# Ewaluacja i interpretowalność

EN: **Evaluation and interpretability** · id `evaluation`

Path: [AI Atlas](https://aiatlasgraph.com/llms.txt) › Ewaluacja i interpretowalność

## Description

- PL: Jak sprawdzać jakość i rozumieć ograniczenia modeli oraz systemów.
- EN: How to check quality and understand the limits of models and systems.

## In one sentence

- PL: To sprawdzanie, jak dobrze model działa, gdzie się myli i co można ustalić o jego zachowaniu.
- EN: This checks how well a model works, where it makes mistakes and what can be learned about its behaviour.

## Problem it solves

- PL: Przekonująca odpowiedź lub udany pokaz nie mówi, czy system będzie działał poprawnie w różnych rzeczywistych sytuacjach.
- EN: A convincing answer or successful demonstration does not show whether a system will work correctly across real situations.

## What it does

- PL: Łączy testy, analizę błędów, oceny ludzi, obserwację działania i badanie modelu, dobierając miary do zadania oraz jego ryzyka.
- EN: It combines tests, error analysis, human assessments, operational monitoring and model studies, choosing measures suited to the task and its risks.

## Everyday example

- PL: Przed użyciem asystenta sprawdzasz typowe i trudne sprawy, a po wdrożeniu oglądasz pomyłki, koszty i zmiany jakości.
- EN: Before using an assistant, you test ordinary and difficult cases; after deployment, you examine mistakes, costs and changes in quality.

## This concerns you if you…

- PL: chcesz sprawdzić działanie modelu poza udanym pokazem
- PL: musisz znać pomyłki, ograniczenia i jakość systemu
- EN: want to assess a model beyond a successful demonstration
- EN: need to understand errors, limits and system quality

## Keywords

regression evaluation · LLM-as-a-judge · error analysis · confidence intervals · calibration · distribution shift · SHAP · LIME · activation patching · sparse autoencoders · agent benchmarks · execution tracing

## How it works

- PL: Kontrolowane próby łączą zadania, odpowiedzi referencyjne i reguły oceny z zapisami wykonania modelu. Analiza błędów, niepewności statystycznej oraz interwencje w modelu pozwalają oceniać zachowanie i sprawdzać hipotezy o jego mechanizmie.
- EN: Controlled trials combine tasks, reference answers and scoring rules with model execution records. Error analysis, statistical uncertainty and model interventions assess behaviour and test hypotheses about its mechanisms.

## Limits

- PL: Wynik zależy od próby, kryteriów i warunków wykonania. Agregaty ukrywają rzadkie błędy, a czytelne wyjaśnienie nie musi wiernie odtwarzać obliczeń modelu.
- EN: Results depend on samples, criteria and execution conditions. Aggregates hide rare failures, and readable explanations may not faithfully represent model computations.

## Key terms

- PL: **metryka**: Reguła przekształcająca wyniki prób w porównywalną ocenę.
- PL: **walidacja**: Sprawdzanie hipotezy lub rozwiązania na danych niezależnych od jego doboru.
- PL: **wierność wyjaśnienia**: Zgodność wyjaśnienia z rzeczywistym sposobem obliczania predykcji.
- EN: **metric**: A rule converting trial outcomes into a comparable score.
- EN: **validation**: Checking a hypothesis or solution using data independent of its selection.
- EN: **explanation faithfulness**: Agreement between an explanation and the actual prediction computation.

## Know first

- [Statystyka i niepewność (Statistics and uncertainty)](https://aiatlasgraph.com/t/probability.md)
- [Uczenie nadzorowane (Supervised learning)](https://aiatlasgraph.com/t/supervised.md)

