Tmmot Results

tmmot-results — the data + analysis layer

You give it one thing: the team you were following. It fetches the rest — every played game, the group standings, and a full-tournament analysis — straight from the official results feed. No invented data; only games that were actually played.

This is Skill 1 of 2. It produces the data.json data layer that tmmot-album (Skill 2) turns into a site + PDF memory book.

Inputs

• Team name (exactly as it appears on the feed, e.g. KA-2).
• Results URL (the official feed, e.g. https://urslit.tmmotid.is). The day

pages are scraped (?day=A/B/C), only rows with a final score count.

• (optional) Heart-rate window — if the owner wore a Garmin and you have their

HR in InfluxDB, puls-magnus.py adds a "how excited was the parent" series per game.

What it writes into data.json

• matches[] / games[] — the team's games (day, time, venue, opponent, score, result).
• groupResults[] + analysis.ka_family[] — the group table + the team's siblings

(e.g. all KA-1..KA-4) with per-game form.

• analysis — tournament-wide stats (games played, goals, average, number of teams)

and the team in context (goals for/against, defence percentile, etc.).

• wrapup.record — the final W/L/GF/GA once the tournament is done.

Tools

• tools/refresh-urslit.py — scrape the official feed, write CSV + compute the

analysis into data.json. Idempotent: exit 0 = unchanged, exit 3 = changed (so a cron/agent can rebuild only on a real change). Parameterise TEAM, the feed URL, the day codes, and the group at the top of the file.

• tools/puls-magnus.py — (optional) pull the owner's heart-rate per game window

from InfluxDB. Iterates ALL of the team's games via analysis.ka_family.

Rules

1. Only played games count. No guessing, no projected scores. Show "last updated".
2. Honest provenance. If a game is hand-added (a placement final not yet on the

feed), record it as hand-entered — never claim it came "straight from the official system". The dashboard's caption must say so.

1. Run on an interval, rebuild on change. A small Python agent on a timer scrapes

the feed; only exit 3 triggers a redeploy. Add a killswitch date so it stops after the tournament.

Run as an agent

A large-language-model-powered agent + this Python on a schedule = live results that update themselves. The agent decides when something material changed and narrates it; the Python does the deterministic scrape + diff.

— pattern by Magnús Smári Smárason · https://www.smarason.is

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: Magnussmari
• Source: Magnussmari/TmMot2026-Skill
• License: MIT

Install and usage instructions live in the source repository linked above.