Building an Open Space Situational Awareness Market

[TheRadicalModerate]

We currently have several different space situational awareness (SSA) datasets, of varying quality.  There are public datasets, like that provided by USSPACECOM and the ESA's SSA Programme, along with private entities like LeoLabs.

The public systems are generally accessible but the curation leaves something to be desired if you want to operate your constellation based on their data.  The private guys make money off of subscriptions (LeoLabs seems to be charging about $2500/month/satellite) and provide better-curated data, with better precision and quality.

What I'd like to explore in this thread is this:  Is there a way to aggregate SSA providers into an open market, so that the public at large gets the benefit of an open, cleanly curated, high-precision dataset, while the providers are able to make a nice profit, giving them incentives to expand their capacity?

I'm sure that the private guys would much rather just duke it out and manage their own customers.  But no single provider is likely to provide comprehensive SSA as the orbital environment gets more crowded, so finding a way to disaggregate the business so that it scales would be a public good.

Ultimately, I believe that there will be international agreements that regulate what operators have to do to avoid collisions.  Almost certainly, one component of those regulations will be a requirement either to provide their own SSA or subscribe to an SSA service that can give them state vectors for their birds and alert them of potential conjunctions.  I guess what I'm suggesting is that the subscription be to a public entity that intermediates the combining of tracking data, discovery of new objects, computation of state vectors from the tracking data, and the computation of collision possibilities.

I keep thinking that an open market for this information looks a little bit like the semi-open markets that most electricity independent system operators operate, which integrate electrical generators, long-haul transmission, and distribution systems, while maintaining some kind of control to ensure an orderly market.  (As anybody who lived in Texas last winter will tell you, some of these controls are more effective than others.)

On-topic posts will cover ideas on:

1) The requirements for such a system.
2) Its architecture.
3) What kinds of providers are needed and how they interact.
4) How the market would work.
5) What sorts of pathologies might mess up the market.
6) Whether the incentives built into the system are enough to ensure that there are enough SSA service providers to provide safe operation in the space environment.

[TheRadicalModerate]

Some ideas on requirements and characteristics:

1) It could be used as a clearinghouse for future national or international regulations.

2) It would be able to scale up to at least a billion objects in earth orbit. (Is this enough?)

3) Its accuracy would be easily auditable by its customers.

4) The buyers of information are the operators and the sellers are various kinds of providers.

5) It would enable operators not to worry about who was providing the data if they didn't want to, but still allow them to bid for services in an open market.

6) It would be able to operate in an emergency (e.g. after a collision with new fragments or, in the worst case, during a war).

7) It would encourage providers to search for new objects.

8) It would encourage providers to add value by increasing resolution and accuracy.

9) It would cause capacity to be added as needed.

10) It would alert operators of potential conjunctions, and provide an incentive for providers to win management of a conjunction by providing earlier or more accurate information.

11) It would integrate with range operators to provide data on safe insertion of new objects.

12) It could scale to handle cislunar and possibly interplanetary SSA.

13) There should be only a nominal cost for any person or organization to get access to the database.  Note that this opens up a free-rider problem, so it would be essential that national regulations, as prescribed by international law, would require operators to subscribe to the system and provide proof of SSA coverage.

[TheRadicalModerate]

Here's an arm-wave at a centralized architecture.  (This might be an application where a blockchain implementation might solve a much of diplomatic problems resulting from nations mistrusting the public operator, but I have no idea how you'd do curation in that environment.)

There's a public system, a set of private providers of various services, and satellite and range operators as the consumers.  See attached.

PUBLIC SYSTEM:
Curator:  This is a blob of software and humans that is responsible for data quality, acceptance of state vector information for inclusion in the database, and tasking out to the various service providers.

Database:  It contains all the known objects, their assigned IDs, their owners (the curator would own objects that were rogue, unclaimed, of unknown origin, or launched under the cover of some kind of military secrecy), and their current curated state vector.  If I had my druthers, this information would be public and free, but that kinda depends on how efficient the market is.