## Subtopics (9 direct, 53 in total)

- [Analiza błędów (Error analysis)](https://aiatlasgraph.com/t/erroranalysis.md): Reading real traces of a system and turning the notes into a list of the ways it fails. Only that list tells you what is worth measuring.
  - [Kodowanie otwarte i taksonomia błędów (Open and axial coding to build a failure taxonomy)](https://aiatlasgraph.com/t/failuretaxonomy.md): Qualitative trace analysis: open coding of failure notes, then axial coding into a taxonomy of failure modes.
  - [Przegląd i próbkowanie śladów (Trace review and sampling)](https://aiatlasgraph.com/t/tracereview.md): Which session records to look at, and in what tool, so a person can review them quickly and carefully.
- [Ewaluacja RAG i agentów (RAG and agent evaluation)](https://aiatlasgraph.com/t/agenteval.md): Checking sources, trajectories and the final result.
  - [Miary jakości wyszukiwania (recall@k, MRR, nDCG) (Retrieval quality metrics (recall@k, MRR, nDCG))](https://aiatlasgraph.com/t/retrievalmetrics.md): Ranking metrics that grade the retrieval stage against labelled relevant documents.
  - [Ocena po stanie końcowym zamiast po przebiegu (Outcome-based vs transcript-based grading of agents)](https://aiatlasgraph.com/t/outcomegrading.md): Grading agents by verifying environment state after a trial, complemented by checks on the transcript.
  - [Powtarzalność agenta: pass@k i pass^k (Agent reliability metrics: pass@k and pass^k)](https://aiatlasgraph.com/t/passk.md): Metrics from repeated trials: pass@k (at least one success in k) and pass^k (all k succeed).
  - [Symulowany użytkownik w testach rozmów wieloetapowych (Simulated users for multi-turn agent evaluation)](https://aiatlasgraph.com/t/usersimulation.md): Evaluating conversational agents with an LLM that role-plays the user from a hidden task instruction.
  - [Wierność odpowiedzi wobec znalezionych źródeł (RAG) (RAG faithfulness and answer relevance metrics)](https://aiatlasgraph.com/t/ragfaithfulness.md): RAG metrics scoring how grounded an answer is in its context, how relevant it is to the question and how relevant the context is.
- [Ewaluatory kodowe (Code-based evaluators)](https://aiatlasgraph.com/t/codeevals.md): A deterministic check in code when a failure can be described by a rule. The cheapest rung of evaluation — before an LLM judge.
  - [Asercje deterministyczne: format, schemat, dopasowanie tekstu (Deterministic assertions: format, schema and string checks)](https://aiatlasgraph.com/t/assertionchecks.md): Code graders that check outputs with rules: exact or regex match, schema validation, length and keyword presence.
  - [Ocena przez uruchomienie testów (poprawność funkcjonalna) (Execution-based grading with unit tests)](https://aiatlasgraph.com/t/executiongrading.md): Grading generated code by executing it in a sandbox against unit or integration tests.
- [Interpretowalność (Interpretability and explainability)](https://aiatlasgraph.com/t/explainability.md): How to understand why a model made a given decision: from models that are transparent by design, through explanations added from outside (SHAP, saliency maps), to looking inside the network.
  - [Interpretowalność mechanistyczna (Mechanistic interpretability)](https://aiatlasgraph.com/t/mechinterp.md): Studying a model's internal components and computations and testing how they contribute to its behaviour.
    - [Obwody i grafy atrybucji (circuit tracing) (Circuits and attribution graphs)](https://aiatlasgraph.com/t/circuittracing.md): Building an approximate graph of features and connections that may mediate the computation behind a model answer. The graph requires intervention-based validation and does not guarantee a complete, faithful reconstruction of the mechanism.
    - [Podmiana aktywacji (activation patching) (Activation patching)](https://aiatlasgraph.com/t/activationpatching.md): An intervention experiment: activations of a chosen component are replaced with activations from another input and the output change is measured. This provides causal evidence whose interpretation depends on the experimental design and metric.
    - [Rzadkie autoenkodery (SAE) i słowniki cech (Sparse autoencoders and dictionary learning)](https://aiatlasgraph.com/t/sae.md): Decomposing model activations into many sparsely active features, some of which can be associated with understandable concepts. Feature interpretability and practical downstream usefulness are not guaranteed.
  - [Mapy istotności dla sieci neuronowych (Saliency maps and gradient attribution)](https://aiatlasgraph.com/t/saliencymaps.md): Visualisations assigning importance to pixels or tokens for a network output using gradients or activations. High importance is not by itself evidence of causal influence.
  - [Modele interpretowalne z natury (Inherently interpretable models)](https://aiatlasgraph.com/t/interpretablemodels.md): Models simple enough that their decision can be read directly — from weights, rules or a tree path — without extra explanation tools.
