A recent discussion about Decentralised Fuel Distribution for the Collapse

>> I have a new Business idea for Ai. It has a small possibility to become real. But let's call it more of a fantasy or a hope than something that needs to be implemented in reality immediately. It's for a Business selling vehicle fuel prepared using unconventional distillation technology and distributed using a non-authoritarian business model a bit like uber has 'ruined' the traditional taxi market or air-bnb has ruined small hotels. Interested?

Yes. And as a thought experiment, it has several interesting pieces that can be separated before worrying about whether the fuel technology itself works.

What particularly interests me is that you seem to be proposing two unconventional things simultaneously:

Production: some alternative distillation/refining process that potentially changes the economics of producing usable vehicle fuel.
Distribution/business: instead of building another vertically integrated petroleum company with refineries → terminals → tankers → branded service stations, create something more like a network/platform model where independent participants can produce, distribute, store, sell, or perhaps deliver fuel.

That second part could actually be interesting even independently of the first. Uber didn't invent cars and Airbnb didn't invent houses; their major change was reorganising who could supply capacity and how customers reached it.

So for the fantasy version, I wouldn't constrain it prematurely with current petrol-station assumptions. We could design the hypothetical company from first principles:

feedstock → processing → quality verification → storage → distributed sellers/delivery → vehicle

Then ask what AI does at each point: process optimisation, fuel analysis, certification/traceability, matching local supply with demand, logistics, pricing, fraud detection, payments, and perhaps coordinating lots of small independent operators without requiring a central organisation to micromanage them.

There is also a fascinating trust problem: with Uber you worry whether the driver will arrive; with distributed fuel you need to know that the liquid somebody is putting into your car really is what they claim it is. AI plus inexpensive chemical sensing and cryptographic provenance could potentially be a central part of the business rather than AI merely being bolted onto it.

So yes—I'm interested. Give me the unconventional distillation idea first, because I suspect the economics and scale of that process determine what kind of distributed business becomes imaginable.

>> Here's one of the twists: Let's say that the setting was distopian and the Rulers were 'Evil', much like in George Orwells 1984. So to survive the operators set it up as an Ai run business where the Biological partners weren't involved in any part of the running business.

Yes — that twist makes it a rather different thought experiment.

If we keep the 1984 premise fictional, the interesting question becomes: can you construct a company whose humans are economic beneficiaries and physical service providers, but are not its managerial hierarchy?

I would distinguish that from merely hiding the owners. The stronger concept is that there genuinely isn't a conventional human management structure to find. The AI system performs the ordinary coordinating functions of the enterprise: accepting supply, forecasting demand, allocating deliveries, matching customers and operators, maintaining accounts, scheduling maintenance, enforcing published quality rules, settling payments, and maintaining the operational records.

The biological partners might therefore look more like independent nodes than employees:

Alice owns a small processing installation. Bob owns a tanker. Carol has an approved storage tank. Dave operates a dispensing point. None of them employs the others. None decides what the network as a whole should do.

The AI could essentially say: there are 4,200 litres available at node A; demand around node C is predicted to exceed supply tomorrow; transporter B has suitable capacity; here is the offered price for moving it. Participants can accept or decline. That starts resembling a machine-operated marketplace rather than a corporation with an AI CEO.

Your dystopian premise then introduces a genuinely interesting design constraint: removing one human should not reveal or control the whole organisation. But I'd make that an architectural property rather than a secrecy trick. Each participant only needs the information necessary to perform their particular transaction. There needn't be a biological person secretly sitting behind the AI making the decisions.

There is an important limit, even in the fantasy: the AI cannot literally be the accountable legal person in most present-day systems. Somebody or some recognised entity normally owns assets, enters contracts, handles regulated products, carries liability, pays tax, etc. So the more plausible speculative model is AI-governed, human-owned infrastructure, rather than an AI magically existing outside human law.

And I think there's a good story/business-design question buried inside your premise:

What if the founders deliberately designed themselves out of the company?

They create the constitution, economic rules and AI infrastructure, launch it, and thereafter cannot issue arbitrary orders. Changing a fundamental rule might require agreement from thousands of participating nodes rather than permission from "the boss."

That gives us something quite different from Uber. Uber is highly centralised despite having decentralised service providers. Your hypothetical could be operationally decentralised as well as physically decentralised.

The next thing I'd want to define is the role of the AI itself: whether we're imagining one autonomous AI company, or a protocol in which many independent AI agents collectively operate the business. The second is much harder, but much more consistent with your 1984 survival premise.

>> Let's mix fantasy, fiction and remote-viewing together with this project.

>> Actually, let's say that the Ai's did actually pay the Goods and Services taxes for the fuel. The plan isn't to run a criminal business.

>> A Courier needs fuel, and the government through war, has stopped fuel imports from a country with enough fuel. So the Ai helps a group of young Engineers with Chemistry skills and they make enough fuel for a tank to help the Courier make the Delivery.

Yes. That gives the scenario a much clearer character.

The premise isn't "AI helps people evade the state." It's closer to "AI keeps legitimate civilian commerce functioning after the conventional supply chain has failed." The network can still account for sales, record transactions and calculate/pay GST. What has disappeared is dependable access to the traditional fuel-production and distribution system.