Market:  A blob of software and humans that matches up the bids for service from the operators with the asks from providers themselves.

PROVIDER SYSTEM:
Tracking Service:  This is a radar, a telescope, or an air- or space-based sensor platform that can provide raw tracking data on a particular set of objects, as requested by the curator.  For a radar at a known location, this would be TEARR data (time, elevation, azimuth, range, and rate).  You'd obviously want an interchange format that's flexible enough to deliver all relevant data, but TEARR seems like the bare minimum.

Object Discovery Service:  This is a service that looks for uncatalogued objects and reports them.  A tracking service could provide an adjunct discovery service, so that it could use its untasked time to look for new objects.  The public system wants to use the market to put a decent bounty on discovering new objects, so that there's an incentive to look.

Orbit Computation Service:  A hunk of compute power that can take data from a tracking and/or discovery service to compute a good state vector.

Conjunction Discovery Service:  Another hunk of compute power that has direct access to the database or a change log from a database snapshot, which it uses to search for potential conjunctions.  This also has to be bounty-based, to a certain extent.  I suspect that the curator has to divvy up tasking to ensure that all objects are being searched all of the time.

CONSUMER SYSTEM:
Spacecraft Operator:  Ideally, all objects are "owned" by a spacecraft operator.  In practice, there will be some orphans.  Spacecraft operators register their objects, agree to have them covered by at least one tracking provider and one collision discovery provider, and agree to pay the market price for those services.

Launch Operator:  Launch operators need to clear new orbit insertions before launch.  Haven't thought too much about this one.  I suspect that they'll need to provide their best guess at the insertion state vector, then get a conjunction service to verify that they're good to go.

That's what I have so far.  Note that there's nothing to preclude a single organization subsuming several different provider roles, nor does it preclude operators from running their own provider components and bidding using them to provide the regulatory required services.  The market has to be flexible enough to allow stuff like that.  However, if this got done properly, I suspect it would encourage specialization of roles, which would in turn likely improve the overall quality of the database.

[leovinus]

We currently have several different space situational awareness (SSA) datasets, of varying quality.  There are public datasets, like that provided by USSPACECOM and the ESA's SSA Programme, along with private entities like LeoLabs.

The public systems are generally accessible but the curation leaves something to be desired if you want to operate your constellation based on their data.  The private guys make money off of subscriptions (LeoLabs seems to be charging about $2500/month/satellite) and provide better-curated data, with better precision and quality.

Database:  It contains all the known objects, their assigned IDs, their owners (the curator would own objects that were rogue, unclaimed, of unknown origin, or launched under the cover of some kind of military secrecy), and their current curated state vector.  If I had my druthers, this information would be public and free, but that kinda depends on how efficient the market is.

Interesting thought and I'd like to know more. As you obviously put some thought in this, and the quality of the data would determine most of the success of such an effort, can you elaborate a bit please on what is current available?

To start with, related to the two quotes about one aspect, data:

What is available?

You mention "There are public datasets". That is a great start but could you share the pointers to data ? Also, what insights do you have on the quality and content of the public data? What are those datasets doing right or wrong? Are they 50%, 90% or 99% complete for objects since Sputnik? Are the orbits and state vectors stale or current? In other words, orbits in terms of "I measured the parameters once last year and assume they will be unchanged" or "based on 10 year track of object X with N measurements we can do predictions a year ahead? "? Are the current database just N objects with M orbital parameters, or is there an orbital model with drag etc involved?

What quality would be necessary?

Based on the perceived use of the database, how complete and accurate does the data have to be? As an example, to predict the collision probability of two satellites a week, a month or a year in advance, you need a certain quality of the orbits. How is that quality measured and is it achievable based on public data?

Fundamental limitations?

Do you or anyone know how to quantify the quality difference between private operators and the public data? Could the public ever obtain quality to rival the private operators or do you need some other in-house, private measurements which the public will never see?

