Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add skills/danieltomaro13/sportsdata-agents/model_developmentnpx skills add DanielTomaro13/sportsdata-agents --skill model_developmentgit clone --depth 1 https://github.com/DanielTomaro13/sportsdata-agentsWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/danieltomaro13/sportsdata-agents/model_development)<a href="https://agentmods.dev/skills/danieltomaro13/sportsdata-agents/model_development"><img src="https://agentmods.dev/badge/skills/danieltomaro13/sportsdata-agents/model_development.svg" alt="Measured on agentmods" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00040 | $0.00852 |
| Opus 5 | $0.00020 | $0.00426 |
| Sonnet 5 | $0.00008 | $0.00170 |
| Haiku 4.5 | $0.00004 | $0.00085 |
Grade A, and why
model_development scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 6d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 71 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Model development — the general method
This is the method for every market (winners, totals, lines, props). Market recipes
(build_a_totals_model, build_a_h2h_model) are worked examples OF this method,
never replacements for it.
1. Frame the problem
- What exactly is being predicted? One binary probability per selection is the default (P(home win), P(over)). Name the market convention (whose line, which book) before any code.
- What is the decision the probability feeds? A backtest entry needs the prob
BEFORE the price moves — timestamps are part of the model contract
(
record_predictionstakespredicted_at; never backdate dishonestly).
2. Sample size BEFORE cleverness
- Rule of thumb: 10–20 outcomes per model parameter as a floor. A logistic regression with 5 features wants 100+ decided events; anything fancier wants several times that.
- "Last 20 games" is rarely enough — it estimates a mean with ±20%+ noise and one parameter eats half of it. Prefer multiple seasons with recency weighting (exponential decay) over tiny recent windows.
- Sport cadence changes everything: an MLB team plays 162 games/season, an NFL team 17 — the same "two seasons of data" is 324 events in one sport and 34 in the other. Count EVENTS, not calendar time.
- Watch regime changes: rule changes, roster turnover, venue moves. Old data is only valuable while the process that generated it still operates — say so when you truncate history and why.
- When the data cannot support the model requested, SAY THAT PLAINLY and build the smaller model that it can support.
3. Features: ask AND measure
- When the request is open-ended, ask the user which stats they believe matter for this market — domain priors are real information and it is their model.
- Independently measure importance from the data (univariate signal, simple permutation importance in run_python) — then report where the user's priors and the data disagree, with numbers. Do not silently drop either.
- Fewer features beat more: every feature is a parameter and §2 already priced parameters. Justify each one in a sentence.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 6d ago First seen · 71 lines · 40 tokens per session scan A 75db9d41db75
model_development is a skill published in the GitHub repository DanielTomaro13/sportsdata-agents (6 stars, last pushed 3d ago), licensed MIT. It adds 40 tokens to every session and 852 once invoked, about $0.0002 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other skills, from other repositories
mle-workflow
Production machine-learning engineering workflow for data contracts, reproducible training, model evaluation, deployment, monitoring, and rollback. Use when building, reviewing, or hardening ML systems beyond one-off notebooks.
weights-and-biases
W&B: log ML experiments, sweeps, model registry, dashboards.
flash-attention
Speed up long-sequence transformer training and inference.
nemo-curator
Curate LLM training data: dedupe, filter, PII redaction.
llava
Vision-language chat: VQA, captioning, image dialogue.
simpo
Reference-free preference alignment, simpler than DPO.