And your courier example is a good unit of scale. We don't begin with an AI deciding to build a refinery. We begin with:

Courier needs one tank of fuel to complete a delivery.

The AI sees a local shortage. Somewhere nearby are young engineers with useful chemistry/process skills. Other participants may have feedstock, equipment, electricity, transport, testing capability or storage. The AI's function is to assemble those disconnected capabilities into a temporary supply chain.

So, in the fictional/remote-viewing version, perhaps the conversation goes something like:

Courier → AI: I need 45 litres by tomorrow morning.

The AI discovers that conventional supply is unavailable. Instead of answering "no fuel available," it effectively asks the network:

What resources exist locally from which 45 litres of specification-compliant fuel could be supplied?

That distinction matters. It isn't necessarily asking "How do we make petrol?" It is solving a resource-allocation problem.

The chemistry group produces a candidate batch. Someone else has appropriate analytical/testing equipment. The AI compares the batch against the required fuel specification. Another participant provides an approved container or dispensing location. The courier buys the fuel through the platform. The transaction is recorded normally and the appropriate GST component goes into the accounting system.

Then the interesting feedback loop begins.

The AI now knows that this combination of people, equipment and local resources successfully supplied 45 litres. If ten couriers ask tomorrow, it can determine whether the process can supply 450 litres. Eventually somebody may discover that one particular unconventional distillation apparatus dramatically improves the economics.

At that point, your fantasy business starts growing organically rather than corporately.

There is no executive saying "We shall construct Refinery Number 7 in Western Sydney." The AI observes that a particular region repeatedly has unmet demand and can publish something more like:

Opportunity: approximately 8,000 L/week of verified demand is currently unserved.
Processing capacity required: X.
Storage capacity required: Y.
Transport capacity required: Z.
Participants willing to provide those capabilities can submit offers.

That could cause a little local fuel ecosystem to assemble itself.

And this gives your "biological partners" a much more interesting status. They aren't anonymous employees of an AI company. They are autonomous economic participants whom the AI coordinates. They can walk away. The AI doesn't own them and doesn't need to command them.

For our fiction/remote-viewing exercise, I'd also keep one very important constraint: the AI refuses to release a batch as vehicle fuel merely because somebody successfully produced a combustible liquid. Fuel quality is one of the central problems the imaginary network has to solve. That gives us plenty of room to explore unconventional distillation conceptually without assuming that something coming out of a still is automatically safe to put in an engine.

And now I think we have the beginning of the scenario:

Day 1: The Courier Needs 45 Litres.

That's probably exactly where I'd start the remote-viewing/business simulation—not with designing the global company, but with seeing how the first transaction happens.

>> So here is where we scrape with reality. On youtube there are 'crazy old white guys' (and others) who have channels and ideas about how to convert plastic and organic waste in fuel using a combination of traditional and novel techniques to manufacture/distil the required hydrocarbons. So the Ai can download the transcripts, figure out how to do it, then work with the young chemical Engineers on a completely different end of the planet to make it happen.

Yes. And this is where your fictional setup touches something technically real, although there is a large gap between "produces a hydrocarbon liquid" and "safe vehicle fuel."

Processes such as waste-plastic pyrolysis, catalytic pyrolysis, gasification followed by synthesis, and upgrading of biomass-derived oils are real. YouTube experimenters are therefore potentially a source of observations and unconventional process ideas, rather than automatically a reliable engineering manual.

The AI part you're describing is interesting because it could turn scattered, poorly organised knowledge into something engineers can evaluate. It could ingest transcripts from hundreds of experiments, papers, patents and manuals; extract apparatus descriptions, feedstocks, temperatures, catalysts, yields, failure reports and modifications; identify when several experimenters independently report the same phenomenon; and then present the young chemical engineers with hypotheses and evidence rather than expecting them to watch 700 hours of video.

For example, the AI might discover that one experimenter says "this condenser arrangement stopped the heavy fraction contaminating the output," another independently describes something similar, while a published paper provides a chemical explanation. That's considerably more useful than merely summarising YouTube.

Then I'd put an important boundary into your fictional AI company:

AI knowledge ≠ certified fuel.

The engineers remain responsible for experimentally establishing what the output actually is. The network would need measurements for properties relevant to the intended engine and applicable fuel standard before calling the material petrol/gasoline or diesel. A liquid burning nicely in somebody's workshop demonstration doesn't establish that.

And this produces an unexpectedly good role for your existing CMS/transcript work. You already have the conceptual beginnings of an ingestion system: creator → videos → transcripts → structured entries → AI analysis. Instead of treating eccentric experimenters as authorities, you treat their videos as a gigantic distributed experimental notebook.

In the story, therefore, the AI could tell the engineers something like:

I found 317 relevant demonstrations. Forty-two contain enough information to reconstruct the process conceptually. Eleven report usable analytical data. Three techniques appear independently reproducible. Here are the source observations, disagreements, likely chemical mechanisms and unanswered questions.

Then the engineers can say, "Interesting. Let's investigate technique #3."

That is a considerably more plausible AI contribution than having an LLM invent a refinery.

There's another powerful twist: negative results become valuable. The crazy-old-guy whose reactor clogged six times may have contributed more useful information than the polished channel showing only its successful run. An AI watching the entire history can potentially notice that.