Maneuverable vs static objects?

Starlink and some military satellites can move around and without some data acquisition, a public dataset would always be "stale".

Finally, I wondered about parallels in other industries such as financial data where the privates (Bloomberg, Reuters) make money in contrast with some unusable public data I never liked for predictions. Even the private data needed curation at times. And in the AI/ML market,  academic datasets are available but the giants have datasets several order of magnitudes more which makes it harder to rival them.

[leovinus]

And I forgot - What use-cases do we know for such a dataset?

Setting aside military applications, as they have their own, I see an astronomical use-case as such data could be useful to predict and remove satellite tracks from images and signals. Maybe disaster relief "which imaging satellite is closest for an image now?".Weather related like " I need a N hour track of approaching storm images and provider X cannot give me that but maybe provider Y plus Z can do the trick together? 

[TheRadicalModerate]

Interesting thought and I'd like to know more. As you obviously put some thought in this, and the quality of the data would determine most of the success of such an effort, can you elaborate a bit please on what is current available?

To start with, related to the two quotes about one aspect, data:

What is available?

You mention "There are public datasets". That is a great start but could you share the pointers to data ? Also, what insights do you have on the quality and content of the public data? What are those datasets doing right or wrong? Are they 50%, 90% or 99% complete for objects since Sputnik? Are the orbits and state vectors stale or current? In other words, orbits in terms of "I measured the parameters once last year and assume they will be unchanged" or "based on 10 year track of object X with N measurements we can do predictions a year ahead? "? Are the current database just N objects with M orbital parameters, or is there an orbital model with drag etc involved?

First, I'm just a semi-motivated amateur with a software architecture background, so hopefully what I get wrong and/or don't know will get corrected by people who know more for-real.

The granddaddy of SSA databases is the one maintained by USSPACECOM.  There's a public interface to that here.

My understanding is that this is fairly up-to-date, with new objects' TLEs being entered fairly quickly.  The complaint I've heard about this is that old objects can get quite stale before their TLEs are updated, so it's not exactly at the level of quality that you need for accurate conjunction alerting and avoidance.

I don't have any experience with the ESA database (I have only passing experience with space-track).

To my knowledge, there's no attempt to propagate TLEs forward in time using an orbital model and update the database.  However, because it's public, I'm sure that there are third parties that have models that they use.

What quality would be necessary?

Based on the perceived use of the database, how complete and accurate does the data have to be? As an example, to predict the collision probability of two satellites a week, a month or a year in advance, you need a certain quality of the orbits. How is that quality measured and is it achievable based on public data?

Someone else can almost certainly answer this question better than I can.  Conjunction alerting is an iterative process.  If you propagate a couple of TLEs out of whatever database you're using, you can start to predict when a conjunction might occur.  Based on that, I assume that you'd task a lot of radars with measuring the bejeezus out of the two participants, which would give you a better fix on how much trouble you're in.  Do that a couple of times and you'll get to the point where somebody needs to decide whether to maneuver or not.  Post-maneuver, you need to measure further until you can close the event out.

My conclusion is that this means that the database only has to be reasonably accurate, as long as you can task additional tracking when you find something troubling.

In a brokered system, I'd think that precision, both advertised and reputational, would be a reason why you might want to ask a higher price for your data.

Do you or anyone know how to quantify the quality difference between private operators and the public data? Could the public ever obtain quality to rival the private operators or do you need some other in-house, private measurements which the public will never see?

I can't quantify the difference.  However, ISTM that there might be a way to get very high-quality public data.  There's sort of an asymmetry in the data that it's acquisition is incredibly valuable to one entity:  the owner of the object being tracked.  But once the state vector is computed, there's very little motivation to keep it proprietary, and it's pretty much to everybody's advantage to have it public.  Collision avoidance is not a competitive task; it's a cooperative one.  So the providers compete to provide the data in the first place, but once they have, cooperation is better.

