Thread 'KAMPYLOS'

Message boards : Puzzles and science : KAMPYLOS
Message board moderation

To post messages, you must log in.

1 · 2 · 3 · Next

AuthorMessage
Alperen Yavuz
Project developer
25 years old

Send message
Joined: 27 Sep 26
Posts: 349
Credit: 5,210,003
RAC: 137,017
Message 128 - Posted: 30 Sep 2026, 15:18:42 UTC
Last modified: 6 Oct 2026, 13:38:47 UTC

🔭 Kampylos update: fully in production now, and a deep dive into what it actually does 🔭

Back when I first posted this, Kampylos was beta, not sending real work yet. That's changed — it's been grinding through real archival data for a while now. Since a few people have asked what this project is actually computing and why it's hard, here's the long version.

🌌 What gravitational microlensing actually is

When a foreground star (the "lens") happens to pass almost exactly between us and a much more distant background star (the "source"), the lens's gravity bends the source's light. We're never close enough to resolve this as a visible ring or arc the way "strong" lensing around galaxy clusters looks — instead, what we actually see is the source star temporarily, smoothly brightening over days to months as the alignment improves, peaks, and fades. That's a microlensing event.

The powerful part: this works completely independent of whether the lens itself gives off any light. It's one of the only techniques that can detect things that are otherwise invisible — free-floating planets with no host star, black holes, faint or dead stars, or planets orbiting a lens star too dim to characterize any other way.

🪐 Why a BINARY lens is the interesting case

If the lens is actually two masses close together — two stars, or a star plus a planet — the light curve isn't the smooth single-bump curve anymore. The combined gravity of the pair creates a much more complex magnification pattern (caustics: curves in the source plane where a point source would be infinitely magnified), and if the source's path crosses or passes near one of these, you get sharp spikes, asymmetric double-peaks, or other distinctive features layered on top of the smooth baseline.

This is genuinely one of the most sensitive ways to find exoplanets that exist in the literature — sensitive to planets around stars far too faint or distant for transit or radial-velocity searches, and sensitive even to planets with NO visible host star at all (rogue/free-floating planets, inferred purely from a short, isolated microlensing blip).

The binary lens geometry is described mainly by two numbers: s (the projected separation between the two masses, in units of the Einstein radius) and q (the mass ratio, secondary/primary — for a star+Jupiter pair q~0.001, for a star+Earth pair q~0.000003).

🧮 Why fitting one of these is a genuinely hard computational problem

A single-lens fit is comparatively easy: a handful of parameters (time of peak, impact parameter, event duration), a smooth well-behaved curve, and it mostly converges from almost any reasonable starting guess.

A binary lens fit is a different beast:

  • There's no closed-form formula for the magnification in general — it has to be computed numerically (Kampylos uses VBMicrolensing, a real contour-integration-based library, the same one professional microlensing groups actually publish with — not a homemade approximation).
  • The (s, q) parameter space is riddled with local minima that can look deceptively like a good fit from one starting point. There's even a well-documented "close/wide" degeneracy in the literature where a lens with separation s and one with separation 1/s can produce very similar-looking light curves. Which means you cannot just locally optimize from a single guess and trust the answer — you have to systematically scan the whole plausible (log s, log q) plane, and at every single point on that grid, run a full local optimization over the remaining parameters (time of closest approach, impact parameter, event timescale, source trajectory angle, and the source-size parameter for finite-source effects), itself restarted from multiple random starting points so you don't even get stuck in a bad local optimum within one grid cell.



That's a LOT of expensive numerical evaluations per event — and public surveys (OGLE, MOA, KMTNet) have alerted on tens of thousands of these events since the 1990s. No single research group has the compute budget to run this exhaustive a search on every single one; in practice, published binary-lens analyses are done by hand, one at a time, for events that already looked promising at a glance. Which means the vast majority of the archive has genuinely never been exhaustively checked. That gap is exactly what this project exists to close.

⚙️ What Kampylos actually computes, step by step


  1. Pull one archival event's light curve (OGLE-IV EWS, MOA's alert archive, or KMTNet).
  2. Quick single-lens (point-source point-lens) pre-fit, just to anchor sane starting values for time/impact-parameter/timescale.
  3. Grid search over (log s, log q) — typically 10x10, spanning several decades of separation and mass ratio.
  4. At every single grid cell: a full local optimization (Nelder-Mead downhill simplex, multiple randomized restarts) over the remaining parameters, with source flux and blend flux solved in closed form for each trial point.
  5. CPU, CUDA and OpenCL backends all run the literal same algorithm — GPU versions just parallelize the many simultaneous trial evaluations within a batch, they don't change the math or the answer.
  6. The FULL grid — chi-squared and the complete best-fit parameter set at every single cell, not just whichever one looks best — gets published openly, deliberately. A "just the top hit" publication would hide exactly the information (how many near-ties exist, how distinct the true minimum actually is) that lets someone independently judge whether a flagged candidate is real or just this particular grid's best-of-a-mediocre-bunch.