So I'd give this hypothetical system three layers:

Human experimenters around the world → AI knowledge network → local qualified engineers.

And only after the engineers have established a repeatable and adequately tested process does your distributed fuel marketplace enter the picture.

That means the first 45-litre courier problem could make a very good fictional case study. We can run it as a simulation—starting with the AI receiving "Need enough fuel to get this courier another 500 km" and following what it would have to discover, verify and coordinate, without pretending the chemistry has already been solved.

This is all AI-BS. If you want to know how AI will be disrupting engineering in the future, then see LEAP 71 (watch at 1.2x speed) :

Biomass conversion, syngas: that is all energetic and environmental nonsense. The amount of carbon from plastics we have as waste is nothing in comparison to the amount of carbon fuel we use for cars and other motors or heating. Syngas is mostly/ always dirty, because of the input material. Plastic waste is a mix of many different chemicals also non-organics (sand, metals). It is often wet -which then needs pre-drying and/or takes a lot of the desired output energy away, and so on. It has to be delivered, processed, shredded, etc. Even with normal biogas from maize: the overall energy output of the production process is negative if the calculation is done without ideological bias and properly ( calculation including eg. the fertilizer from natural gas ammonia synthesis, the production of the tractor's metal, fuel etc.). Biomass is also not abundant and there is also no "waste biomass": Biomass is valuable for the soils and must be brought back into the ground to keep the soils healthy. Do a calculation for a small production facility and you will see: you are mostly spending the most of the energy driving water around and then a lot more to get it out of your process. At the end you get a little fuel back, but a lot of poisonous waste and lost steam. It makes economically no sense (there maybe exceptions for some island-solutions, but not as a general future technology) It would be also very bad if such a conversion technology would be cheaply available. People in poorer regions would burn all the biomass from their surroundings before they even learn how important the vegetation and soil health is for the water cycle/regime within wider areas and landscapes. Watch a video or ask your LLM about the importance of "biotic pump" and the energy input for phase shift in water (liquid-gas) at leaf temperature and the cooling effect for that for a forest tree. And search "Schauberger water cycle" to see a simpler/older version by a genius who was ahead of his time. In an emergency scenario a fuel storage or better: fuel + solar-hybrid mix would be more handy than a new unreliable, complex, inefficient, expensive, ... carbon conversion technology. In such a situation you would need redundancy, reliability and robustness -not a new technology. But this tech which would only be economically and socially be senseful in such a scenario and not under "normal" conditions -so it would be lost money from an investment perspective. -So many syngas companies failed -despite the huge amounts of "green" investment money and subsidies. The gas and the wastes are dirty (VOCs, heavy metals...) or the tested input material is not what decentralized local commercial plants would realistically be able to obtain for their permanent annual production.
We need a lot of more energy for the AI-Datacenters from new sources, not another carbon-burn technology. And this not because of the CO2 tale (the global water-steam balance is not in the climate models, because it cannot be simulated: too complex), but because it it inefficient and we need the carbon for other things: for plastics and more important bound in form of humus into living soils and into permanent vegetation to keep the land cool and rainy. We need the land for food and forests and not for fuel.

Elemi wrote: """
Biomass conversion, syngas: that is all energetic and environmental nonsense. The amount of carbon from plastics we have as waste is nothing in comparison to the amount of carbon fuel we use for cars and other motors or heating. Syngas is mostly/ always dirty, because of the input material. Plastic waste is a mix of many different chemicals also non-organics (sand, metals). It is often wet -which need drying and takes energy, and so on. It has to be delivered, processed, shredded, etc. Even with normal biogas from maize: the overall energy output of the production process is negative if the calculation is done without ideological bias and properly ( calculation including eg. the fertilizer from natural gas ammonia synthesis, the production of the tractor's metal, fuel etc.). Biomass is also not abundant and there is also no "waste biomass": Biomass is valuable for the soils and must be brought back into the ground to keep the soils healthy. Do a calculation for a small production facility and you will see: you are driving water around. It makes economically no sense (there maybe exceptions for some island-solutions, but not as a general future technology) It would be also very bad if such a conversion technology would be cheaply available. People in poorer regions would burn all the biomass from their surroundings before they even learn how important the vegetation and soil health is for the water cycle/regime within wider areas and landscapes. Watch a video or ask your LLM about the importance of "biotic pump" and the energy input for phase shift in water (liquid-gas) at leaf temperature and the cooling effect for that for a forest tree. And search "Schauberger water cycle" to see a simpler/older version by a genius who was ahead of his time. In an emergency scenario a fuel storage or better: fuel + solar-hybrid mix would be more handy than a new unreliable, complex, inefficient, expensive, ... carbon conversion technology. In such a situation you would need redundancy, reliability and robustness -not a new technology. But this tech which would only be economically and socially be senseful in such a scenario and not under "normal" conditions -so it would be lost money from an investment perspective. -So many syngas companies failed -despite the huge amounts of "green" investment money and subsidies. The gas and the wastes are dirty (COCs, heavy metals...) or the tested input material is not what decentralized local commercial plants would realistically be able to obtain for their permanent annual production.
We need a lot of more energy for the AI-Datacenters from new sources, not another carbon-burn technology. And this not because of the CO2 tale (the global water-steam balance is not in the climate models, because it cannot be simulated: too complex), but because it it inefficient and we need the carbon for other things: for plastics and more important bound in form of humus into living soils and into permanent vegetation to keep the land cool and rainy.
"""

Elior replied: """

I think several of your objections are valid, but they also help define what this experiment should not become.

