational knowledge diffuses within the sector; technology knowledge strands at the sector boundary.

When Cooperative Research Centres close — and they do close, by design, at the end of their funding term — their project records typically disappear with them. There is no mandated transfer to a persistent public repository. The knowledge produced by a decade of collaborative research between universities, industry, and government becomes inaccessible the moment the program ends. It is not destroyed. It is simply no longer findable.

The Australian Research Council funds approximately 670 new projects per year across its major schemes. Its data infrastructure is largely inaccessible to automated tools, limiting the discoverability of publicly funded academic research to those who know where to look and how to navigate the system manually.

Across all of these cases, the pattern is the same. Public or quasi-public funding produces research. Research produces knowledge. Knowledge is recorded — in project reports, in program databases, in acquittal documents. And then the recording system, designed for compliance rather than discovery, ensures that the knowledge is practically inaccessible to the people who could translate it.

In whose interest?

ACARP has never been independently reviewed for technology diffusion or commercial translation outcomes. The program describes itself as a unique and highly successful mining research program — and by its own internal metrics, that assessment is defensible.^[10] Australian coal mines are demonstrably safer and more productive. The operational research worked.

But no external party has ever asked whether the technology research served its purpose: whether the knowledge diffused beyond the industry, whether any of the 30,000 reports contributed to products commercialised elsewhere, whether the access model was ever fit for anything other than operational research.

For operational research, the question of who benefits has a clean answer: the coal industry, appropriately, benefits from research it funded to solve its own problems. For technology research, the answer is less comfortable. The value created by the original research — the decades of accumulated knowledge, the failed approaches, the working solutions — left little trace in the commercial outcome, because the commercial outcome was built by people who never knew the research existed.

The archive was built to show the industry where its money went. It was not built to ensure that the money produced lasting value beyond the industry fence. For operational research, those two things are the same. For technology research, they are not — and the gap between them is where Australia’s stranded knowledge lives.

What needs to change

The solution is not complicated in principle, even if it requires genuine institutional will to implement.

For technology research, the distinction between operational and technology outputs should be made explicit at the project level — at funding, not at archiving. Technology research outputs should be open-indexed as a condition of funding: findable by search engines, accessible to AI tools, visible to the construction engineers and agricultural roboticists and defence systems integrators who would never think to search a sector-specific portal.

Operational research can retain its current access model. The coal industry’s safety and productivity knowledge does not need to be on the open web. But the autonomous excavator guidance system does.

More broadly, Australia needs what it has conspicuously avoided building: a national research project registry. Not a publications database. Not a grants portal. A registry of projects — what was funded, by whom, with what objectives, involving which researchers, producing what results — maintained as a national knowledge asset rather than an administrative byproduct. The model is CORDIS. The domestic precedent is Geoscience Australia, which has maintained the national resource mapping function for physical assets for decades on exactly the same logic: you cannot develop what you cannot find.^[11]

The ACARP case makes the cost of inaction concrete. Thirty years of research. Fifty-eight reports on autonomous excavation alone. Two American startups that raised $350 million to solve the same problem from scratch. A coal industry that will purchase the commercial product without ever knowing that Australian researchers built the prototype three decades ago.

The archive exists. The research is real. The knowledge is valuable.

It just needs to be findable.

Australia has built institutions to fund knowledge creation, but not institutions to preserve and diffuse that knowledge once the funding program ends.

And this problem has human consequences for the people who create knowledge.

Postscript: The Pointless Answer

I should be honest about why this matters to me, because it is not the reason people tend to assume.

I am not looking for financial reward. The research was publicly funded — that was the arrangement, and I have no claim on it. I am not looking for acknowledgement either, though I will admit it would be nice to know my name was attached to something that went on to matter.

What I want is smaller than either of those, and harder to satisfy. I want to know that the work was of use. That somewhere, someone building the thing I once helped build found what we had learned and did not have to start from nothing. That the years were not spent producing reports that would be sealed behind a wall and never read.

One of the most quietly frightening programs on television is the BBC quiz show Pointless. The goal is not to give the best-known answer but the most obscure correct one — the answer that, when a hundred people were surveyed, nobody gave. A “pointless answer” is the prize: something entirely true, entirely valid, that scored zero because no one knew it.

That is the image I cannot shake. Not the failure of a wrong answer, but the strange fate of a right one that scored zero. The studio celebrates it as a triumph. Strip away the lights and the applause and it is something else entirely: a correct answer that, but for a single unlikely person, no one on earth would have said aloud. Being correct and being found are completely different properties. A fact can be right, real, and known to no one at all.

