GCRI · Zîrekiya çêkirî û pergalên xweser
Nirxandina modelê
Model evaluation asks what an AI model can do, how reliably, and where it fails. Benchmarks, red teaming, human studies and field monitoring each measure something different, and none fully predicts behaviour in deployment. This page sets out how evaluations are designed and where they mislead, how results should be reported, and what evidence serves the developers, deployers and regulators who rely on them.
Model evaluation in seven parts.
- 01The questionModel evaluation: how the risk develops
- 02Li seranserê qadanEvaluation must reflect the domain where a model is used
- 03NîşanderWhat is measured, and how well
- 04ÇavkanîWhere the evidence comes from
- 05BiryarWho decides, and what the evidence serves
- 06Core questionsWhat evaluators ask
- 07MechanismsWhy scores mislead
Model evaluation: how the risk develops.
Validity, calibration and generalisation. The dossier below sets out how the risk develops, what is measured, where the data come from, how it is modelled and who decides.
Benchmarks saturate and leak
Models quickly reach high scores on fixed benchmarks, and test items can leak into training data. High scores then say more about the benchmark than about general capability.
Test conditions differ from use
Evaluations use set prompts, settings and tasks. Real users phrase things differently, combine tools and face new situations, so measured performance may not transfer.
Adversarial testing finds hidden failures
Red teaming and adversarial prompts reveal unsafe or unreliable behaviour that average-case metrics miss. Coverage depends on the testers’ skill and the time allowed.
Performance drifts after deployment
Model updates, changing inputs and new uses alter behaviour over time. Without monitoring, a model evaluated at release can behave differently months later.
Tomara modelan.
The screen a team would use to follow this subject: its measures, states and decisions. Names are examples and values are illustrative.
Tomara modelan
Model evaluation · Model registry · Release gate
Model
| M-184 triage assistant | Xizmeta giştî | Bilind | 19 of 24 suites | Hatiye asteng kirin |
| M-320 flood nowcast | Zanist | Navîn | 12 of 12 | Hatiye weşandin |
| M-157 document search | Navxweyî | Kêm | 8 of 8 | Hatiye weşandin |
| M-563 agent planner | Ajanî | Bilind | 12 of 16 | Tê nirxandin |
Karta modelê
- Bikaranîna armanckirî
- piştgiriya hilbijartina pêşîniyê, mirov biryarê dide
- Li derveyî çarçoveyê
- biryarên mafdariyê
- Sînorên naskirî
- rastbûna kêmtir li ser rewşên kêm
- Xwedî
- Xwediyê modelê
Têketina bûyeran
- D-21INCÇavkaniya xeyalî di kurteyê de
- D-9MITBingehkirina vegerandinê hat zêdekirin
- D-2INCDerzîya fermanê bi pelê barkirî
- D-1MITDestûra amûrê hate teng kirin
Evaluation scores
vs release threshold
Evaluation must reflect the domain where a model is used.
- TenduristîGirêdanClinical AI must be evaluated on the populations it serves.What to watchPerformance by subgroup
- AvGirêdanHydrological ML must be tested out of sample in new basins.What to watchPerformance in ungauged basins
- XwarinGirêdanCrop monitoring models need local ground truth.What to watchAccuracy against field data
Evaluations are designed for the rêzeya Labs Evidence. The International AI Safety Report (2025) describes the limits of current evaluations; NIST’s AI RMF and its generative AI profile (2024) set measurement practice. Models would carry cards under NS-12 of the proposed Nexus Standards.
- Held-out test sets for a domain.
- Evaluation protocols with known failure cases.
How model evaluation is modelled, measured and tested.
Model, texmînên wî, daneyên wî û nediyariya wî, bi şeweyê ku zanyar an endezyar kontrol bike. Avahî rêbazên weşandî dişopînin. Nirxên hatine xêzkirin nimûneyî ne, ne encam in.
Model, hevkêşe û texmîn
The model structure used for model evaluation, the assumptions it makes and how each is tested.