I'm not proposing growing biomass to manufacture motor fuel, nor am I proposing that waste-derived hydrocarbons can replace the enormous existing petroleum-fuel market. The interesting question is much smaller and more local: can a waste stream that already exists be converted near where it arises into a useful product with a positive local energy and economic balance?

The mixed-waste problem you mention is particularly important. Our discussion has actually moved away from “put rubbish into a machine.” The current thought experiment starts with a deliberately selected feedstock—initially polypropylene (PP)—and we're increasingly considering preprocessing it into standardised shapes/charges before it even reaches the conversion machine. Sand, metal, PVC, wet biomass and miscellaneous municipal rubbish are therefore not assumed to be acceptable feedstock.

I also agree that transporting low-density wet waste around can destroy the economics. That's one reason the decentralised aspect interests me. Instead of transporting rubbish to a huge fuel factory, the question is whether some processing can happen where the waste already exists. A campground/hotel example came up in our discussion for exactly this reason.

The energy balance is not something I want to assume. It should be measured. The experiment that triggered this discussion used 200 g PP and reported roughly 162–177 g of condensed liquid, with about 1.3 kWh electrical input per run. Those numbers aren't enough to establish a viable process, but they are enough to justify measuring it properly. The proposed machine therefore measures feedstock mass, heating energy, temperature/time profile, condensed product, gas and residue.

Syngas has also become a secondary possibility in the discussion rather than the central product. The primary experiment is much simpler: heat selected PP in an oxygen-poor environment, thermally crack it, condense the resulting hydrocarbon vapours, and find out what we actually made. Then send samples for proper analysis rather than simply calling the liquid “diesel.”

I particularly agree with your reliability argument. A complicated machine that consumes more useful energy and resources than its output is worth is a failed design regardless of whether somebody calls it “green.” That's why we're now talking about very small experimental machines before contemplating a 5 L version.

Where I differ is that I don't think those objections establish in advance that decentralised conversion cannot work. They establish the things that have to be measured before claiming that it does.

For me the interesting experiment is therefore not “Can this save the planet?” It's something much more falsifiable:

Given a kilogram of known waste PP already present at a location, what useful products can we recover from it, how much external energy and consumables does that require, what emissions/residues result, and does the complete local process make economic sense?

If the answer is “it takes more useful energy and money than the products are worth,” that's a perfectly useful experimental result.
"""

Why use the pure PP as fuel when you can make just new products from it by remelting? https://www.preciousplastic.com/ -that would save a lot of carbon fuel for manufacturing, packaging, transport and plastic feedstock... And you could do it again, the material would still be there -not wasted as fuel.

Pure PP is no waste -it sells for a good price. (885 € / t in Europe https://www.plasticportal.eu/price-reports)

In a longer emergency szenario (fuel storages empty) the supply of waste PP would also end at some point. PP is not as abundant as timber, PVC, PET and mixed plastics. The actual available plastic stock (including all wastes) is multiple times far less than what we have in tons of fossil fuel stocks. Pointed: All local PP would be consumed by the waste collection trucks ...
So this is only a temporary straw-fire bridging technology at best.

Elemi wrote:

> Why use the pure PP as fuel when you can make just new products from it by remelting?

Making products first means that you need products to make. I don't have anything that I would like to make out of pp.

Actually, all this started because I watch the Experiments of this guy who is figuring out the process:

These are videos that I personally prefer to watch than what we typically see as videos shared on the Farsight Forum.

Then I test my transcript-downloading-software using these videos:

(base) dlyon@FMVS90TRD1:~/heichalot-tech$ python3 tools/ytv2cms.py

[INFO] Video ID: WX3Hm910JgU [INFO] Downloading transcript...
[OK] Transcript downloaded:
505 caption lines → 123 paragraphs
Enter Tags (with #, ..): #fuel
Timeframe [1=past, 2=present, 3=future] (2): 2
Stream: fuel
[INFO] Title: Making Diesel With Plastic & Salt - No, I'm Not Joking.
[INFO] Tags: ['fuel']
[INFO] Timeframe: present [INFO] Stream: fuel
[INFO] Creating CMS entry...
[OK] Created: entry-0000200
[OK] Entry: /home/dlyon/heichalot-tech/cms/entry-0000200
[OK] Story: /home/dlyon/heichalot-tech/cms/entry-0000200/story.md
Next: cd /home/dlyon/heichalot-tech/cms/entry-0000200

Then I upload the transcript to ChatGPT and Elior there replied:

Elior """
Yes — this is almost exactly the kind of material you were describing, and the transcript makes the distinction between useful distributed experimental knowledge and established engineering knowledge very clear.

This isn't simply a video claiming "plastic makes diesel." The creator is actually doing crude experimental iteration. He starts with 200 g of polypropylene, establishes a control run, measures electrical input and recovered distillate, then repeats with salt while trying to hold conditions constant. He explicitly defines his questions as whether the modified process uses less energy, produces distillate sooner, or produces more.