I feel for the person whose answer is branded pointless. A person no one knows, holding a truth no one came for. How cruel a word.

That is the quiet fear underneath all of this. Not that the work was wrong — but that it was right, and scored zero. And I have come to suspect that if I do not tell my own story, nobody else is likely to tell it for me.

Perhaps the uncomfortable truth is this: innovation in Australia has not stopped creating value. We have simply stopped making that value visible.

Which is, in the end, why I built iwiki.au - an open visible site of Australian Innovation — and why I keep writing. Not for me. So that the contribution of others — the inventors never named, the projects quietly closed, the reports sealed behind walls — need never be branded pointless. So that somewhere in the record there is someone willing to reach for the answer no one else thought to give.

References

Built Robotics closed a US$64 million Series C in April 2022, led by Tiger Global, bringing total funding to US$112 million. See “Built Robotics Raises $64M to Transform Excavators into Robots”, PR Newswire, 7 April 2022; and “Built Robotics raises another $64M to make construction equipment autonomous”, TechCrunch, 7 April 2022.↩︎
Bedrock Robotics, founded in 2024 by former Waymo leaders including the former head of Waymo’s autonomous trucking program, emerged from stealth in July 2025 with US$80 million in Seed and Series A funding, and raised a US$270 million Series B on 4 February 2026, co-led by CapitalG (Alphabet’s independent growth fund) and the Valor Atreides AI Fund, with participation from NVentures (NVIDIA’s venture arm), Tishman Speyer, MIT, and others — bringing total funding to over US$350 million at a US$1.75 billion valuation. See “Bedrock Robotics Raises $270 Million in Series B Funding”, PR Newswire, 4 February 2026; and “Bedrock Robotics Emerges from Stealth with $80M”, PR Newswire, 16 July 2025.↩︎
Both the Built Exosystem and the Bedrock Operator are retrofit kits — combining LiDAR, GPS, motion sensors, cameras and onboard compute — that mount to existing excavators in hours or a day without permanent modification. See “Bedrock Robotics raises $270M in red-hot AI sector”, Construction Dive, 11 February 2026; and the Built Robotics Series C sources at note 1.↩︎
Report count and identifiers drawn from the ACARP report catalogue, Open Cut / Overburden Removal category, acarp.com.au. Project numbers C3007 (Dragline Automation) through to recent automation projects such as C27063 span the program’s excavation-automation research thread.↩︎
CORDIS — the Community Research and Development Information Service — has been the European Commission’s primary public repository of EU-funded research since 1990. See cordis.europa.eu.↩︎
On absorptive capacity as the binding constraint on Australian research translation, see my article The Country That Mistook Its Ecosystem for a Pipeline, which traces a decade of Australian evidence — from the 2013 absorptive capacity survey through successive Australian Innovation System Reports and the Productivity Commission’s 2023 Innovation for the 98% — identifying knowledge diffusion and absorptive capacity as critical to productivity growth and research translation. The concept was first formalised by Cohen and Levinthal in 1990.↩︎
ACARP was established on 22 January 1992 and is fully funded by a levy of five cents per tonne of saleable coal paid by all Australian black coal producers. As of its 2025 report, ACARP had awarded $473.22 million in direct funding to 2,140 projects since inception. The archive additionally holds pre-1992 NERDDC reports and the progress and final reports generated across the program’s history; the approximate figure of 30,000 reports refers to the total holdings of the report archive rather than the number of funded projects. See the ACARP 2025 Report and acarp.com.au.↩︎
Pricing observed on the ACARP report catalogue at acarp.com.au: reports are provided free to Australian black coal producers who fund the levy, with per-report charges applying to other users. Prices quoted in Australian dollars and current at time of writing.↩︎
ACARP Project C14043, Accurate Guidance and Measurement for Excavators using a Laser Scanner. Listed in the ACARP report catalogue at acarp.com.au.↩︎
The description “a unique and highly successful mining research program” is ACARP’s own, reproduced across its published materials and partner listings. See, for example, the Australian Coal Industry’s Research Program profile, AusIMM, October 2025.↩︎
Geoscience Australia maintains the Commonwealth’s national geoscientific and resource-mapping functions as a long-term public asset. See ga.gov.au.↩︎

Elliot Duff

Discussion about this post

Ready for more?