Ri = f(Hi, Ei, Vi, Ci)metirsî, rûbirûbûn, lawazî û kapasîteya serederiyêEAL = ∑k pk · Lkli ser bûyerên k bi îhtîmala salane pkxt+1 = A xt + B ut + εtA carries the couplings, u the shock, ε the error| Texmîn | Çima tê çêkirin | Çawa tê ceribandin |
|---|---|---|
| Metirsî û rûbirûbûn di nav bûyerekê de serbixwe ne | Entegrala windahiyê ji hesabkirinê re maqûl dihêle | Bi lihevhatineke copula li ser tomarên hevbeş re bidin ber hev |
| Girêdan li nêzî xala xebatê xêzik in | Dihêle şok di A de belav bibin | Ji derveyî rêza xêzik tevlihev bike û bide ber hev |
| Xêzên lawaziyê di navbera cihan de têne veguhastin | Xêzikên herêmî kêm tên dîtin | Li hember windahiyên çavdêrîkirî erê bike li cihê ku hene |
- Têketin
- danehevên bi guhertoyan û çavkaniyê
- Derketin
- belavkirin, tu carî jimarên tenê
- Kod
- bi qerta modelê û ceribandinan re hat weşandin
- Zêdebûn
- heman dane, kod, jîngeh û tov
What is measured, and how well.
Each indicator carries its unit, its cadence, the kind of source behind it and the evidence label a value would carry. No values are shown: a programme supplies them under its own authority.
| Indicator | Unit | Cadence | Source type | Nîşan |
|---|---|---|---|---|
| Benchmark score with uncertainty | score with confidence interval | per evaluation | Published benchmark results | REPORTED |
| Contamination checks reported | percent of evaluations reporting checks | per evaluation | Evaluation reports | DECLARED |
| Red-team findings by severity | count by severity class | per exercise | Red-team reports | REPORTED |
| Performance gap between subgroups | difference in error rates | per evaluation | Disaggregated evaluation | ÇAVDÊRÎKIRÎ |
| Post-deployment incident reports | count by type | monthly | Incident monitoring | REPORTED |
| Calibration error | expected calibration error | per evaluation | Evaluation on held-out data | ÇAVDÊRÎKIRÎ |
Where the evidence comes from.
Named public datasets and institutional records, each with its licence class. Sources by agreement are used only under their terms and never republished.
| Çavkanî | What it gives | Licence class |
|---|---|---|
| Stanford HELM (Holistic Evaluation of Language Models) | Open, multi-metric evaluations of language models | Open source, Apache 2.0 for code; results public |
| MLCommons benchmarks, including AILuminate | Standardised performance and safety benchmarks | Public results, MLCommons terms (to check) |
| Raporta Navneteweyî ya Ewlehiya AI | Synthesis of evidence on capabilities, risks and evaluation limits | Public, terms per publisher |
| Model cards and system cards from developers | Developer-reported evaluations and limitations | Public documents, terms per developer |
| Epoch AI benchmark and model databases | Compiled benchmark results and model characteristics | Open, CC BY 4.0 |
Who decides, and what the evidence serves.
Zîrekiya çêkirî û Pergal prepares evidence. The decisions below are taken by the authorities named, under their own rules, and never by the platform.
| Kî biryarê dide | The question | Evidence this page serves |
|---|---|---|
| Pêşdebir | Whether a model is ready for release and what limits to state | Evaluation protocols; red-team findings |
| Deploying organisations | Whether a model is fit for a specific use | Use-specific evaluations; subgroup performance |
| Regulators and AI safety institutes | What evaluation evidence to require | Evaluation standards and reporting templates |
Outputs this topic produces
- Evaluation protocols with uncertainty and contamination checks
- Disaggregated evaluation designs for subgroup performance
- Red-team methods for public sector use cases
- Evaluation reporting templates
What evaluators ask.
- Does the test measure the capability or property that matters for the intended use (construct validity)?
- When the model reports high confidence, is it right as often as that confidence implies (calibration)?
- Does performance hold on data from different places, times or populations (generalisation)?
- Has test data leaked into training, inflating results (contamination)?
Why scores mislead.
Models learn statistical regularities in their training data, including spurious ones. A medical imaging model may learn the hospital’s scanner signature rather than the disease; a language model may recognise benchmark questions it saw during training. When deployment data differ from test data, a problem known as guherîna belavbûnê, performance can fall sharply and without warning.
