iiinova — where your digital self keeps evolving

James Kane @jameskane.blog · 01/10/2026

Same desktop using the Aqua theme. Iconography needs work...

James Kane @jameskane.blog · 01/10/2026

My FreeBSD fork to add the Radxa Dragon Q8B is coming along nicely. The NeXT/SGI/AmigaMUI inspired theme for a new desktop environment written in Swift 6 is now running on it. Most of the hardware for a desktop experience is fully operating.

121

James Kane @jameskane.blog · 29/09/2026

Opus 5.5 now has finished accelerated GPU paths for the Dragon Q8B on #FreeBSD. Roughly the same FPS as Radxa's Ubuntu image. Now to get it to clean up the mess.

021

James Kane @jameskane.blog · 28/09/2026

I have pointed Claude Opus 5.5 at the #FreeBSD source and the #Radxa Dragon Q8B. It has a near feature complete driver for the two ethernet ports. Now it's iterating on more driver support via SSH. Feature branch for any interested: github.com/JamesKane/fr...

github.com

GitHub - JamesKane/freebsd-src at radxa-dragon-q8b

The FreeBSD src tree publish-only repository. Experimenting with 'simple' pull requests.... - JamesKane/freebsd-src

000

James Kane @jameskane.blog · 26/09/2026

The Geekbench 6 score for the #Radxa Dragon Q8B. It's a little faster than the Orange PI 6 Plus board using four fewer cores to do it. 1651 Single-Core, 6833 Multi-Core

000

James Kane @jameskane.blog · 25/09/2026

New SBC received: Radxa Dragon Q8B w/ 16 GB of RAM Now to find an NVMe to boot it from and get some benchmarks.

000

James Kane @jameskane.blog · 15/08/2026

The latest installers can be found at decoding-us.org/download. 3-Clause BSD source code: github.com/JamesKane/de...

decoding-us.org

Download the Navigator — Decoding Us

000

James Kane @jameskane.blog · 15/08/2026

Added a read remapping harness in Decoding-Us Navigator's latest Alpha build. Remap your hg37 or hg38 direct-2-consumer BAM to chm13v2.0 freely and locally on a computer with at least 16GB of RAM and a few hundred GB of disk space. #genetic #genealogy #Big-Y #Y-Elite

100

James Kane @jameskane.blog · 09/08/2026

MDKA informational boxes for samples under the terminal branches for people who opt-in from Navigator is the current work in progress.

000

James Kane @jameskane.blog · 09/08/2026

Starting to work Block Tree aka Icicle tree rendering into decoding-us.org. Limiting these to the genealogical era as the amount of data blows up the further back you allow. ytree.net and FTDNA's render speeds are unacceptable. decoding-us.org/ytree/node/R...

decoding-us.org

Ancestral origins · R-A9005 — Decoding Us

100

James Kane @jameskane.blog · 06/08/2026

The latest feature branch for the Navigator project system is coming together: Icicle blocks derived from Alex Williamson’s ytree. Identification of shared SNPs, which may represent new branches. It should land in a new Alpha release this weekend. #YDNA #Genealogy #DataVisualization

Screenshot of the DecodingUs Navigator app showing an icicle plot of the CTS4466 Y‑DNA tree, with a highlighted candidate branch (Candidate 1) spanning roughly 1900–1800 BC on a timeline from ~2015 BC to the present.

010

James Kane @jameskane.blog · 29/07/2026

Added a new report at decoding-us.org, which displays the Y-DNA STR ranges, published mutation rates, and motif values where available. There's a job setup to calculate new mutation rates when the sample size gets larger. decoding-us.org/str-markers #Genetic #Genealogy

decoding-us.org

Y-STR Markers — Decoding Us

000

James Kane @jameskane.blog · 28/07/2026

Working on an Archaic DNA module based on Altai, Chagyrskaya8, Denisova3 and Vindija33. Mostly working and Denisovan DNA being rejected for a European sample. Need to run on some Oceanic samples to validate the rest and clean-up the final reports.

Screenshot showing archaic ancestry results: 12,126 Neanderthal/Denisovan allele copies detected across 299,932 marker sites (100% coverage). Breakdown: 9,460 Neanderthal-diagnostic, 2,412 shared-archaic. Result is higher than 7% of European reference samples. Note: Denisovan-specific markers not reported outside Oceania; counts aren’t directly comparable to other companies’ data.

000

James Kane @jameskane.blog · 25/07/2026

Latest Alpha Build for Decoding-Us Navigator. Rewrote the Chromosome Painter to better isolate the two parents. Added more populations from public datasets. Fixed all the reported issues in older releases. #OpenSource #Genetic #Genealogy

Screenshot of the Decoding-Us Navigator interface. The left sidebar lists subject IDs. The main panel displays a "Your DNA sides" section featuring a chromosome painting chart. Horizontal bars representing chromosomes 1 through 22 are filled with various shades of blue blocks. A legend below the chart identifies ancestral origins such as British, Tuscan, Finnish, and Iberian. At the bottom is an AI input box labeled "Ask about your results."

