# Alexey Gavrilov — Research > Research on discrete latent representations, compressed text generation, and inspectable control for software regeneration. The canonical human-readable pages below contain verified bibliographic metadata, full abstracts, key results, limitations, and citation files. Do not infer a public code release when a page states that no implementation is available. ## Canonical indexes - English LLM research index (en) Canonical compact machine-readable index for the English source edition. URL: https://aogavrilov.com/llms.txt - Human-readable site index The four editorial source editions in one human-readable crawl path; all 41 fixed language variants remain available through reciprocal hreflang and the XML sitemap. URL: https://aogavrilov.com/sitemap/ - Publication archive URL: https://aogavrilov.com/publications/ - Machine-readable publication index URL: https://aogavrilov.com/publications.json - FAIR interoperability catalog URL: https://aogavrilov.com/.well-known/api-catalog - OpenAPI retrieval description URL: https://aogavrilov.com/openapi.json - OAI-PMH 2.0 repository URL: https://aogavrilov.com/oai?verb=Identify - Scholarly-object Signmap URL: https://aogavrilov.com/signmap.xml - ResourceSync capability list URL: https://aogavrilov.com/resourcesync/capabilitylist.xml - ResourceSync resource list URL: https://aogavrilov.com/resourcesync/resourcelist.xml - English Atom feed (en) URL: https://aogavrilov.com/atom.xml - English RSS feed (en) URL: https://aogavrilov.com/rss.xml - English JSON Feed (en) URL: https://aogavrilov.com/feed.json - Русский Atom feed (ru) URL: https://aogavrilov.com/ru/atom.xml - Русский RSS feed (ru) URL: https://aogavrilov.com/ru/rss.xml - Русский JSON Feed (ru) URL: https://aogavrilov.com/ru/feed.json - 中文（简体） Atom feed (zh-Hans) URL: https://aogavrilov.com/zh/atom.xml - 中文（简体） RSS feed (zh-Hans) URL: https://aogavrilov.com/zh/rss.xml - 中文（简体） JSON Feed (zh-Hans) URL: https://aogavrilov.com/zh/feed.json - 中文（繁體） Atom feed (zh-Hant) URL: https://aogavrilov.com/zh-hant/atom.xml - 中文（繁體） RSS feed (zh-Hant) URL: https://aogavrilov.com/zh-hant/rss.xml - 中文（繁體） JSON Feed (zh-Hant) URL: https://aogavrilov.com/zh-hant/feed.json - Español Atom feed (es) URL: https://aogavrilov.com/es/atom.xml - Español RSS feed (es) URL: https://aogavrilov.com/es/rss.xml - Español JSON Feed (es) URL: https://aogavrilov.com/es/feed.json - العربية Atom feed (ar) URL: https://aogavrilov.com/ar/atom.xml - العربية RSS feed (ar) URL: https://aogavrilov.com/ar/rss.xml - العربية JSON Feed (ar) URL: https://aogavrilov.com/ar/feed.json - हिन्दी Atom feed (hi) URL: https://aogavrilov.com/hi/atom.xml - हिन्दी RSS feed (hi) URL: https://aogavrilov.com/hi/rss.xml - हिन्दी JSON Feed (hi) URL: https://aogavrilov.com/hi/feed.json - Français Atom feed (fr) URL: https://aogavrilov.com/fr/atom.xml - Français RSS feed (fr) URL: https://aogavrilov.com/fr/rss.xml - Français JSON Feed (fr) URL: https://aogavrilov.com/fr/feed.json - Português Atom feed (pt) URL: https://aogavrilov.com/pt/atom.xml - Português RSS feed (pt) URL: https://aogavrilov.com/pt/rss.xml - Português JSON Feed (pt) URL: https://aogavrilov.com/pt/feed.json - Deutsch Atom feed (de) URL: https://aogavrilov.com/de/atom.xml - Deutsch RSS feed (de) URL: https://aogavrilov.com/de/rss.xml - Deutsch JSON Feed (de) URL: https://aogavrilov.com/de/feed.json - 日本語 Atom feed (ja) URL: https://aogavrilov.com/ja/atom.xml - 日本語 RSS feed (ja) URL: https://aogavrilov.com/ja/rss.xml - 日本語 JSON Feed (ja) URL: https://aogavrilov.com/ja/feed.json - 한국어 Atom feed (ko) URL: https://aogavrilov.com/ko/atom.xml - 한국어 RSS feed (ko) URL: https://aogavrilov.com/ko/rss.xml - 한국어 JSON Feed (ko) URL: https://aogavrilov.com/ko/feed.json - Bahasa Indonesia Atom feed (id) URL: https://aogavrilov.com/id/atom.xml - Bahasa Indonesia RSS feed (id) URL: https://aogavrilov.com/id/rss.xml - Bahasa Indonesia JSON Feed (id) URL: https://aogavrilov.com/id/feed.json - Türkçe Atom feed (tr) URL: https://aogavrilov.com/tr/atom.xml - Türkçe RSS feed (tr) URL: https://aogavrilov.com/tr/rss.xml - Türkçe JSON Feed (tr) URL: https://aogavrilov.com/tr/feed.json - Italiano Atom feed (it) URL: https://aogavrilov.com/it/atom.xml - Italiano RSS feed (it) URL: https://aogavrilov.com/it/rss.xml - Italiano JSON Feed (it) URL: https://aogavrilov.com/it/feed.json - Polski Atom feed (pl) URL: https://aogavrilov.com/pl/atom.xml - Polski RSS feed (pl) URL: https://aogavrilov.com/pl/rss.xml - Polski JSON Feed (pl) URL: https://aogavrilov.com/pl/feed.json - Nederlands Atom feed (nl) URL: https://aogavrilov.com/nl/atom.xml - Nederlands RSS feed (nl) URL: https://aogavrilov.com/nl/rss.xml - Nederlands JSON Feed (nl) URL: https://aogavrilov.com/nl/feed.json - Українська Atom feed (uk) URL: https://aogavrilov.com/uk/atom.xml - Українська RSS feed (uk) URL: https://aogavrilov.com/uk/rss.xml - Українська JSON Feed (uk) URL: https://aogavrilov.com/uk/feed.json - فارسی Atom feed (fa) URL: https://aogavrilov.com/fa/atom.xml - فارسی RSS feed (fa) URL: https://aogavrilov.com/fa/rss.xml - فارسی JSON Feed (fa) URL: https://aogavrilov.com/fa/feed.json - Tiếng Việt Atom feed (vi) URL: https://aogavrilov.com/vi/atom.xml - Tiếng Việt RSS feed (vi) URL: https://aogavrilov.com/vi/rss.xml - Tiếng Việt JSON Feed (vi) URL: https://aogavrilov.com/vi/feed.json - ไทย Atom feed (th) URL: https://aogavrilov.com/th/atom.xml - ไทย RSS feed (th) URL: https://aogavrilov.com/th/rss.xml - ไทย JSON Feed (th) URL: https://aogavrilov.com/th/feed.json - বাংলা Atom feed (bn) URL: https://aogavrilov.com/bn/atom.xml - বাংলা RSS feed (bn) URL: https://aogavrilov.com/bn/rss.xml - বাংলা JSON Feed (bn) URL: https://aogavrilov.com/bn/feed.json - اردو Atom feed (ur) URL: https://aogavrilov.com/ur/atom.xml - اردو RSS feed (ur) URL: https://aogavrilov.com/ur/rss.xml - اردو JSON Feed (ur) URL: https://aogavrilov.com/ur/feed.json - मराठी Atom feed (mr) URL: https://aogavrilov.com/mr/atom.xml - मराठी RSS feed (mr) URL: https://aogavrilov.com/mr/rss.xml - मराठी JSON Feed (mr) URL: https://aogavrilov.com/mr/feed.json - తెలుగు Atom feed (te) URL: https://aogavrilov.com/te/atom.xml - తెలుగు RSS feed (te) URL: https://aogavrilov.com/te/rss.xml - తెలుగు JSON Feed (te) URL: https://aogavrilov.com/te/feed.json - தமிழ் Atom feed (ta) URL: https://aogavrilov.com/ta/atom.xml - தமிழ் RSS feed (ta) URL: https://aogavrilov.com/ta/rss.xml - தமிழ் JSON Feed (ta) URL: https://aogavrilov.com/ta/feed.json - Bahasa Melayu Atom feed (ms) URL: https://aogavrilov.com/ms/atom.xml - Bahasa Melayu RSS feed (ms) URL: https://aogavrilov.com/ms/rss.xml - Bahasa Melayu JSON Feed (ms) URL: https://aogavrilov.com/ms/feed.json - Filipino Atom feed (fil) URL: https://aogavrilov.com/fil/atom.xml - Filipino RSS feed (fil) URL: https://aogavrilov.com/fil/rss.xml - Filipino JSON Feed (fil) URL: https://aogavrilov.com/fil/feed.json - Kiswahili Atom feed (sw) URL: https://aogavrilov.com/sw/atom.xml - Kiswahili RSS feed (sw) URL: https://aogavrilov.com/sw/rss.xml - Kiswahili JSON Feed (sw) URL: https://aogavrilov.com/sw/feed.json - Čeština Atom feed (cs) URL: https://aogavrilov.com/cs/atom.xml - Čeština RSS feed (cs) URL: https://aogavrilov.com/cs/rss.xml - Čeština JSON Feed (cs) URL: https://aogavrilov.com/cs/feed.json - Română Atom feed (ro) URL: https://aogavrilov.com/ro/atom.xml - Română RSS feed (ro) URL: https://aogavrilov.com/ro/rss.xml - Română JSON