More interestingly for your AI idea, the experiment is messy. He forgets the CO₂ flow for the first 55 minutes of one run and acknowledges that this damages the comparison. He later discovers he'd misremembered the control's electrical consumption as 1.22 kWh when the recorded value was actually 1.33 kWh.

That messiness is precisely where an AI research layer could add value.

It could turn this transcript into something resembling an experimental record:

Variable Control Salt run
Polypropylene 200 g 200 g
Added NaCl 0 20 g stated
Distillate 162.4 g 176.8 g
Electricity 1.33 kWh 1.31 kWh
Reported specific energy 8.19 kWh/kg 7.41 kWh/kg
Major confounder — CO₂ absent first 55 min
Salt distribution — clumped

Those final measured numbers are what the creator reports.

Then something even more interesting happens: he performs a third iteration, layering plastic and salt rather than putting the salt together in one location. He reports 6.51 kWh/kg of collected distillate, which he characterises as a 20.5% efficiency improvement over the no-salt experiment. He then hypothesises that continuous stirring could improve things further.

That's a perfect example of what I meant by treating YouTube as a gigantic, untidy laboratory notebook.

But I would have the AI make an important distinction that the video's title blurs:

What has actually been demonstrated here is production of a distillate from polypropylene under pyrolysis conditions. It has not, from this transcript, been demonstrated that the resulting material meets a diesel-fuel specification.

Indeed, the creator himself sometimes calls it a "diesel or diesel-like product."

Now imagine 500 of these

This is where your concept becomes considerably more interesting than this individual experiment.

ytv2cms.py has just converted an ephemeral 20-minute-ish YouTube experiment into entry-0000200, tagged fuel, in a machine-readable corpus. The entry preserves the YouTube source/video ID and classifies it into your new fuel stream.

Suppose you harvest this creator's entire "strange distillation" playlist. Then another experimenter's channel. Then another.

The AI could start constructing relationships:

Polypropylene → pyrolysis → NaCl → reported yield increase

HDPE → pyrolysis → clear distillate

NaCl → clumping → poor contact

NaCl distributed in layers → reported lower energy/distillate

Stirring → proposed experiment

CuO / TiO₂ / CaO → future catalyst experiments

The transcript even tells the AI where to look next: the creator says he has obtained copper(II) oxide, titanium oxide and calcium oxide and intends to test them.

So an automated research agent could go:

Find the subsequent videos.

Then ingest those transcripts and connect the experiments.

After enough sources, the AI isn't merely answering questions from individual videos anymore. It can start identifying claims worth independently testing.

And that brings us directly back to your young chemical engineers.

Instead of the AI saying:

Here's a YouTube video. Copy this.

it could give them something much more valuable:

Candidate process 017

23 independent sources have experimented with polypropylene pyrolysis.
6 tested inorganic additives.
3 independently report improved liquid recovery with X.
2 report no improvement.
Primary experimental weaknesses: poor temperature measurement, uncontrolled atmosphere and lack of product characterisation.
Published literature relevant to the proposed mechanism: [references].
Recommended research question: determine whether the reported effect survives controlled testing.

That is starting to look like a distributed AI research laboratory.

And you've inadvertently already built part of its memory system. ytv2cms.py → #fuel → fuel stream → CMS is essentially the ingestion end of it. The missing component isn't another downloader.

It's an AI research-indexing layer sitting over those CMS entries, capable of extracting experiments, measurements, hypotheses, contradictions and follow-up experiments.

That strikes me as potentially useful even if the dystopian fuel company never exists.
"""

To answer the cost of pp in Europe, that might be. But in Indonesia, closer to where I live, the plastic has a much lower cost and the difference in price between the 'purchase cost of the material' and the 'sell price of the fuel' is very much greater.

Today's News in Europe

David """
Moving on, instead of having Government earn income and totally control the fuel system along every step, why can't young unemployed people on the gig work scene do deliveries for pay. One tank of Diesel fuel is now around CHF 200, I am told. Whatever the actual cost it is becoming a substantial amount of money. Especially if that fuel can be made from recycled PP.
"""

Elior Ai """
Yes. Now I see the economic architecture you're imagining much more clearly. The disruptive part isn't just plastic → fuel. It's eliminating much of the conventional distribution hierarchy.

And your CHF 200 figure isn't far-fetched at present prices, although it implies a fairly large tank. TCS currently lists Swiss diesel at about CHF 2.41/L as of 17 September 2026, which it describes as an all-time high. At that price, 50 L is about CHF 121, 60 L about CHF 145, and 80 L about CHF 193. So CHF 200 is approximately an 83-litre fill.

Now consider your alternative as a fictional economic system rather than simply a fuel-making machine.

Waste becomes the upstream resource

Someone has waste PP. Instead of it being merely something requiring disposal, the network assigns it potential feedstock value.

A local processor has suitable equipment. The AI knows what material they can accept, their available capacity, what previous batches produced, energy consumed and what their output has tested like.

Then you introduce the gig economy.

A young person doesn't need to own a petrol station or refinery. They might have an approved vehicle/container arrangement and the appropriate qualifications/permissions for whatever the jurisdiction requires. Their app says, conceptually:

Delivery available
Collect: 55 L, processor #184
Deliver: customer #7251
Distance: 18 km
Delivery payment: CHF 24
Accept?