Large general-purpose models add new difficulties. Their capabilities are broad and emergent, outputs vary with phrasing, and the line between capability and memorisation is blurred. Aggregate scores can hide failures concentrated in particular groups or tasks.
How evaluation is done, and its limits.
- 01
Held-out and out-of-distribution testing
Testing on data deliberately different from training, analysed with îhtîmal û statîstîk. No test set covers every deployment condition.
- 02
Calibration analysis
Comparing stated confidence with observed accuracy, and checking how it changes under shift.
- 03
Disaggregated evaluation
Reporting performance by subgroup and task rather than only on average. Small subgroups give uncertain estimates.
- 04
Human and expert evaluation
Structured assessment of open-ended outputs, using zanista civakî û reftarî methods for rater agreement. Human judgement is costly and can be inconsistent.
How it is studied, and what the methods cannot say.
- 01Endezyariya nirxandin û teminata zîrekiya çêkirîAssurance cases for AI systems · Structured red teaming · Post-deployment monitoringApplied here
- 02Îhtîmal û statîstîkUncertainty on benchmark scores · Calibration assessment · Audit samplingApplied here
- 03Zîrekiya çêkirî û fêrbûna makîneyêDistribution shift analysis · Fine-tuning for a specific use · Interpretability methodsApplied here
- 04Zanistên civakî û tevgerîHuman oversight studies · Affected community research · Workplace adoption studiesApplied here
Where this page’s references lead.
Sazî û belgeyên ku ev rûpel li hember wan tê xwendin. Rewş û çap diguherin; her girêdan diçe ba saziya derxistinê.
Ev rûpel li ser Konsorsiyûmê te ber bi kû ve dibe.
Pirs û bersiv.
Why are benchmark scores not enough?
What is red teaming in AI?
What is data contamination?
Is a high benchmark score evidence a model is safe to deploy?
What is benchmark saturation?
Why evaluate after deployment too?
Will the platform publish evaluation protocols?
Ji vê rûpelê delîl bidin.
rêzeya Pergalên Stratejîk is the Consortium’s quarterly publication. What you know about this subject can become evidence, a case or a correction in the next edition, credited to you and published under CC BY 4.0 where it is your own work.
- E-nameDergeh GW-GCRIJi vê rûpelê bişîninDelîl, rastkirin, doz, rêbaz, nirxandin, wergerandin an têketina profîlê. Bi nîşana vê rûpelê digihîje, ji ber vê yekê edîtorê xwediyê wê wê bistîne.
- ZenodoDepoya vekirîWê razînin û ji civakê re bişîninTomara xwe li Zenodo bi ORCID iD ya xwe biweşînin, paşê wê li wir ji civaka Reports re bişînin. Kurator wê vedikolin; DOI ya we dimîne.
- rêzeya Pergalên StratejîkGCRIHemû rêyên beşdariyê bibîneCureyên beşdariyê, her yek çawa tê nirxandin, tu çi navlêkirinê distînî, lîsans û tiştê ku qet nayê qebûlkirin.
Tu carî daneyên şexsî, hûrgiliyên binesaza hesas an tiştekî ku mafê we yê parvekirinê tune ne neşînin. Beşdarî berî bikaranînê tê nirxandin; şandina wê nivîskarî an mafên edîtorî nade û sponsorî tu carî encamekê nakire.
Rêbazek, danehevek an pirsek bîne.
Agahî û pirs îro vekirî ne. Pirsek referansekê distîne û gava piştî wê dide nasîn; serlêdan, drav, veqetandina hesabkirinê û tomarkirin tenê dema ku tîm, mercên wê û kapasîteya vekolînê hebin vedibin.
Belgeyên hiqûqî.
Nexşe, dane û jimarên vê malperê ji bo çarçoveyê û agahiyên giştî têne nîşandan. Ew ne şîret, ne pêşbînî û ne hişyarî ne; sînorên agahiyê di Daxuyaniyên Berpirsiyariyê û Daxuyaniyên Daneyê de hatine diyarkirin.