Skip to content
Skills Library
Data & analytics · Planning

ml-problem-framing

/ml-problem-framing

Frame ML problems around a decision, a measurable target, and honest error costs before touching models. Use when scoping an ML project or reviewing whether one should exist.

Repository
AI-SKILLS ↗
Licence
MIT
Last repository change
05.09.2026

Assessment and security

90/ 100Class: good
Format25 / 25
Frontmatter with name and description, valid name, description up to 1,024 characters, name matching the folder.
Discoverability25 / 25
Description explains when to use the skill, has a reasonable length and uses third person.
Content25 / 35
At least 150 words and at most 500 lines, headings, step lists, examples or code.
Hygiene15 / 15
No placeholders such as TODO, no Windows paths, with a licence.
Riskno signals detected in the scan
Files in this package
1
Scripts
0
HTML comments
0
URLs in SKILL.md
0
Deprecation

No signals detected. Repository changed on 05.09.2026, without references to retired models or APIs.

Licence

MIT, open: use and modify while retaining author attribution.

What it does and when to use it

Frame ML problems around a decision, a measurable target, and honest error costs before touching models. Use when scoping an ML project or reviewing whether one should exist.

Original author description

Frame ML problems around a decision, a measurable target, and honest error costs before touching models. Use when scoping an ML project or reviewing whether one should exist.

How it works

  1. Method
  2. Boundaries
Original SKILL.md headings · 55 lines, 405 words
SKILL.md preview
---
name: ml-problem-framing
description: Frame ML problems around a decision, a measurable target, and honest error costs before touching models. Use when scoping an ML project or reviewing whether one should exist.
---

# ML problem framing

Most failed ML projects were framed wrong, not modeled wrong. The frame
is: what decision changes, what exactly is predicted, what does each
kind of error cost, and would something simpler make the same decision.

## Method

1. **Start from the decision, not the data.** Name the action the
   prediction triggers ("flag order for review", "rank these items",
   "set this price") and who or what consumes it. A prediction nobody
   acts on is a dashboard; if the action is unclear, stop here.
2. **Specify the target variable ruthlessly.** "Churn" becomes
   "no login within 30 days of day X, for accounts active in the
   prior 60". Definition choices (window, population, exclusions)
   change the problem more than any model choice will; write them
   down and get the domain owner to sign (see warehouse-modeling on
   metric definitions).
3. **Verify the label's availability and honesty.** When does ground
   truth arrive (instant, 30 days, never)? Is it censored (you only
   see fraud you caught), biased by the current process (loan
   defaults observed only for approved loans), or drifting? Label
   latency bounds your retraining loop; label bias caps what any
   model can learn.
4. **Price the errors asymmetrically.** False positive cost vs false
   negative cost, in money or user harm, from the decision owner.
   This drives the metric (see model-evaluation), the operating
   threshold, and often the architecture (a cheap model plus human
   review can beat an expensive autonomous one).
5. **Establish the non-ML baseline first.** The current heuristic,
   a rules list, a single feature threshold, or "predict the
   historical mean" (see ml-baselines). The project's real question
   is the delta over this baseline at the decision, not the model's
   absolute score; sometimes the baseline wins and the project
   should become three IF statements.
6. **Define success upstream and downstream.** Offline: metric and
   threshold on a proper holdout (see train-test-discipline).
   Online: the business metric the decision should move, measured by
   experiment (see ab-test-design), plus guardrails (latency budget,
   fairness slices, override rate). Agree on the kill criteria
   before building; projects without them run on sunk cost.

## Boundaries

- Problems needing 100% correctness, full explainability by law, or
  where errors are catastrophic and unauditable are rules-and-review
  territory, not ML.
- Insufficient or unlabelable data is a data-collection project
  first; committing model timelines before the label pipeline exists
  reverses the dependency.
- Frame once, revisit at every scope change; silent target drift
  ("churn now means downgrade too") invalidates every past
  evaluation.

Versions

The database has no other skills named ml-problem-framing.

Database snapshot: 13.11.2026 · package version 93b9943bd9bc (SHA-256 digest of the entire package)

Downloads in the last 30 days: 0 · ranking

Report a problem with this skill: harmful content, incorrect descriptions or an author’s removal request.

Similar skills

  • 100framing-ml-problems
    Frames a data/ML problem before any modeling — the decision, the action, the cost of a wrong call, task type, target, and success metric. Use when picking a project direction, writing a research question, or mapping a problem onto an ML task type.
  • 100problem-framing-canvas
    Guide teams through MITRE's Problem Framing Canvas. Use when you need a clearer problem statement before jumping to solutions.
  • 90problem-framing
    Problem Framing. Turn a business brief, PRD, dataset description or stakeholder question into a one-page ANALYSIS_MANIFEST.md before any modeling. Use when someone asks what to predict, how to measure success, which metric or baseline to use, the cost of errors, or how to scope…
  • 100021-problem-framing
    Use when a problem or issue needs explicit framing before deeper analysis begins — establishing the Problem statement, Current state, Desired state, Stakeholders, and Success criteria. This should trigger when an issue's Problem Framing point of view needs evaluation, or when a…
  • 90problem-framing
    Use when a team is converging on solutions before agreeing on the problem, when a brief reads as a feature request, when symptoms and root needs are tangled, or when assumptions need surfacing before design work proceeds. Do NOT use for code-level bug triage, runtime failure…
  • 100problem-framing
    Use when the user needs to sharpen a vague problem into pain points, reframed questions, and ranked opportunities before designing solutions. Triggers on "frame the problem", "pain points", "how might we", "HMW", "opportunity assessment", "what problem", and the same intent in any language.

Feedback and reports

Your message goes to Bartosz Sokoliński. An email address is needed only for a reply. See the privacy policy.