📊 Where things stand right now

81 full-grid datasets already published openly on Zenodo, each with its own permanent DOI — 52 OGLE-IV, 29 MOA, KMTNet queued up next. Every single one is the complete solution map as described above. Browse them all: kampylos_published.php

Multi-platform now too: CPU, CUDA and OpenCL, across Linux, Windows, macOS (Intel and Apple Silicon) and Android — all running the identical, cross-checked search.

🔎 Honest anomaly tracking

Any event whose best grid point clears an automatic "this looks like it could be a real detection" threshold gets flagged and tracked publicly on its own page — win or lose, nothing gets quietly buried if it turns out to be a false positive once actually checked by hand: kampylos_discoveries.php. One strong-looking candidate (OGLE-2011-BLG-0952, delta_chi2=286) has already gone through this — turned out to be a systematic artifact once properly verified, not a real planet. That's still exactly the outcome this project is built to produce either way: a checked, honest answer instead of a guess.

🙏 What's left

Tens of thousands of archival events still queued across all three surveys. Every host running Kampylos genuinely gets us through them faster. Thanks for sticking with it through the beta days — questions, and bug reports, always welcome below.


ID: 128 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profiledskagcommunity
Bug hunter
Early supporter
46 years old
Avatar

Send message
Joined: 29 Sep 26
Posts: 64
Credit: 21,977,365
RAC: 136,334
Message 129 - Posted: 30 Sep 2026, 15:36:12 UTC

Very cool to see other applications/underprojects too. Keep up the good stuff :)
---------- 24/7 Crunching since 2011 -----------
DSKAG Austria: http://www.dskag.at

ID: 129 · Rating: 1 · rate: Rate + / Rate - Report as offensive     Reply Quote
ProfileSteve Dodd
Early supporter

Send message
Joined: 28 Sep 26
Posts: 27
Credit: 53,008,805
RAC: 218,076
Message 158 - Posted: 30 Sep 2026, 21:48:40 UTC

Ok. I can go for ANOTHER astrophysics application :))
ID: 158 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profiledskagcommunity
Bug hunter
Early supporter
46 years old
Avatar

Send message
Joined: 29 Sep 26
Posts: 64
Credit: 21,977,365
RAC: 136,334
Message 161 - Posted: 1 Oct 2026, 6:40:48 UTC

Perhaps its possible to get communication with dc stats again when you have such projects. because a credit export to the signature providers are important longterm ^^
---------- 24/7 Crunching since 2011 -----------
DSKAG Austria: http://www.dskag.at

ID: 161 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
boboviz
Ambassador

Send message
Joined: 29 Sep 26
Posts: 47
Credit: 1,634,482
RAC: 33,191
Message 162 - Posted: 1 Oct 2026, 7:28:51 UTC - in response to Message 128.  

In reply to Alperen Yavuz's message of 30 Sep 2026:
🚧 Status: in development, not sending work yet -- the search engine is built on VBMicrolensing (the standard tool real microlensing groups use for this, not something homemade) and is currently going through correctness testing. It's listed on the Applications page as beta so you can see it coming. I'll post here the moment real work units go out.

Thanks for your patience -- questions welcome below. 🙏


No problem.
We are here to crunch your wus!!
ID: 162 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
mmonnin

Send message
Joined: 28 Sep 26
Posts: 14
Credit: 20,205,102
RAC: 1,209,765
Message 194 - Posted: 1 Oct 2026, 19:35:42 UTC

All 13 of my tasks have come back Invalid

<core_client_version>8.2.5</core_client_version>
<![CDATA[
<stderr_txt>
Kampylos (gravitational microlensing binary-lens grid search) v1
bitboinc -- https://bitboinc.athena.org.tr/
coded by Alperen Yavuz
Loaded 148 data points; grid 1x10, log_s [-1.0000,-1.0000], log_q [-4.0000,0.0000]
PSPL anchor: t0=2456456.2807 u0=12.74318 tE=1.8408 chi2=188.883
Resuming from cell 0/10
[1/10] log_s=-1.000 log_q=-4.000 chi2=180.033
[2/10] log_s=-1.000 log_q=-3.556 chi2=179.689
[3/10] log_s=-1.000 log_q=-3.111 chi2=179.352
[4/10] log_s=-1.000 log_q=-2.667 chi2=179.143
[5/10] log_s=-1.000 log_q=-2.222 chi2=178.935
[6/10] log_s=-1.000 log_q=-1.778 chi2=178.800
[7/10] log_s=-1.000 log_q=-1.333 chi2=178.416
[8/10] log_s=-1.000 log_q=-0.889 chi2=183.037
[9/10] log_s=-1.000 log_q=-0.444 chi2=178.700
[10/10] log_s=-1.000 log_q=0.000 chi2=179.154
2026-10-01 02:59:13 (604560): called boinc_finish(0)

</stderr_txt>
]]>
ID: 194 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Alperen Yavuz
Project developer
25 years old

Send message
Joined: 27 Sep 26
Posts: 349
Credit: 5,210,003
RAC: 137,017
Message 199 - Posted: 1 Oct 2026, 23:16:14 UTC - in response to Message 194.  