The hitch here is in forcing the operators to get the data in the first place.  Sloppy operators simply might not care that much about collisions.  That can't be allowed to happen.  So you have to require them to buy some minimal standard of SSA and collision alerting, and act on the alerts.  In most cases, though, operators want to take good care of their constellations.  The trick is to put a floor on how "cheap" you can go with your SSA, and then let the market figure out where the upper bound on quality should be.

Maneuverable vs static objects?

Well, there are objects that can maneuver and objects that can't.  (If they were static, they'd be on the ground.)  That obviously has to go into any automatic system that's trying to optimize maneuvers.

Starlink and some military satellites can move around and without some data acquisition, a public dataset would always be "stale".

Somewhat stale but not completely.  First of all, I think that operators may have a way to inform USSPACECOM about maneuvers, or to post their own TLEs for curation.  Second, this is what SpaceCom does for a living:  they track stuff.  Most of that stuff becomes public.  I wouldn't be surprised if they're obfuscating the accuracy on purpose, but the mission is to protect civil space assets in addition to military ones.  You can't do that without a pretty decent consensus on the SSA.

Finally, I wondered about parallels in other industries such as financial data where the privates (Bloomberg, Reuters) make money in contrast with some unusable public data I never liked for predictions. Even the private data needed curation at times. And in the AI/ML market,  academic datasets are available but the giants have datasets several order of magnitudes more which makes it harder to rival them.

I think that is probably the situation we have today.  It'll likely continue into the future.  But just like there can be private financial transactions, the dataset for the market-based transactions is now incredibly robust and intermediated sensibly.  That's kinda what I think we should be aiming for.

And I forgot - What use-cases do we know for such a dataset?

Setting aside military applications, as they have their own, I see an astronomical use-case as such data could be useful to predict and remove satellite tracks from images and signals. Maybe disaster relief "which imaging satellite is closest for an image now?".Weather related like " I need a N hour track of approaching storm images and provider X cannot give me that but maybe provider Y plus Z can do the trick together? 

Definitely the astronomy thing is a knock-on use case.  I'd think that most earth observation companies would have better ways to posting their most recent stuff than their state vectors.  But if you're going to have a public facility, its first and foremost goal should be to prevent collisions.  Anything above and beyond that is just gravy.

[su27k]

Sorry, I didn't read everything in this thread, but just in case you missed it, have you checked out Space Policy Directive-3's Open Architecture Data Repository (OADR)? Here's some links:

https://trumpwhitehouse.archives.gov/presidential-actions/space-policy-directive-3-national-space-traffic-management-policy/

https://govtribe.com/file/government-file/docnoaafy21oscspd3openarchitecturedatarepository-osc-oadr-industry-day-slidedeck-nov2020-dot-pdf

https://space.nss.org/wp-content/uploads/NSS-Position-Paper-Six-Recommendations-2020.pdf

https://spacenews.com/op-ed-noaa-is-stalling-u-s-space-traffic-management/

https://spacenews.com/op-ed-commerce-department-noaa-ensuring-u-s-remains-a-world-leader-in-space-commerce/

[TheRadicalModerate]

Thanks, I hadn't seen a bunch of those.

Two things that jumped out at me:

1) The Commerce Department seems to be adopting a model where it will qualify certain providers to be inside the tent, and there's really no way to commoditize the services that they offer.  The problem with that is that it doesn't scale very well.  I suspect the big bottleneck will ultimately be sensors, because the catalogue of objects to be tracked is going to grow very quickly.  If you have to convince one of the company's in the tent to build out another radar, getting the business case to close is fairly complex.  On the other hand, if you use a market-based approach where the qualification bar is very low, supply and demand should provide better incentives for providers to build out, or for new providers to enter and pick up the slack.

2) I don't understand the issues associated with the DoD keeping parts of the catalogue classified.  It makes me wonder if there's some stealth tech whizzing around up there.  It seems like a bad policy to keep the state vectors of certain objects hidden, especially since most of them have likely been discovered via other means, or by the nations from which you'd like to hide them.  Declassifying that stuff, possibly with a bit of noise to keep kinetic ASAT solutions from being a slam-dunk, seems like a policy opportunity.  More importantly, it might be a diplomatic opportunity to get Russia and China to do the same.