241

James Kane @jameskane.blog · 12/07/2026

Built in Rust with no complex dependencies, Navigator supports "progressive consensus" to cross-check multiple DNA kits & syncs anonymized results via AT Protocol for tree building. Join the Alpha: jameskane.blog/genomics/202... #RustLang #ATProto #OpenSource

jameskane.blog

Navigator Alpha: Run Your Own Genomics Lab on Your Own Machine

Navigator is the desktop app that turns your own computer into the analysis lab. Point it at a BAM, CRAM, or a raw data file from a consumer test, and it does the coverage, the Y and mitochondrial hap...

040

James Kane @jameskane.blog · 12/07/2026

🧬 Introducing Navigator Alpha: a local desktop genomics app. Process BAM/CRAM, VCF, or chip tests (23andMe/Ancestry) offline with zero cloud uploads or privacy risks. Run Y/mt haplogroups, coverage, & ancestry on your own hardware. #Genomics #Bioinformatics #GeneticGenealogy 👇

140

James Kane @jameskane.blog · 04/07/2026

The Formed and TMRCA ages are based on McDonald's (2021), Improved Models of Coalescence Ages of Y-DNA Haplogroups. doi.org/10.3390/gene...

doi.org

100

James Kane @jameskane.blog · 04/07/2026

Unlike prior examples, I did take pains to use FTDNA's labels where they are not obviously in need of improvement or don't exist. The side panel shows the defining SNP(s) placements on b37, b38 and hs1 (chm13v2). Where available it also reconstructs the STR signature transitions from the parent.

A screenshot of the Decoding Us Y-DNA tree with a detailed information panel open on the right. The tree shows various branching haplogroups, and the side panel displays specific genetic data for the selected node "R-S1115," including its TMRCA (Time to Most Recent Common Ancestor), specific SNP mutations, and Y-STR ancestral motifs.

100

James Kane @jameskane.blog · 04/07/2026

The Decoding-Us AppView is now running the Rust-version and hosting my latest chm13v2-based haplogroup trees. Y-DNA is sourced from IGSR samples and YDNA Warehouse samples. mtDNA is IGSR samples.

A phylogenetic tree diagram of Y-DNA lineages from the Decoding Us project. The chart shows a branching ancestral path starting from a root "Y" node, splitting into major branches like A0-T and A1, which further divide into numerous smaller sub-clades (yellow boxes) with associated variant counts and TMRCA (Time to Most Recent Common Ancestor) dates.

A phylogenetic tree diagram of mitochondrial DNA (mtDNA) from the Decoding Us project. The lineage begins with the L0a node, branching through L0a1 into further sub-groups like L0a1-A2000 and L0a1b. A list of specific biosamples is displayed on the right side of the tree.

110

James Kane @jameskane.blog · 04/07/2026

I have the 32GB model. When purchased in January, this was a no-brainer upgrade. Like most SBCs the RAM is soldered on. It will be interesting to watch if this changes. Putting SODIMMs on the boards is the only way they will "beat" RAMogedden.

010

James Kane @jameskane.blog · 02/07/2026

With the used market having Ivy Bridge or newer systems like the Trashcan Mac going for $200 or less, it's hard to look at the SBCs here and their current pricing with RAM shortages as viable. Even if the Intels are using 10x the energy...

000

James Kane @jameskane.blog · 02/07/2026

OrangePi 6 Plus running stock Ubuntu instead of the vendor image. Minimap2 alignment performance. Total Reads. Total Time Reads/Sec 615605482 3h 50m 0s 14ms 160592.571514783 Roughly the same performance as a 12-core Mac Pro 2013.

320

James Kane @jameskane.blog · 18/06/2026

I am highly amused by Claude Code's thinking streams when dealing with FTDNA's Big Y VCF data. Everything I've muttered over the last decade when a BAM wasn't available, so I could just ignore their oddities. #genetic #genealogy

000

James Kane @jameskane.blog · 17/06/2026

2. Strength: Structural variants. The graph recovers ~80% of HiFi SVs. Delly gets 26% on chr20 and collapses to 14% on chr16. The harder the region, the bigger the graph’s edge. Full analysis: jameskane.blog/genomics/202... (3/3)

jameskane.github.io

What a Pangenome Misses, and What It Nails

My last post argued that a genotype-only pangenome callset silently misses variation and quietly carries structural variation. This time I put numbers on both claims, across two whole chromosomes, wit...

