Decent future location for Rowing data analysis

[JaapvanEkris]

Hi all,

RowsAndAll.com will close its doors end of 2026. Its intended path is to integrate into Intervals.icu. However, I see some issues there:

• Intervals.icu really messes with lap and split data (see https://forum.intervals.icu/t/intervals-icu-messes-up-fit-lapdata/127532/11 and https://forum.intervals.icu/t/question-on-fit-file-parsing-misaligned-splits-summary-first-vs-summary-last-handling/123523/21), making checking adherence to new SPM's nearly impossible.
• Rowing specific things, like the force curve, don't seem to be on the priority list of intervals.icu, at all. I've been on the platform quite a while, and it feels like cycling is first, rest comes later.
• Options to plot things together (like strokerate and distance per stroke) isn't to be found anywhere.
• The GUI is a complete mess of options and buttons and things. Certainly not novice friendy, and even as an experienced user, not friendly

Ideally it would be hosted on the ORM Raspberry Pi itself, making it accessible via Web. Alternative might be on desktop or as a webservice.

I'm seriously looking for alternatives that will provide a good user experience. For example FIT dashboard, it looks nice, but still is in early development (the guy has a track record in building this stuff (see Garmin Grafana, so I am confident this will deliver if it gains traction.

Any other alternatives we might integrate to to perfect the experience?

[Abasz]

The GUI is a complete mess of options and buttons and things. Certainly not novice friendly, and even as an experienced user, not friendly

I do not have particular preference here, but this statement is really from my heart :D I feel the same. In fact its so overwhelming that I cant find the motivation to start learning it. I showed the owner of Old Danube he said it looks great and he said we should integrate vai the api upload, so I will implement that (as it is better than the pure browser storage that is non-transferable) but I wonder when a standard coach would see this would not get lost.

I did check FIT dashboard its visually pleasing though one needs time to get used to the different charts. It reminds me when a company was proposing our firm's finance team with dashboard so we can see everything at a glance and essentially it turned out its too much for management :D and they did not even take the time customise it as there were too many charts. I have the feeling of grasps all, looses all a little bit.

Furthermore, backing up session to an online platform is safer in terms of persistence.

[JaapvanEkris]

I found a potential option: Tredict (see https://www.tredict.com/app/). They have a decent GUI, are hosted in Germany, and they seem to respect lps etc. Any ideas before I reach out to them?

[JaapvanEkris]

I completely had it with Intervals.icu. They don't move at all, and their GUI is a convoluted mess of options. They keep messing with data. And key fields are still missing (force curve...). Not a good approach.

In the meantime I've been working on an MCP addition to ORM. This added the benefit that:

• reporting and analysis can span the entire period of maintained records. I've added quite some functions. So basic reporting needs can be solved by running a local MCP on the Pi, and connecting to it (doable, even for novices). Added benefit is integration into other datastreams (in my case environmetal monitoring and Garmin data).
• We can expand the dataset, including DFA-Alpha1 HR series, etc., without worrying how to get it into a specific platform.
• the file-writers will move to a database instead of in-memory recording. The MCP has an underlying database, in essence the superset of the FIT-data and the RowsAndAll data. This makes data persisted even during an OS crash
• the file-writers (still a ToDo) will move to the database as basis. This allows the creation of any past session trivial, even if that session has crashed, or whatever.
• Claude has implemented some scripts for migration purposses. These scripts will be testen with over a 1000 sessions. These also provide a backdoor for feeding in Garmin or other data, allowing the database to become a single source of truth.
• I am anticipating the MCP to be able to deliver a session's data in fit-file, csv etc..
• I am considering a report, similar to ESPRowingMonitor, to be directly reachable. Many data is just presentation, s the backend would do the heavy lifting.

[Abasz]

Are you going to use node database managers or issue sql commands directly? Who will run/maintain the database server/service? ORM or separate server?

I dont fully understand this MCP server approach either. That requires AI that interacts with the server so I suppose that will not be free right?

On intervals.icu I full agree on the part of it being too many options and difficult to navigate...

[JaapvanEkris]

Are you going to use node database managers or issue sql commands directly? Who will run/maintain the database server/service? ORM or separate server?

In essence, it will become a recorder, and everything is controlled from node.js. It will be controlled from node.js as sqlite has a pretty decent interface. The idea is that instead of 3 possible recorders (exluding the raw recder here) all recording their own state in memory, the database will just centrally record everything, and allow recorders to use that to produce their files. I designed it such that will prune data after a given period (configurable), so the footprint and memory usage are limited.

I really didn't like the memory usage of the recorders, as the rowingdata recorder and fitrecorder both create huge files. And if the app crashed, the data is often gone. In the default configuration it will retain a week of data, enough to look back a bit, and retrieve a failed upload, but light enough to work on a Pi Zero.

Currently I'm testing with full history of over a 1000 sessions, including the underlying data Garmin captured (raw heartrate and breathingrate). That is curently poblematic as the database is huge and the migation scripts slowly stop performing. It might be the migation-scripts, which lean heavily on the database to integrate Garmin data into the existing RowingData. So some analysis work to be done there.

I dont fully understand this MCP server approach either. That requires AI that interacts with the server so I suppose that will not be free right?

You can run a MCP locally on your own network. All it takes is installing node.js on the client machine to act as a proxy. So you can connect to the ORM's MCP and get direct access to all its data, and even set up the training plan. So the user has to set up a Connector (in Claude terms).

Installing Node.js for use as a proxy is a bit overkill but also extremely doable. So it will remain free and secure. Only downside is that mobile/web clients can't access it.

I understood that cloudflare has a service (called cloudflared) that can run on a RPi and essentially forwards the service. But finding out how to secure this well, and setting it up, is something for a later moment.

The MCP allows storing the knowledge in the database, so data analysis (like usual tresholds etc.) can be stored on the MCP as context, instead of the memory of the AI tool. It actually allows for decent reporting of a session, as well as good trend analysis and data comparison.