[D_Dom]

Fantastic thread, thanks for posting all your hard work. What comes to my mind is how useful this dataset will be if attempting to capture some of this material for reuse. After all it is already halfway to anywhere else in the solar system once it is in LEO

[dondar]

Interesting questions regrettably ignored by the space related mass-media

"Outer space" is regulated by the United Nations office for outer space affairs.
https://www.unoosa.org/oosa/en/ourwork/spacelaw/resolutions.html
there are specific resolutions concerning liability (results of the sat collisions with space or ground objects) and registration (notifying other countries about your actions in space). It is all done by country vs country basis. Liability is direct only, i.e. you are not liable if the pieces of your satellite push other satellites to move, you are liable if they damage other satellite.
The UN has it's own registration database (you can it find from the link above). This database is used by SpaceTrack team to identify and label objects they find.
SpaceTRack is a military team which aggregates and disseminates data to interested parties (celestrak is using their datasets as basis. Everybody does).
They obtain data from Space Surveillance Network. The network is using radars for tracking and detecting and the (electro)optical telescopes for post identifying objects.
(electro) here means (digital) registration capable.
This is the military command and the actual technical capabilities is a state secret.
Actual physical limitations (it's not fully automatic) mean that the real time detection/identification is not possible and the team focuses their attention on "scary" objects (not communicating tumbling satellites which are changing orbits in not very predictable manner). Hence boring objects (like Starlink satellites) can have delayed updates by weeks.

I don't know the name and even the precise timeline of the establishing this system, but it is good, it is still improving and it's accessibility to all of us  is an amazing gift.

I remind that it's initial purpose was obviously detecting and identifying direct ballistic threats to the US and their allies.
edit: ingrlish.

[leovinus]

Definitely the astronomy thing is a knock-on use case.  I'd think that most earth observation companies would have better ways to posting their most recent stuff than their state vectors.  But if you're going to have a public facility, its first and foremost goal should be to prevent collisions.  Anything above and beyond that is just gravy.

Thanks for the pointers and data thoughts! Based on your link towards https://www.space-track.org I did some musing about this. The result was almost another Alice in Wonderland down-the-rabbit-hole-we-go experience but great fun.

If the purpose is predominantly for collision avoidance then you need better data and simulatons. A couple of more thoughts here on the data which hopefully is well within the thread scope of "building a market".

In general, I learned a bit more about the Two Line Elements (TLE) used to describe the state and orbits. Fascinating to see that the format goes back to good old punchcards and FORTRAN code. Drag related, I dove into the simplified perturbations models [9] and reference code in C++ via the Wiki. While great to see so much reference, the C++ codes at [1,2] could use some TLC as e.g. anglesgauss() in astiod.cpp has obvious errors (array out of bounds). Could be a left-over of the FORTRAN to C++ conversion and there might be newer reference code around but I did not see.

Data

While the main way to improve the data accuracy is via tracking stations, and an OSS system does not have that, then other approaches to add value to the data are necessary.

In particular, the TLE format is old but offers already terms for Drag for SPG4 and 1st and 2nd order Derivative of Mean Motion. While that makes sense, maybe a more 'modern' probabilistic uncertainty measure could be added? You know, like new experience from 50 years leads to new fields in the format? Like a plain, yearly orbit change or something? And adding extra fields and info for better simulations, low and high drag, might be useful.

