nusadua.dev

Independent by design

Hi.

Hello / From BaliResearch follows
SCROLL TO EXPLORE01 / 10

Drag to rotate the study. With the study focused, use arrow keys to rotate and Home to reset. Double-click also resets the view. In the 164-point configuration, hover a point to read its coordinates, click to hold it, and press Escape to release. The objects rotate automatically on two axes; automatic motion pauses during dragging or point inspection and is disabled with reduced motion. The coordinates are also available as a text download.

nusadua.dev✳ Independent by design

STUDY 01 / Independent research

Then we
research.

A little curiosity.
A better question.
A different way through.

Follow the questions ↗
STUDY / 001INTERFERENCE
001 / A study in interferenceGo on. Play a little.
↓ Browse publications
000 / Serious work. Curious minds.

In the open

Publications.

Papers, open data, and contributions to shared mathematical references.

01 / Papers & data

Zenodo

7 preprints + 1 dataset

Full texts, versioned records, and open data. See the Publications tab.

03 / Configurations

Achim’s archive

No-three-in-line archive

9 new half-turn-symmetric solutions (rot2) for n = 33, 36, 37, 39, credited to Aleksei Kudriashov on .

Zenodo bibliography 08View publicationsClose bibliography
  1. PreprintVersion 3.4

    Nineteen certified configurations, eighteen independent bounds for the no-four-coplanar problem in the cube, and what they do not establish (OEIS A280537)

    DOI 10.5281/zenodo.22272371
  2. PreprintVersion 1.9

    The error of the Guy-Kelly heuristic, measured against exact counts

    DOI 10.5281/zenodo.22063379
  3. PreprintVersion 1.1

    The direction spectrum of no-three-in-line solutions: a line model derives its shape, not its scale

    DOI 10.5281/zenodo.22278951
  4. DatasetVersion 1.1

    Witness configurations and lower bounds for the no-three-in-line problem in the n x n x n grid (OEIS A399138): a(8) >= 94, a(9) >= 116, a(10) >= 138, a(11) >= 164

    DOI 10.5281/zenodo.22271462
  5. PreprintVersion 1.5

    Certified exact values for the no-three-in-line problem in the three-dimensional grid

    DOI 10.5281/zenodo.22273425
  6. PreprintVersion 1.3

    Balanced orbit defects in half-turn-symmetric no-three-in-line configurations, with new 2n-point configurations for n = 36, 37, 39

    DOI 10.5281/zenodo.22064192
  7. PreprintVersion 1.18

    Extremal no-three-in-line subsets of a modular hyperbola in the Hall-Jackson-Sudbery-Wild window

    DOI 10.5281/zenodo.22275500
  8. Preprint

    Pre-registered stakes against a tuned baseline: a one-day ledger of grokking interventions that mostly failed

    DOI 10.5281/zenodo.22196294

Development support

Tools to keep building.

In-kind support for development.

OpenAI

Daily API token credits for open-source development.

Google

Virtual machine and Gemini credits.

Small proofs.
Made to travel.

Inside the atelier: experiments, constructions and working notes. We make our assumptions explicit, keep the counterexamples, and leave something another person can build on.

For the curious.15 papers, experiments & working notes

15 entries

Papers, constructions, experiments and the methods behind them.

01 / MATHEMATICSPreprintModular hyperbolasExact structure inside the Hall–Jackson–Sudbery–Wild window.02 / MATHEMATICSPreprintNo three on a line. In a cube.Exact small cases and independently certified lower bounds.03 / MATHEMATICSPreprintNo four in a plane.Certified configurations and bounds for the cubic grid.04 / MATHEMATICSPreprintWhen symmetry leaves a gap.Balanced orbit defects in half-turn-symmetric configurations.05 / MATHEMATICSPreprintHow wrong is a good heuristic?The Guy–Kelly estimate, measured against exact counts.06 / MATHEMATICSPreprintShape is not scale.A line model for the direction spectrum of geometric solutions.07 / MATHEMATICSVerified constructionEvery point, accounted for.164 exact coordinates in an 11 × 11 × 11 grid.08 / MACHINE LEARNINGExperimental studyThe structure of what we learn.Data selection, spectral predictors and conditional generalisation.09 / MACHINE LEARNINGExperimental studyA rule learned. Then erased.Auxiliary memory, dying gradients and the loss of generalisation.10 / MACHINE LEARNINGExperimental studyFirst, fix the ruler.Grokking accelerators against a tuned, sustained baseline.11 / MACHINE LEARNINGExperimental studyGive the experiment time.Activation, final LayerNorm and the training-budget effect.12 / RESEARCH METHODSResearch methodSmall proofs. Made to travel.From a computational need to a lemma someone else can use.13 / RESEARCH METHODSResearch methodA different way through.Thinking lenses with invariants, predictions and recorded failures.14 / MATHEMATICSOpen investigationCan every vertex share one journey?Hamiltonian cycles, graph transformations and Barnette’s conjecture.15 / MATHEMATICSOpen investigationCan a hard configuration split in two?A doubling-lemma approach to the Erdős–Szekeres problem.
Point data: Aleksei Kudriashov · A399138 witnesses · CC BY 4.0.
The ribbon and square study are visual interpretations. Point constructions use source coordinates; scene notes distinguish reported data, model predictions and illustrations. Neon decorations carry no research data.

IN PLAIN WORDS

Okay, what do
you actually do?

We’re an independent studio in Bali. We do research and build software for businesses.

First, we meet on Google Meet. You walk us through your business and its processes: what’s hard, what slows you down, what drives you mad. Together, we find a way forward with software and science.

Research

Independent questions in mathematics and machine learning.

Software

Internal platforms for finances, approvals and working together.

HOW WE SHARE

Good work.
Shared good.

Half of what you pay for our services goes to us. The other half goes to a local children’s home in Jimbaran, Bali.

50%

Our work

50%

A children’s home
in Jimbaran

A Google Meet call / A beginning

Let’s talk.
Human to human.

Your business, the awkward processes, the idea you can’t leave alone. Let’s find a useful place to start.

  1. Tell us a little.
  2. Choose a day that works.
  3. We confirm the time and send your Google Meet link.
studio@nusadua.dev ↗
01 / A little about you02 / A time to talk
A little about you
Where should we reach you?

A username is enough. No phone number needed.

We use your details to arrange the call. Requests are stored privately and sent to our studio mailbox. Your draft stays in this browser tab.