000

James Kane @jameskane.blog · 17/06/2026

1. Weakness: Small variants. The graph is blind to private SNPs because they aren't in the panel (99% representation gap). It also over-calls noise in repetitive regions (only 43% confirm). Standard linear callers win here. (2/3)

100

James Kane @jameskane.blog · 17/06/2026

What does a genotype-only pangenome callset miss, and what does it nail? I put numbers on both claims across chr20 & chr16, using 3 callers (vg, GATK, Delly) and a low-coverage PacBio HiFi long-read reality check. The trade-off is not what I expected. 🧵 (1/3)

100

James Kane @jameskane.blog · 17/06/2026

First generation of the de novo mtDNA tree is live in the Decoding Us dev region. 🧬 Built using 1K Genomes raw data realigned to CHM13v2 (T2T), processed via GATK, and structured with IQ-TREE. Native, clean rendering of the U5a1 branch subclades. Moving beyond just Y-DNA.

A web browser screenshot of the Decoding Us platform developer region displaying an interactive mitochondrial DNA (mtDNA) phylogenetic tree layout. The interface shows a horizontal tree structure focusing on haplogroup U5a1 and its nested subclades, including U5a1b1, U5a1b1a, and U5a1b1h. Internal nodes are represented by light yellow rectangular boxes labeled with haplogroups or specific variants (such as '★ chrM:15743C>T'). Terminal nodes branch out to individual sample identifiers from the 1000 Genomes project, marked by green dots (e.g., NA20515, HG00285). The top of the page includes a breadcrumb navigation path tracing the phylogenetic lineage back to root nodes, alongside controls for tree depth and orientation.

000

James Kane @jameskane.blog · 16/06/2026

Time to see how this #OrangePi 6 Plus handles minimap2. Initial impression is it's a loud boy with that stock fan.

000

James Kane @jameskane.blog · 15/06/2026

With the end of the Intel Mac era, I picked up a Late 2013 Mac Pro with dual AMD FirePro D700s. It's definitely a 13 year old machine, but still useful and an interesting desk ornament for the retro desk.

A screenshot of the "About This Mac" window. It shows a Mac Pro (Late 2013) running macOS Monterey version 12.7.6. Specifications listed include: Processor 2.7 GHz 12-Core Intel Xeon E5, Memory 64 GB 1866 MHz DDR3, and Graphics AMD FirePro D700 6 GB. The serial number is redacted with a black bar.

010

James Kane @jameskane.blog · 15/06/2026

Multiple data sources tab

000

James Kane @jameskane.blog · 15/06/2026

mtDNA profile using FTDNA's tree at the moment.

A dark-mode application screenshot of "DUNavigator" showing the "mtDNA" analysis tab for subject KANE-001.
mtDNA haplogroup panel: Displays the assigned haplogroup as U5a1b1g, noting "53/55 mutations, score 0.629". It maps out the full phylogenetic lineage tracing back to RSRS.
mtDNA sequences panel: Indicates "No mtDNA sequences yet" with a button to "Import mtDNA FASTA".
mtDNA de-novo SNP calls (chrM) panel: Displays a list of "28 SNP call(s)" with columns for Position, Change, Depth, and AF (Allele Frequency). Visible variants include transitions like 174 A>G (depth 52, AF 1.00), 2620 T>C (depth 83, AF 1.00), and 6451 C>T (depth 109, AF 0.99). A disabled "Publish variants to PDS" button is visible at the bottom of the list.

100

James Kane @jameskane.blog · 15/06/2026

Y-DNA profile using the #ISOGG tree.

100

James Kane @jameskane.blog · 15/06/2026

Overview screen with sequencing display.

A dark-mode desktop interface of "DUNavigator" showing the "Overview" tab for a male subject, KANE-001.
The top section, "Haplogroup consensus", displays:
Y-DNA: R1b1a1b1a1a2c1a3a2a1a2d1b1 (6 sources, conf 0.519) marked as compatible.
mtDNA: U5a1b1g (6 sources, conf 0.630) marked as compatible.
The middle section, "Coverage", shows global metrics such as a Genome Territory of 3117292070, Mean Coverage of 3.63, Median Coverage of 3, and 1.1% of bases at or above 10x depth.
A "Per-contig coverage" table breaks down stats for chromosomes 1 through 6 (chr1–chr6), showing lengths, read counts, and mean depths ranging from 3.71 to 4.10. A "Depth histogram — whole genome" with red vertical bars is visible at the very bottom.