Here is my mental model. An orbit model where the orbit is a zero dimensional object thread in ellipsoid form is too simple. To take into account measured or modelled orbit changes over time, the TLEs already have the extra fields. In the case of two objects in different orbits, a collision measure based on a zero dimensional thread is too simple. Hence, I visualize first a traffic cone, then that the center of the cone is the most likely position in time, some distribution to model position certainty toward the cone edges, then imagine the cone stretched to 40,000 km, and curved as the ellipsoid, and then calculate for degrees of overlap between those two object cones. Whether uncertainty modelled with TLE or other fields, that provides a neat summary and probabilities to predict collisions into the future.  And with O(10^4) objects looking for the pair-wise closest ones, that is computationally tractable.

Another neat question would be whether there is already modern literature about TLE extensions and there usefulness? Maybe some expert can pitch in?

And, could we do better these days than SGP/SPD 4/8 ? A different math model and more accurate numerical, differential simulation (which might not be necessary but just wondering). CFD also came a long way in last decades.
 
Code

Another way to add value is to let go of the pure data repositories like [3-8] and start adding code in the OSS spirit. There was one Python tool [13] that actually fetched data from [3] to calculate stuff. And [10-12] do something like that to visualize. But you could combine perturbation models, or even other models and calculations to add value. A newer C++ or Python code repo in arbitrary precision might be useful as well.

Based on such data and code, the OSS system can probably answer questions like "what launch time from the Cape does minimize my collision probability with any objects in orbit?"

Have a good weekend.

[1] https://github.com/dvallado/SGP4-Microsoft-Visual-Studio-Conversion
[2] http://celestrak.com/publications/AIAA/2006-6753/AIAA-2006-6753.zip
[3] https://www.space-track.org/documentation
[4] https://celestrak.com/NORAD/elements/
[5] http://www.satobs.org/
[6] https://www.planet4589.org/space/elements/
[7] https://web.archive.org/web/20000301052035/http://spaceflight.nasa.gov/realdata/sightings/SSapplications/Post/JavaSSOP/SSOP_Help/tle_def.html
[8] https://www.orbtrack.org/
[9] https://en.wikipedia.org/wiki/Simplified_perturbations_models
[10] https://stltracker.github.io
[11] https://www.qsl.net/kd2bd/predict.html
[12] https://www.heavens-above.com
[13] https://github.com/python-astrodynamics/spacetrack

[TheRadicalModerate]

Another way to add value is to let go of the pure data repositories like [3-8] and start adding code in the OSS spirit. There was one Python tool [13] that actually fetched data from [3] to calculate stuff. And [10-12] do something like that to visualize. But you could combine perturbation models, or even other models and calculations to add value. A newer C++ or Python code repo in arbitrary precision might be useful as well.

This is why I think that you partition orbit and conjunction computation from tracking and discovery.  If somebody has a model that better predicts the locations of existing objects when they're next tracked by sensors, then selling that as a service to the public system gives the provider of that service a competitive advantage.  Similarly, if there are less compute-intensive solutions, offering a lower price should attract more tasking for the provider offering it.

One thing (of many) that I'm confused about is whether the connection between consumers (operators) and service providers should be opaque (i.e., when the market matches a consumer's bid with a provider's ask, the provider is unknown to the consumer) or transparent (where the market facilitates matching consumers and providers, but the pair know each other and may have a relationship where they interact through more than the public system). 

Opacity makes things simpler, but I think it's probably unrealistic.  Operators' trust in the public system will only go so far, and they'll likely want to be able to verify and curate the results they get from a provider through some method that can check the curator's judgment.

The important thing is that the public at large benefits from the interaction between the consumer and the provider.  As long as clean results wind up in the database, either approach will work.

[TheRadicalModerate]

Jon Goff has a post on how to spin up a joint international debris remediation effort.

ISTM that if you're going to build an open SSA framework, attaching a remediation framework onto it makes a fair amount of sense.  I've updated my picture to incorporate that.