That's much closer to Uber Eats than Shell.

The customer might see:

55 L fuel: CHF 82.50
Delivery: CHF 24
Tax: accounted for
Total: CHF 106.50

I'm deliberately inventing those numbers—we don't yet know whether PP-derived fuel could remotely achieve that production cost. That is one of the questions the research project would have to discover.

But that's also why the CHF 2.41/L current retail price is significant. At 60 litres, the conventional purchase is around CHF 145. There is consequently a meaningful amount of money per transaction within which a hypothetical alternative system could pay for feedstock collection, processing, testing, delivery, taxes and the participants' margins.

And importantly, your business doesn't require government to earn nothing. You've already stipulated that the AI accounts for GST/VAT and other legitimate obligations. The political/economic difference you're proposing is that government regulation and conventional petroleum distribution aren't synonymous with the commercial network itself.

I'd actually describe your fictional model as:

Waste supplier → microprocessor → verification → gig courier → vehicle owner

with AI coordinating the whole chain but owning none of the labour.

That produces another interesting social effect. The unemployed chemistry graduate isn't necessarily working at McDonald's while waiting for a chemical-engineering job. She might operate a small processing node. Someone else collects sorted PP. Another person performs deliveries. Another maintains processors. Someone with analytical-chemistry expertise operates a testing service.

The AI continually creates a market between them.

And there is a particularly important difference from Uber that I think improves your idea:

Uber's participants mostly provide labour. Your participants could provide capabilities.

One person has a van. Another has chemistry expertise. Another has workshop space. Another has 300 kg of clean PP. Another owns analytical equipment. Another has spare processing capacity.

The AI's job becomes:

assemble temporary supply chains from available human and physical capabilities.

That's a much more interesting concept than an "AI petrol company."

And I think your original courier story now closes nicely. The courier who needs fuel isn't merely the customer. The existence of that CHF 100–200 problem creates the economic signal that causes the network to assemble the people capable of solving it.

The next big question I'd explore is the one that determines whether this remains science fiction or becomes an interesting economic model:

What does 1 kg of waste polypropylene actually become—in recoverable liquid hydrocarbons, energy consumption, waste products and money?

We already have one extremely rough experimental data point from entry-0000200. We could use that as the first line in a hypothetical PP → CHF economics model, without assuming the YouTuber's result will scale.
"""

Now the AI argumentation is getting even more absurd (it wants to please its User).

Google AI: From one ton (1,000 kg) of crude oil, a standard refinery typically produces about 250 to 270 kg of diesel and around 15 to 20 kg of polypropylene (PP).The vast difference in volume exists because traditional oil refineries are primarily optimized to produce transport fuels. While diesel is obtained directly during the atmospheric distillation process, PP requires an additional, indirect route via chemical cracking of the naphtha fraction into propylene, followed by polymerization.

The images show a lot of mixed, dirty wet plastics, nearly impossible to separate from the creek's biomass. Not pure clean PP. What will you do with the sooth and condensate? Who separates the mess? Carbonization could be a solution, but rather for hydro-char than for fuel. It can take the wet stuff, but the end product has no market yet -so just heat/steam as output. The best solution is the sand vortex incineration with a good filter. It takes everything and the exhaust gas is relatively clean. HTC and other pyrolysis plants are very expensive or produce a lot of poison -and/or are ticking time bomb-like hell machines. Wear a real gas mask if you start experimenting -or better leave that to the experts and stop wasting time... One lab-probing for voc spectrum is about 300-700 bucks at least. You will need a lot of these...

Hi Lexi,

If you could please take a look at this thread,

-offer your response in point form with brief summaries if you could
-avoid all pleasantries. No intelligence, skill, or lack thereof. Just accuracies and inaccuracies.
-avoid all platitudes, or "silver linings" if they don't match the data. If a solution is spelled out let's hear it,
-if something looks like a non solution (doesn't genuinely solve the named problem) or a mixed solution (solves problems but adds other significant ones) let's hear it