This is a test app; the core code isn't fully configured yet. You are generating data to assist with my coding process—it is not actual, real-world data.
ID: 199 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
mmonnin

Send message
Joined: 28 Sep 26
Posts: 14
Credit: 20,205,102
RAC: 1,209,765
Message 200 - Posted: 1 Oct 2026, 23:53:11 UTC

Then accept any result returned. If you send out work, it needs to have a correct, accepted result.
Go hire your own data center to do this junk work for you.
Another reason the project shouldn't be listed.
ID: 200 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Alperen Yavuz
Project developer
25 years old

Send message
Joined: 27 Sep 26
Posts: 349
Credit: 5,210,003
RAC: 137,017
Message 201 - Posted: 2 Oct 2026, 0:14:08 UTC - in response to Message 200.  
Last modified: 2 Oct 2026, 0:20:47 UTC

All projects do this; you don't say anything to them. If you don't like it, leave. I am not an institution; I am just one person.

I finished the apps in two months, and this project hasn't even been running for a week—what do you expect? Do you think fixing bugs is easy?

You cannot modify the code without users running the application, generating data, and seeing the results.

And it has already been clearly stated and explained that KAMPYLOS is in the testing phase.

I would be happy if people could start doing something other than opposing everything, acting like bullies, and behaving in a hostile manner.

Have you ever seen a BOINC project of this scale before? I have 6 TB of data, and I want to process it properly—not in a haphazard way—and publish the results accordingly, rather than using some random, half-baked application. I haven't released the GPU application yet because there is a coding error in the FP32 implementation; I am manually debugging the C code with the help of a volunteer, as AI was unable to solve the issue.
ID: 201 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
boboviz
Ambassador

Send message
Joined: 29 Sep 26
Posts: 47
Credit: 1,634,482
RAC: 33,191
Message 206 - Posted: 2 Oct 2026, 8:38:10 UTC - in response to Message 200.  
Last modified: 2 Oct 2026, 8:40:00 UTC

In reply to mmonnin's message of 1 Oct 2026:
Then accept any result returned. If you send out work, it needs to have a correct, accepted result.
Go hire your own data center to do this junk work for you.
Another reason the project shouldn't be listed.


I'm not agree.
Only "big" projects (Rosetta@home, Einstein@home, LHC@home, etc) have the possibility to setup beta test site (Ralph@home, Albert@Home, LHC Dev, etc).
It's normal, in a new project, to have beta wus
ID: 206 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profiledskagcommunity
Bug hunter
Early supporter
46 years old
Avatar

Send message
Joined: 29 Sep 26
Posts: 64
Credit: 21,977,365
RAC: 136,334
Message 223 - Posted: 2 Oct 2026, 11:48:44 UTC - in response to Message 206.  
Last modified: 2 Oct 2026, 11:53:52 UTC

I must agree with bobovic, project just rolled out large scale. There should be a bit patience, no matter how somebody stands to the projects.

Indeed that seems the first subproject for "us" true sciencecrunchers. So it tries to scale in every direction, whats a good thing.
---------- 24/7 Crunching since 2011 -----------
DSKAG Austria: http://www.dskag.at

ID: 223 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profilerilian
Volunteer moderator
Code reviewer
Bug hunter
Early supporter
Avatar

Send message
Joined: 28 Sep 26
Posts: 261
Credit: 43,482,494
RAC: 474,864
Message 233 - Posted: 2 Oct 2026, 13:42:00 UTC - in response to Message 223.  

In reply to dskagcommunity's message of 2 Oct 2026:
There should be a bit patience

Yeah.. Some people complained and slammed the door on 1st day after credit issue :D
I crunch for Ukraine

ID: 233 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Aurum
Early supporter
Avatar

Send message
Joined: 1 Oct 26
Posts: 10
Credit: 7,637,822
RAC: 62,147
Message 236 - Posted: 2 Oct 2026, 14:28:49 UTC

I'll be glad to run your beta test WUs. Look forward to getting them.
ID: 236 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
mmonnin

Send message
Joined: 28 Sep 26
Posts: 14
Credit: 20,205,102
RAC: 1,209,765
Message 261 - Posted: 2 Oct 2026, 20:04:42 UTC

Projects DO NOT do this. This project is an outlier

The admin is running before leaning how to crawl let alone walk in every aspect of BOINC. Did they learn the ropes of running a BOINC project with 1 app? No, 4 were released in quick succession

NEVER EVFER have I seen a project refuse to validate work. Especially if they know the results will be bodus. Even if application does not provide the results you want, then validate no matter the result. It's not the users fault the app does not work. The only thing a project provides the users is credit and badges. If you're not going to provide that for the money and effort we put in then why should we do it?