  - [Ogólne zachowanie modelu: ważność cech i wykresy zależności (Global model explanations: feature importance and dependence plots)](https://aiatlasgraph.com/t/globalexplanations.md): A description of model behaviour across a dataset: which features matter most to predictions and how model output varies as a feature changes. Such plots do not by themselves establish causality.
  - [Sondy klasyfikacyjne (probing) (Probing classifiers)](https://aiatlasgraph.com/t/probing.md): A classifier trained on a model's internal representations to test whether a property can be decoded from them. Strong probe performance alone does not prove that the model uses that information in its behaviour.
  - [Sterowanie aktywacjami (wektory sterujące) (Activation steering and representation engineering)](https://aiatlasgraph.com/t/activationsteering.md): Finding directions or representations in activations associated with a concept and amplifying or suppressing them at run time to study or change model behaviour without retraining.
  - [Wyjaśnianie pojedynczej decyzji (SHAP, LIME) (Local feature attribution (SHAP, LIME))](https://aiatlasgraph.com/t/featureattribution.md): Methods that explain an individual prediction by assigning contributions to input features. They may treat the model as a black box or exploit its structure.
  - [Wyjaśnienia „co by było, gdyby” (kontrfaktyczne) (Counterfactual explanations)](https://aiatlasgraph.com/t/counterfactualexpl.md): The smallest change to the input that would flip the model's decision — e.g. what would have to change in an application for the loan to be approved.
- [Metryki i benchmarki (Metrics and benchmarks)](https://aiatlasgraph.com/t/metrics.md): Choosing a measure that fits the task and comparing models reliably: classic metrics, standardised benchmarks, leaderboards and statistical uncertainty in results.
  - [Benchmarki modeli językowych (LLM benchmarks)](https://aiatlasgraph.com/t/llmbenchmarks.md): Standardised task sets used to compare language models. They are grouped by the skill measured, such as knowledge and reasoning, coding or agent work.
    - [Benchmarki agentowe (Agent benchmarks)](https://aiatlasgraph.com/t/agentbench.md): Environments where a model carries out multi-step tasks with tools — in a browser, an operating system or a conversation with a simulated customer — and is scored on whether the goal was achieved.
    - [Benchmarki długiego kontekstu (igła w stogu siana i dalej) (Long-context benchmarks (needle-in-a-haystack and beyond))](https://aiatlasgraph.com/t/longcontextbench.md): Synthetic and realistic tests of effective context length: retrieving planted facts, multi-hop tracing and aggregation.
    - [Benchmarki programistyczne (Coding benchmarks)](https://aiatlasgraph.com/t/codingbench.md): Tasks in which a model writes or fixes code and tests check the result — from single functions to real issues in repositories.
    - [Benchmarki wiedzy i rozumowania (Knowledge and reasoning benchmarks)](https://aiatlasgraph.com/t/knowledgebench.md): Closed-ended questions across many fields, from school level to expert level, graded automatically. Older sets saturate quickly, so ever harder ones are created.
    - [Przeciek danych i nasycenie benchmarków (Benchmark contamination and saturation)](https://aiatlasgraph.com/t/benchcontamination.md): Why a benchmark score can mislead: test questions may have leaked into training data, and popular tests stop separating top models once all score near 100%.
  - [Metryki klasyfikacji i regresji (Classification and regression metrics)](https://aiatlasgraph.com/t/classmetrics.md): Classic measures for models that predict a class or a number. The choice depends on the cost of errors and whether classes are balanced.
  - [Niepewność statystyczna wyników ewaluacji (Statistical uncertainty in evaluations)](https://aiatlasgraph.com/t/evalstats.md): An evaluation is an experiment on a sample of tasks, so its score is uncertain. Without estimating that uncertainty, a small gap between models may be indistinguishable from sampling variation.
  - [Porównanie tekstu z wzorcem (BLEU, ROUGE, BERTScore) (Reference-based text generation metrics)](https://aiatlasgraph.com/t/textgenmetrics.md): Automatic scoring of generated text by comparing it to a reference answer — from counting shared words to meaning similarity in embeddings.
  - [Rankingi z głosami ludzi (areny) (Human-preference leaderboards (arenas))](https://aiatlasgraph.com/t/arenas.md): People anonymously compare two model answers and choose the better one; many such comparisons produce a ranking. The result nevertheless depends on vote collection, models' access to the arena and comparison sampling.
- [Monitoring produkcyjny (Monitoring and observability)](https://aiatlasgraph.com/t/monitoring.md): Watching a system after deployment: traces, cost, regressions.