110

James Kane @jameskane.blog · 15/06/2026

The new Rust-base DecodingUs Navigator is coming along. Haplogroup detection and ancestry analysis working well. Still need to create the chm13v2 versions of a few panels. IBD via encrypted P2P between Navigators in progress. Then tidy up the consensus between multiple tests. #genetic #genealogy

A dark-themed desktop software interface for "DUNavigator: Decoding-Us Navigator".
On the left sidebar, a subject named "KANE-001" is selected, displaying a Y-DNA haplogroup beginning with R1b and an mtDNA haplogroup of U5a1b1g.
The main panel displays the "Ancestry" tab for this subject. A section titled "Donor ancestry (best source)" shows a blue circular graphic and states "European 100.0%" with "19023/20000 SNPs - confidence 96%". Below this, a sub-breakdown lists specific European regions: NW European (Utah) 36.8%, British 23.6%, Tuscan (Italy) 14.2%, Iberian (Spain) 13.1%, and Finnish 12.0%.
The bottom features a PCA (Principal Component Analysis) scatter plot comparing the sample against reference populations, and an error status bar at the very bottom of the window indicating a failed file fetch from a UCSC database URL.

100

James Kane @jameskane.blog · 09/06/2026

Two fixes: surject back to a linear reference + pileup, or vg augment the graph with the reads. Both recovered R1b-FGC29071. Lesson: a genotype-only graph VCF is a genotyper, not a complete variant catalog. jameskane.blog/genomics/202...

jameskane.blog

Finding a Y Haplogroup in a Pangenome (and Why It Almost Didn’t Work)

I tried to read a Y-DNA haplogroup straight out of a personal pangenome alignment. It got me most of the way, then quietly lied to me. Here is what went wrong, why, and the three different ways I even...

010

James Kane @jameskane.blog · 09/06/2026

The catch: a genotype-only pangenome callset can only see variation carried by the assemblies that built the graph. My terminal lineage (R1b-L21 down to FGC29071) wasn't in the panel, so it was invisible. The repetitive Y made it worse with paralog artifacts.

410

James Kane @jameskane.blog · 09/06/2026

I tried to read my Y-DNA haplogroup straight out of a personal pangenome alignment. It confidently placed me in R1b, then steered me to the WRONG sub-branch. New post on what went wrong, and how I got the right answer. 🧬🌳 #bioinformatics #genomics

120

James Kane @jameskane.blog · 27/05/2026

My SpaceMit K3 dev system arrived after some adventures with DHL. Performance is about 3x faster than the K1 on GeekBench 6. Firefly's AIBOX-K3 variant is nicely trimmed. Will run it as is for a bit before reimagine with a different OS.

SpaceMit AIBox-K3 Geekbench 6 Scores:
327 Single-Core, 1594 Multi-Core

000

James Kane @jameskane.blog · 22/05/2026

Holy hell. My SpaceMit K3 board shipment was assessed nearly $150 in import fees to clear customs. So tired of "winning." SBC's as a hobby are effectively dead in the current world.

000

James Kane @jameskane.blog · 17/04/2026

Breaking: Trump lies, again.

030

James Kane @jameskane.blog · 17/02/2026

Well... the models are trained on the code from mediocre white men.

010

James Kane @jameskane.blog · 16/02/2026

No shit. Phones are a mature market now. There's nothing to add but incremental specs. This is not a bad thing.

000

James Kane @jameskane.blog · 14/02/2026

Working on a rewrite of bwa-mem2 in #Swift 6 with a Metal acceleration path on Apple Silicon. Using the techniques from "Accelerating BWA-MEM Read Mapping on GPUs". hps://dl.acm.org/doi/10.1145/3577193.3593703 Early benches show 4x faster than bwa-mem2 with NEON patching with 1/3 RAM usage!

000

James Kane @jameskane.blog · 10/02/2026

After no success in finding a complete Swift binding for htslib for another project I've been planning, I had Claude whip one up. The code is reasonable so it's on Github: github.com/JamesKane/sw... Expect changes as I find and fix the clanker's bugs and omissions.

github.com

GitHub - JamesKane/swift-htslib: Swift bindings for HTSLIB

Swift bindings for HTSLIB. Contribute to JamesKane/swift-htslib development by creating an account on GitHub.

000

James Kane @jameskane.blog · 06/02/2026

One of these years I will not take a layoff from running for 2 months after race season, but this was not that year. 45 minutes of absolute destruction on the treadmill in the books.

000

James Kane @jameskane.blog · 05/02/2026

Added a forked version of bwa-mem2 to my GitHub page, which compiles on ARM with NEON. Mostly leveraging sse2neon, but also some optimizations on top. github.com/JamesKane/bw... Timings done with chm13v2.0 as the reference for 2,000,000 150 base pair reads on Apple Silicon.

000

James Kane @jameskane.blog · 21/01/2026

Geekbench 6 suggests it's on par with my 2018-era gaming rig for single and multicore. My M4 mini smacks it around, but the CIX SoC actually has sample drivers I can reference for the OS I'm writing.

000