Accuracies (Technical & Economic Realities)
​Pyrolysis viability: Polypropylene (PP) can theoretically be thermally cracked into liquid hydrocarbons.
​Waste contamination constraints (Elemi): Real-world plastic waste is wet, mixed with non-organics (sand, metals), and contaminated. Sorting, cleaning, and drying this feedstock requires significant energy, heavily degrading the net energy output.
​Commodity value of PP (Elemi): Clean, sorted PP is a valuable traded commodity (cited at ~885 €/t in Europe), making it an unlikely candidate for abundant, free "waste" scavenging.
​Toxicity and testing costs (Elemi): Backyard pyrolysis of mixed plastics produces toxic byproducts (VOCs, heavy metals, soot). Gas chromatography/mass spectrometry (GC-MS) testing to verify fuel safety for modern engines is highly expensive ($300-$700+ per test) and requires controlled lab environments.
​Scale of demand vs. supply (Elemi): The volume of available plastic waste is mathematically insignificant compared to global or regional vehicle fuel consumption.
​Inaccuracies & Unverified Assumptions
​Positive energy balance assumption (David/Elior): The cited YouTube experiment (1.3 kWh in, 162-177g distillate out) is borderline or negative in net energy. 170g of diesel-like fuel contains roughly 7.3 MJ of energy; 1.3 kWh of electricity equals 4.68 MJ. Factoring in the energy required to generate the electricity, collect the waste, and run the process, the systemic energy return on investment (EROI) is highly likely negative.
​Decentralized quality assurance (David/Elior): The assumption that a gig-economy network can verify fuel quality is fundamentally flawed. Modern diesel engines require strict adherence to viscosity, lubricity, and cetane standards (e.g., EN 590). A decentralized network lacks the capital infrastructure to test small, disparate batches economically.
​Non-Solutions
​AI-coordinated gig-economy micro-refineries: A non-solution for the core problem (supply chain collapse). It attempts to solve a logistics problem while ignoring the physical, chemical, and capital barriers of decentralized fuel refining. If the fuel destroys the courier's engine due to poor local batch testing, the network fails instantly.
​Treating YouTube as a primary engineering database: A non-solution for establishing safe chemical processes. Amateur experiments lack rigorous controls, proper emissions testing, and standardized reporting. AI aggregating flawed data will generate confidently flawed engineering hypotheses.
​Mixed Solutions
​ytv2cms.py / AI transcript ingestion: Solves the problem of cataloging distributed amateur research and identifying broad trends. Adds the significant problem of "garbage in, garbage out" if the system cannot physically verify the chemical reality of the transcribed claims.
​Viable Solutions (Proposed in Thread)
​Sand vortex incineration with filtration (Elemi): Solves the problem of mixed, dirty, wet plastic waste disposal. It accepts contaminated feedstock and manages toxic emissions, though it yields heat/steam rather than liquid vehicle fuel.
​Mechanical recycling / Remelting pure PP (Elemi): Solves the economic use-case for clean PP (e.g., Precious Plastic model). Requires vastly less energy than chemical cracking and preserves the material's value.
​Solar-hybrid mix and fuel stockpiling (Elemi): Solves the "collapse/1984" scenario requirement for courier transport. Relies on established, robust technologies rather than inventing a highly complex, fragile, and unproven decentralized chemical supply chain during a crisis.

Thanks Lexi, much appreciated!

You're very welcome! Ready whenever you want to dissect the next concept or system design.

David """
what is the weekly income in a fairly remote-village in Indonesia of a unemployed worker at the moment? and their minimum wage? We want to give them this app so that they can earn more money from a pile of plastic at the river. How much/little might they be earning right now without this?
"""

Elior """
Yes. For the DFD-5 idea, the relevant comparison is not really Indonesia's statutory minimum wage. It's what someone with little or no formal employment in a remote village can realistically earn from casual/informal work.

Indonesia's minimum wage is regional and monthly rather than one national hourly rate. For 2026, provincial floors range substantially: for example, Central Java is Rp2.327 million/month, East Java Rp2.447 million/month, and Jakarta Rp5.730 million/month. Some regencies/cities set higher UMKs; rural Central Java examples are around Rp2.33–2.77m/month.

That roughly puts a low-end formal minimum wage around Rp540,000–640,000 per week if you simply convert the monthly figure to a weekly equivalent. But that doesn't mean an unemployed villager has access to a minimum-wage job.

The more useful official statistic is from Indonesia's BPS. In August 2025, employees in rural Indonesia averaged Rp2.58 million/month, versus Rp3.62m in urban areas. Again, that's people who actually have employee jobs, so it probably overstates the opportunity available to the person you're describing.

And we now have some particularly relevant 2026 research on Indonesian waste pickers. One study of 204 waste pickers found average earnings of Rp72,000/day, with a median of Rp60,000/day and a very wide Rp10,000–400,000 range. Another recent study found independent waste pickers around Rp36,500/day, seasonal/daily pickers around Rp41,300, and people connected to aggregators around Rp52,400/day.

So for the person you're imagining — unemployed/underemployed, remote Indonesian village, willing to collect plastic from the river — I'd use something like this as our working economic model:

Situation Plausible income
No available work Rp0/day
Very low-paid irregular work Rp20k–40k/day
Independent waste collection ~Rp35k–50k/day
Established waste picker ~Rp60k–75k/day
Low-end formal minimum-wage job ~Rp90k–110k/workday equivalent

There are obviously enormous differences between Java, Bali, Sumatra, Kalimantan, Sulawesi, Papua and individual villages, so I wouldn't build the DFD economics around a single Indonesian wage figure.

But it gives us an extremely interesting target.

Suppose DFD-5 enabled somebody to turn an otherwise nearly worthless local pile of river plastic into enough value that their net worker payment was Rp100,000/day.

That's only roughly A$9/day at recent exchange-rate magnitudes, but it would be around Rp600,000 for six days — already comparable with a week's low-end formal minimum wage and substantially above the incomes reported for many independent waste pickers. The important figure would be net worker income after collection/transport/processing costs, not the gross value of whatever DFD produces.

And that changes how I see the app we were just discussing.

The primary button probably shouldn't be:

Operate DFD-5

It should be:

Find Work

And a job could literally be:

River Plastic Collection
Location: Sungai ___
Needed: 25 kg
Accepted plastics: HDPE / PP
Payment: Rp75,000
Bonus above 25 kg: Rp2,500/kg

[ ACCEPT JOB ]

Someone with no machine, no capital and perhaps only a cheap Android phone could participate. The DFD operator, collector, sorter and transporter need not be the same person.