  - [Ewaluacja online na ruchu produkcyjnym (Online evaluation on production traffic)](https://aiatlasgraph.com/t/onlineevals.md): Continuous grading of sampled production traffic with code graders and LLM judges, complementing offline evals.
  - [Wykrywanie dryfu danych i pojęć (Data and concept drift detection)](https://aiatlasgraph.com/t/driftdetection.md): Detecting that the production-data distribution or the relationship between inputs and correct outputs has changed relative to training data, which may require diagnosis or model adaptation.
  - [Śledzenie wywołań modeli i agentów (tracing) (LLM and agent tracing)](https://aiatlasgraph.com/t/llmtracing.md): Instrumenting LLM applications to record model, tool and agent calls as a tree of spans with attributes.
- [Ocena ludzi i modeli (Human evaluation and model judges)](https://aiatlasgraph.com/t/judges.md): Judging answers, with quality control on the judging itself.
  - [Model językowy jako sędzia (LLM-as-a-judge) (LLM-as-a-judge)](https://aiatlasgraph.com/t/llmjudge.md): Using a language model as an automatic evaluator in single-answer, pairwise or reference-guided mode.
    - [Uprzedzenia sędziów LLM: kolejność, długość, własny styl (LLM judge biases and mitigation)](https://aiatlasgraph.com/t/judgebias.md): Systematic errors of LLM judges (position, verbosity, self-preference) and techniques to reduce them.
  - [Ocena według rubryki i listy kontrolnej (Rubric-based and checklist grading)](https://aiatlasgraph.com/t/rubricgrading.md): Grading answers against explicit criteria with level descriptions or binary checklist items, done by people or an LLM judge.
  - [Walidacja ewaluatorów (Evaluator validation against human labels)](https://aiatlasgraph.com/t/judgevalidation.md): Testing an LLM judge or classifier against human labels, on a set that played no part in tuning it.
- [Odporność i niepewność (Robustness and calibration)](https://aiatlasgraph.com/t/robustness.md): Behaviour on unusual data, and how trustworthy the confidence is.
  - [Kalibracja pewności modelu (Confidence calibration)](https://aiatlasgraph.com/t/calibration.md): Whether a predicted probability or other confidence score matches how often the model is right — e.g. whether 80%-confidence predictions are correct about 80% of the time — and how to correct a mismatch.
  - [Predykcja konformalna (Conformal prediction)](https://aiatlasgraph.com/t/conformalprediction.md): Methods that equip any model with prediction sets or intervals having controlled marginal coverage under exchangeable data, without parametric distribution assumptions.
  - [Wykrywanie halucynacji (Hallucination detection)](https://aiatlasgraph.com/t/hallucinationdetect.md): Checking whether a model is inventing facts: against the provided source, against outside knowledge, or by how much its answers vary when the question is repeated.
  - [Wykrywanie przykładów spoza rozkładu (OOD) (Out-of-distribution detection)](https://aiatlasgraph.com/t/ooddetection.md): Recognising inputs that do not fit the training-data distribution, so the system can raise an alert, abstain or hand the decision to a person.
- [Zbiory testowe i regresje (Evaluation datasets and regression tests)](https://aiatlasgraph.com/t/evalsets.md): Repeatable checks on behaviour after changes.
  - [Ewaluacja agentów zgodna z przepływem zadań (Task-flow aligned agent evaluation)](https://aiatlasgraph.com/t/taskfloweval.md): Method turns issue-derived tasks into multi-turn interactions matching measured user task flows.
  - [Frameworki do uruchamiania ewaluacji (Evaluation harnesses and frameworks)](https://aiatlasgraph.com/t/evalharness.md): Tools that run test sets against a model or agent repeatably — with the same prompts, settings, sandbox and logs — so results can be compared and reproduced.
  - [Syntetyczne przypadki testowe z wymiarów zapytań (Synthetic test case generation for evals)](https://aiatlasgraph.com/t/synthevalcases.md): Building eval sets from query variation dimensions, value tuples and a separate conversion step into natural language.

## Sources

- [Full Stack Deep Learning (2022, archiwum)](<https://fullstackdeeplearning.com/course/2022/>) `fsdl` · reference: Pełny cykl systemu ML: dane, testy, wdrażanie, monitoring, zespoły i etyka.
- [Hugging Face · Agents Course](<https://huggingface.co/learn/agents-course/en/unit0/introduction>) `agents` · reference: Agenci, narzędzia, frameworki, agentic RAG i ewaluacja.
- [Stanford CS229 · Machine Learning (2018, archiwum)](<https://cs229.stanford.edu/syllabus-autumn2018.html>) `cs229` · reference: Klasyczne ML, statystyka i podstawy uczenia. Użyte jako stabilny szkielet, nie przegląd nowości.
- [DeepLearning.AI — course catalogue](<https://www.deeplearning.ai/courses>) `dlai` · course catalogue: 42 topic labels with counts, about 130 courses.

---

AI Atlas revision 122 (2026-10-30) · HTML: https://aiatlasgraph.com/t/evaluation · index: https://aiatlasgraph.com/llms.txt · everything: https://aiatlasgraph.com/llms-full.txt · updated 2026-10-30