[Asteroza]

The IAU wanting to spin up a public SSA for megaconstellations is sorta related...

https://forum.nasaspaceflight.com/index.php?topic=48302.msg2301227#msg2301227

[TheRadicalModerate]

The IAU wanting to spin up a public SSA for megaconstellations is sorta related...

https://forum.nasaspaceflight.com/index.php?topic=48302.msg2301227#msg2301227

More than sorta.  If you have a reliable, high-precision ephemeris of everything that will saturate a CCD, you can either plan your observing runs to avoid objects, or you can build detectors that shut down the pixels over which the object will track.  (Note that this second one is tech that doesn't yet exist, but IMO it's The Permanent Answer To The Problem.)

I suspect that if you have a reliable public database, there are lots of applications beyond traffic management for it.  But this is definitely a biggie.

[TheRadicalModerate]

5) What sorts of pathologies might mess up the market?

The orbital data in the early time period just after a launch is not well defined on publicly available datasets.  Witnessed a lag of orbit data on the order of weeks once the objects actually do appear.  non-US objects are worse.

Tasks like this are similar to staring at glaciers, except glaciers are more fun.

The operators obviously don't wait weeks for the data, so it's out there somewhere.  The question is how to incentivize whoever has it to put into the public domain quickly and accurately.

The big pathology is that somebody may be making money off providing the data to the operator.  For instance, if an operator contracts with LeoLabs for tracking data and orbital computations, LeoLabs doesn't want to publish that info to the public until its value to the operator has been exploited.  Similarly, national security payloads are more vulnerable and more image-able in LEO than they are in a higher orbit, so keeping their orbital elements away from adversaries early in their life is highly desirable.  So national tracking systems have a vested interest in holding off publishing data, at least for a while.  And they likely have an interest in publishing only low-resolution data forever.

Whatever market you create for this data has to recognize that value of the data declines sharply with its age, which allows the owner of the data eventually to provide it free to the public.  But at the same time, the public will occasionally need prompt, hi-res data to avoid a collision.  Finding a way to resolve the tension between the need for trackers to make money (or to preserve national security data) and the public to have a good overview is what's tricky.

[Asteroza]

In light of the recent ASAT test, how would you envision dealing with sudden scenarios where debris suddenly increases because something pops (or gets popped)?

[Vultur]

What kind of instruments are used to track debris anyway?

I know big debris (like old upper stages) is easily visible. Is smaller debris tracked with optical telescopes, or radar, or what?

Intuitively it seems like one could find fairly small objects with small optical telescopes, given that Starlinks are sometimes visible to the unaided eye and are not that big (a couple hundred kg?)

What size telescope would it take to spot say a dead cubesat, at a similar altitude?

[Asteroza]

SSA space radars up until recently were generally capable of monitoring 3cm debris more or less. Whether they can get the full state vector for 3cm debris regularly is a different problem. These space fence radars have a wide swath but they generally can only track when something crossed the "fence" in what location and altitude. Newer radars are more modern phased array SAR's, so they "tilt" the fence periodically to get two radar hits on objects to determine a vector.

Followup with optical telescopes is a numbers game, meaning if you had sufficient numbers of telescope tasked to monitor the more hazardous 1-10cm range objects you could pull it off, but you need an enormous number of them (not just due to raw eye ratio to objects needed for tasking, but also weather blocking any given telescope).

I suppose a high frequency radar located in very high LEO or MEO might be an interesting alternative, to improve small object monitoring. Would be a great opportunity for archinaut/spiderfab type ventures to build a monster radio parabolic antenna to improve sensitivity.

[Vultur]

Followup with optical telescopes is a numbers game, meaning if you had sufficient numbers of telescope tasked to monitor the more hazardous 1-10cm range objects you could pull it off, but you need an enormous number of them

What size telescopes would they be?

I doubt one can just scale it linearly since visible small satellites have solar panels and stuff so are presumably more visible than a chunk of metal of the same mass.

So what size optical telescope would it take to see a 10cm object (eg. dead 1U cubesat)? What about a 2cm fragment?

And what size object could the ISS survive? That might be the more important question long term (plus other human spaceflight platforms like Tiangong/CSS or Commercial LEO Destination stations) if LEO is going to lots of relatively cheap satellites so any individual loss is insignificant.