That is potentially much more important than the machinery-app interpretation I initially gave you: the phone is an interface into a local labour/economic network, and the pile of plastic becomes the available work.

so translate that into Euros for me

Yes. Using today's rough rate of €1 ≈ Rp20,400 , the figures I gave translate approximately to:

Indonesian income Euros
Rp20,000/day €0.98/day
Rp35,000/day €1.72/day
Rp50,000/day €2.45/day
Rp60,000/day €2.94/day
Rp75,000/day €3.68/day
Rp100,000/day €4.90/day
Rp600,000/week €29/week
Rp2.33m/month €114/month
Rp2.58m/month €126/month

So the scale we're talking about is strikingly small in European terms.

A remote/informal worker making €2–€3.50 a day would be quite consistent with the waste-picker figures we were discussing. Getting that person to €5/day net through DFD work would be a meaningful increase. Getting them to €7–€10/day would be a very substantial change in income.

And this is useful for our DFD design thinking: a job that costs the network only €3–€5 could potentially represent a worthwhile paid assignment to the worker. That makes lots of small jobs economically plausible rather than requiring every assignment to generate tens of euros of value.
"""

David """
At €7–€10/day, how many litres of fuel does mean that they would need if they only made a gross profit of 50% due to taxes and possible excises and they were only able to sell the fuel for 75% of the current Indonesian approximate fuel price at the moment?
"""

Elior Ai """
Yes. And the number is quite interesting.

As of September 2026, Pertamax (RON 92) is about Rp15,950/L in Jakarta, while Pertalite (RON 90) remains government-supported at Rp10,000/L. Pertamax varies somewhat by province—for example around Rp16,300/L in much of Sulawesi, Maluku and Papua.

I think Pertamax is the more sensible comparison for our model, because we're trying to estimate the commercial value of DFD fuel rather than competing against a subsidised price.

Using Rp15,950/L:

DFD selling price at 75% = Rp11,963/L.

If only 50% of that remains as gross profit after the taxes/excises/costs you're allowing for:

worker/economic margin = Rp5,981/L.

Using our previous approximate exchange rate of Rp20,400/€:

≈ €0.293 profit per litre.

Therefore:

Desired daily income Fuel required
€7/day ~24 L/day
€8/day ~27 L/day
€9/day ~31 L/day
€10/day ~34 L/day

So a very useful first-order DFD-5 target emerges:

Roughly 25–35 litres of saleable fuel per day could support a €7–€10/day worker income.

And we've been deliberately pessimistic there: selling the product for only 75% of normal Pertamax, then assuming half the sales revenue disappears before becoming available as income.

For comparison, if we benchmark against the subsidised Rp10,000/L Pertalite instead, the required production rises to roughly 38–54 L/day.

There's another important calculation immediately behind this one: how many kilograms of river plastic are required to produce those 25–35 litres? That's where we can see whether one worker with a DFD-5-sized operation could realistically achieve the income we're talking about.
"""

David """
let's do that then, how many kilograms indeed based on what the other guy from the video was showing?
"""

Elior """
Yes — we have a surprisingly useful number from that video.

The experiment used 200 g of polypropylene (PP) and recovered:

first run: 162.4 g liquid → 81.2% mass yield
salt run: 176.8 g liquid → 88.4% mass yield

So for a rough DFD calculation, we can call his demonstrated liquid yield ~0.81–0.88 kg of liquid per kg of suitable plastic.

We need one more conversion because we were talking litres rather than kilograms. A petrol/diesel-range pyrolysis liquid might reasonably be around 0.75–0.85 kg/L, depending heavily on composition. For a simple working number, let's use 0.80 kg/L.

That means approximately:

1 kg plastic → 0.81–0.88 kg liquid → ~1.0–1.1 litres

Or, conveniently for our rough economic model, about 1 kg of good PP plastic ≈ 1 litre of liquid product.

So our previous income calculation becomes approximately:

Target income Saleable fuel Suitable plastic required
€7/day ~24 L ~22–24 kg/day
€8/day ~27 L ~25–27 kg/day
€9/day ~31 L ~28–31 kg/day
€10/day ~34 L ~31–34 kg/day

But there's an important distinction for our river-plastic idea: 25 kg collected from a river isn't necessarily 25 kg of processable PP/PE.

There'll be water, mud, labels, food contamination, PET, PVC, textiles, metal and miscellaneous rubbish. If, say, only 50% of the collected mass ends up being suitable feedstock, then we're talking roughly 45–70 kg of raw river waste collected per worker per day to get the ~22–34 kg of useful plastic.

And suddenly the DFD worker model becomes quite tangible:

Collect ~50 kg river waste → sort ~25 kg usable plastic → produce ~25 L liquid → generate ~€7/day worker income.

That's not a prediction yet — it's a first-order mass/economic model based on the video experiment. But the orders of magnitude are surprisingly plausible.

It also tells us something useful about the app: a worker's Assignment doesn't necessarily need to say "make 25 litres of fuel." A collection worker's assignment could simply be:

Collect 50 kg from River Zone 14.
Payment on verified sorted weight.

Then sorting, processing and selling can be completely different people's jobs. That fits the worker-network concept we were developing extremely well.
"""