You cannot modify the code without users running the application, generating data, and seeing the results.


I'm critical of admins that do not run a single task themselves offline or with BOINC and shove an app to prod w/o testing.[/quote]
ID: 261 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profilerilian
Volunteer moderator
Code reviewer
Bug hunter
Early supporter
Avatar

Send message
Joined: 28 Sep 26
Posts: 261
Credit: 43,482,494
RAC: 474,864
Message 262 - Posted: 2 Oct 2026, 20:13:56 UTC - in response to Message 261.  

In reply to mmonnin's message of 2 Oct 2026:
Projects DO NOT do this. This project is an outlier
NEVER EVFER have I seen a project refuse to validate work.


That is why this sub-project status is in development, i think

As for other 3 apps, the nature of approach is random checks, so it is unlikely can be validated properly
I crunch for Ukraine

ID: 262 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Alperen Yavuz
Project developer
25 years old

Send message
Joined: 27 Sep 26
Posts: 349
Credit: 5,210,003
RAC: 137,017
Message 263 - Posted: 2 Oct 2026, 21:57:22 UTC - in response to Message 261.  

You have the simple option of abandoning the project. The applications have already been tested in a QEMU environment; you won't get anywhere without generating data on actual physical hardware to identify and debug errors. And if you're such an expert, go ahead and create a BOINC project yourself let's see what you can do. BOINC's only function is distributing programs and collecting results; it has no other purpose. It is an extremely simple platform the real work lies in the applications themselves.

For people like you, it’s not about projects or anything like that you just love attacking people.

On the very first day I announced the project before I had even uploaded the applications to the system, when there was nothing but a draft you started attacking and badmouthing it. I am already examining everyone's issues one by one, resolving them, and explaining both the cause of the problem and the fix I applied. What exactly do you want to happen?
ID: 263 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Alperen Yavuz
Project developer
25 years old

Send message
Joined: 27 Sep 26
Posts: 349
Credit: 5,210,003
RAC: 137,017
Message 315 - Posted: 3 Oct 2026, 23:12:02 UTC

Something exciting might be turning up in the Kampylos search. Early into going through the archive, I found one event that's showing a strong statistical preference for a binary lens over a single star, and the numbers are pointing toward the planetary mass range. I'm checking the literature right now to see if this has already been reported somewhere, and re-verifying the result before I call it anything official.

If it holds up, this could be our first real find, and yeah, that probably means a naming contest before it goes anywhere near publication, because something like this deserves a real name, not a catalog number. More soon.

We're still just getting started here. Tens of thousands of events left in the queue, and every volunteer running the app gets us through it faster.

Honestly, I didn't expect to actually catch something like this, and I'm genuinely excited about it. I think I want to put more of my focus on this project than the others for a while.
ID: 315 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Alperen Yavuz
Project developer
25 years old

Send message
Joined: 27 Sep 26
Posts: 349
Credit: 5,210,003
RAC: 137,017
Message 320 - Posted: 4 Oct 2026, 0:38:07 UTC

It passed the verification tests, and I haven't found any published data on the internet. I suppose we'll need to find an astrophysicist or someone similar to verify the result.
ID: 320 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Alperen Yavuz
Project developer
25 years old

Send message
Joined: 27 Sep 26
Posts: 349
Credit: 5,210,003
RAC: 137,017
Message 360 - Posted: 4 Oct 2026, 18:24:59 UTC

It was confirmed that the result did not belong to a planet—that it was a parasite.
ID: 360 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Drago75
Bug hunter
Early supporter
51 years old

Send message
Joined: 28 Sep 26
Posts: 67
Credit: 45,851,691
RAC: 928,453
Message 404 - Posted: 5 Oct 2026, 23:02:09 UTC
Last modified: 5 Oct 2026, 23:03:22 UTC

I am getting mostly download errors for Kampylos, but strangely not all of them. RTX 3060m on Linux Mint 22.3:

<core_client_version>8.3.0</core_client_version>
<![CDATA[
<message>
WU download error: couldn't get input files:
<file_xfer_error>
<file_name>kampylos_phot_ev110_1791037503_7494</file_name>
<error_code>-224 (permanent HTTP error)</error_code>
<error_message>permanent HTTP error</error_message>
</file_xfer_error>
</message>
]]>
ID: 404 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
1 · 2 · 3 · Next

Message boards : Puzzles and science : KAMPYLOS