OptimallyAxel Ha - I noticed the GTA-6 parallel too. Honestly, they don't have the luxury of releasing a limited "V0" focusing on key new technology demonstration like I do. My V1 is probably 1 year delayed too - good. This is actually getting really intense and constricted by the 12/13 deadline because the floating licensing system needs to be perfect and support all the future industrial packages and flexible SKUs - I don't know if I will have RadarHalo by 12/13 and I absolutely don't care, because what's already here is spectacular, and my velocity will take off in the winter. This is so clearly the upcoming world's most optimized and performant, fastest platform for all geospatial visualization I need to absolutely hold the course, cut zero corners, and make sure that, at a minimum, the "world's greatest GOES-19 raster timeseries view in original, full resolution with zero downscaling on beautiful dynamic range ultra-resolution elevation global basemap" minimal deliverable you've seen in the clip is ready on 12/13 for many users. Keep in mind I am working in a 100x more complex and serious and high-difficulty environment. The same realtime server that serves the data with millisecond latency also does the licensing. It either runs 24/7/365 with no issues or else the entire product is offline for everyone. So I have recently pivoted towards testing and finalizing the floating licensing system (immediate VM supporting cross-device zero-activation usage) and getting the historical data backfill + realtime update data model absolutely perfect.
Actually yesterday I had a huge epiphany which results in possibly a 2 week delay to my existing work. The previous model for the Websockets-based locally cached metadata backfill was a "sliding window" model where, for each abstract "realtime data feed", the client would request server for frame history per-feed in 30 minute chunks going backwards until the full window is loaded. The realtime update path was separate and not implemented yet. When I started the realtime update path I realized the critical issue my model didn't cleanly handle when NOAA outages happen and retroactive data refills occur.
The whole point of tactical is you are paying $20/month for the most absolute mission-critical, optimized, zero gaps, full resolution data service that delivers the most valuable subset of operational meteorological data for realtime storm tracking at the fullest resolutions and lowest latencies.
This means that the service has to transcend the old, "sorry, NOAA was out". It needs to handle NOAA outages as a controlled reality and implement absolutely perfect, industrial strength backfill logic with the same realtime infrastructure as the regular new frames. In other words, every second where a new NetCDF or GRIB2 source file is uploaded from government but not yet visible in the application with the full post-processed smooth/fast loading pyramid tiles we generate from raw bloated rasters, is a moment of failure.
So I need to switch to an unbounded length, temporal LOD bitmask-based model where all reatime datafeeds are quantized into one-second bins, 86400 per day, implemented over Websockets as ultra-efficient ZSTD compressed bitsets, and the same single wire message format for regular loop data backfill also serves post-outage refill and regular realtime updates.
This will make it possible for a forensic product later that could, in theory, loop over weeks, months, or years of absolute full resolution data and rapidly adjustable timescales - only downloading the frames per-view that correspond to visual temporal resolution at a separate license-mediated "data FPS", which can be 40 FPS in the base license (your eye struggles to tell the difference faster than this and it would waste more data). The actual map motion and overall rendering remains unbounded with 120fps monitor recommended.
V0 is for the "CONUS Standard Realtime" package where I am optimizing the costs to shove 10x+ data as Professional into a reduced price $20/month subscription. The Band 2 CONUS satellite imagery is the centerpiece of this as literally all the other data is an order of magnitude smaller size than this. The raws are nearly 100mb per 5 minute frame. The next-gen pyramid engine will download full-res loops using view-constrained tiles at rates expected to be unsurpassed by any other existing web-based or desktop software currently in existence, for a given fixed internet connection speed.
V0 on 12/13 is really about finally conquering the biggest gap in my software abilities for realtime weather analysis in the past decade - the Band 2 CONUS full-resolution 0.5km satellite imagery. WSV3 Professional downscales this to 1/4th dimension, 1/16th data size - because it has no multires tile pyramid engine.
RadarHalo Polyradial Mosaic will come directly after - likely December but possibly January 2026. This is where we have multiple Level 2 radar sites up concurrently and automatically paged in with camera view, blending into the MRMS mosaic in grayscale offscreen pass before color palette application at far ranges. This also requires full 24/7 realtime ingest and processing of the entire NWS Level 2 feed for the whole country - which I am also packing into the $20/month. First-gen just downloads the massively inefficient raw AR2V files straight from public sources. They are wasting 50x the bandwidth in the common case of just viewing 0.5-deg base super-res reflectivity.
So the new base license for WSV3 Tactical Mesoanalyst will be way higher value than WSV3 Professional in terms of data - but it will be narrower and woefully incomplete on 12/13 compared to what I intend to add to it in 2026.
So, it is absolutely a GTA-6 situation. Actually, it's my "Halo 3" moment. The whole WSV3 product line and going back to WeatherStudio (my Halo 1) was inspired by how compelled I was by playing Halo 3 as a 13 year old and forming an interest in realtime GPU graphics programming. Halo 3 was the perfect third installment where Bungie did not rush (unlike Halo 2) and had time to enact perfectionism. The parallel is absolutely fitting, because WSV3 Professional 2015 (my Halo 2), was incredibly rushed and released prematurely.
V0 will be in some other ways much more than what I envisioned in September 2023 when starting this. Mostly relating to architecture/engineering and how I have redefined modern Win32 software development with this project. You will download a sub-200kb binary that is written in bare metal Pascal with zero language RTL, zero language standard runtime library, zero dynamic memory allocation, and using pure static .bss section preallocated buffers. This is unheard of even for game engines and C++ software. It makes C++ native apps using the C++ standard library look like Javascript. This is the kind of development practices for NASA space rovers and medical/automotive/aerospace embedded projects, brought to Win32 graphics software. Zero-installer, instant launching, downloads-on-demand what it needs - current virgin launch times are sub 2.0s on regular internet and subsequent post-initial launch times sub-300ms, depending mostly on iGPU. Tactical means "zero waiting" and massive immediacy. And the whole D3D12 rendering engine rewrite was incomprehensible to me if you'd told me 2 years ago we'd do it before December 2025.
So all of that data-oriented-design-extremist insane level re-engineering cost probably 1.25 years of development from 9/2023 to now - for good cause. I basically threw out and did not continue using 95%+ of my code written from 9/2023 through 5/2024. Even most 6/2024 - early 2025 code is gone due to phases of perfectionism, culminating in April 2025 achievement of the complete excision of all dynamic memory and recompiling System.pas with no allocator. Now we have an application that will simply never crash. Once we get through the weeds of expectable graphics driver quirks and edge cases I'm sure we will hit in the first few weeks, it will hit a level of eternal stability very quickly. The proof is first-gen is still relatively stable even with its convoluted mess of dynamic memory allocation and high-level object-oriented code which is anathema to me nowadays.
WSV3 Tactical Mesoanalyst is the first publicly accessible native desktop GPU viewer for geospatial and weather data that is implemented as a true embedded realtime system with constant server connection and zero dynamic memory. V0 will be by far enough to communicate this unmistakably. Of course I do regret not having more high-level actual content on 12/13 as originally planned, but I would not have persisted in the high-level D3D11 Delphi-based OOP-programmed app which I had at the time of the early June 2024 crossroads (and decision to start from mostly scratch again on bare metal design) if given the choice. We even had the 3D terrain engine then I still plan to reimplement in 2026. If I hadn't done the unplanned restart in June 2024, we would have a slower and less reliable but graphically impressive version of what we will have in just a handful of months in the most robust, mission-critical, stable, performant, low-resource-usage, and truly tactical-grade way.