Feed (ro) URL: https://aogavrilov.com/ro/feed.json - Ελληνικά Atom feed (el) URL: https://aogavrilov.com/el/atom.xml - Ελληνικά RSS feed (el) URL: https://aogavrilov.com/el/rss.xml - Ελληνικά JSON Feed (el) URL: https://aogavrilov.com/el/feed.json - עברית Atom feed (he) URL: https://aogavrilov.com/he/atom.xml - עברית RSS feed (he) URL: https://aogavrilov.com/he/rss.xml - עברית JSON Feed (he) URL: https://aogavrilov.com/he/feed.json - Svenska Atom feed (sv) URL: https://aogavrilov.com/sv/atom.xml - Svenska RSS feed (sv) URL: https://aogavrilov.com/sv/rss.xml - Svenska JSON Feed (sv) URL: https://aogavrilov.com/sv/feed.json - Magyar Atom feed (hu) URL: https://aogavrilov.com/hu/atom.xml - Magyar RSS feed (hu) URL: https://aogavrilov.com/hu/rss.xml - Magyar JSON Feed (hu) URL: https://aogavrilov.com/hu/feed.json - Dansk Atom feed (da) URL: https://aogavrilov.com/da/atom.xml - Dansk RSS feed (da) URL: https://aogavrilov.com/da/rss.xml - Dansk JSON Feed (da) URL: https://aogavrilov.com/da/feed.json - Suomi Atom feed (fi) URL: https://aogavrilov.com/fi/atom.xml - Suomi RSS feed (fi) URL: https://aogavrilov.com/fi/rss.xml - Suomi JSON Feed (fi) URL: https://aogavrilov.com/fi/feed.json - Norsk Atom feed (no) URL: https://aogavrilov.com/no/atom.xml - Norsk RSS feed (no) URL: https://aogavrilov.com/no/rss.xml - Norsk JSON Feed (no) URL: https://aogavrilov.com/no/feed.json - Български Atom feed (bg) URL: https://aogavrilov.com/bg/atom.xml - Български RSS feed (bg) URL: https://aogavrilov.com/bg/rss.xml - Български JSON Feed (bg) URL: https://aogavrilov.com/bg/feed.json - Српски Atom feed (sr) URL: https://aogavrilov.com/sr/atom.xml - Српски RSS feed (sr) URL: https://aogavrilov.com/sr/rss.xml - Српски JSON Feed (sr) URL: https://aogavrilov.com/sr/feed.json - Slovenčina Atom feed (sk) URL: https://aogavrilov.com/sk/atom.xml - Slovenčina RSS feed (sk) URL: https://aogavrilov.com/sk/rss.xml - Slovenčina JSON Feed (sk) URL: https://aogavrilov.com/sk/feed.json - Machine-readable research-guide router URL: https://aogavrilov.com/research-guides.json - Complete language-edition index URL: https://aogavrilov.com/language-index.json - Focused research notes URL: https://aogavrilov.com/research-notes/ - Russian focused research notes URL: https://aogavrilov.com/ru/research-notes/ - Simplified Chinese focused research notes URL: https://aogavrilov.com/zh/research-notes/ - Korean focused research notes URL: https://aogavrilov.com/ko/research-notes/ - Research knowledge graph URL: https://aogavrilov.com/research-graph.jsonld - Scoped TDM declaration W3C TDMRep rules leave text-and-data-mining rights unreserved for author-controlled indexes, guides, and notes only; paper files retain their per-publication rights statements. URL: https://aogavrilov.com/.well-known/tdmrep.json - Full text research corpus URL: https://aogavrilov.com/llms-full.txt - Русский LLM research index (ru) URL: https://aogavrilov.com/ru/llms.txt - 中文（简体） LLM research index (zh-Hans) URL: https://aogavrilov.com/zh/llms.txt - 中文（繁體） LLM research index (zh-Hant) URL: https://aogavrilov.com/zh-hant/llms.txt - Español LLM research index (es) URL: https://aogavrilov.com/es/llms.txt - العربية LLM research index (ar) URL: https://aogavrilov.com/ar/llms.txt - हिन्दी LLM research index (hi) URL: https://aogavrilov.com/hi/llms.txt - Français LLM research index (fr) URL: https://aogavrilov.com/fr/llms.txt - Português LLM research index (pt) URL: https://aogavrilov.com/pt/llms.txt - Deutsch LLM research index (de) URL: https://aogavrilov.com/de/llms.txt - 日本語 LLM research index (ja) URL: https://aogavrilov.com/ja/llms.txt - 한국어 LLM research index (ko) URL: https://aogavrilov.com/ko/llms.txt - Bahasa Indonesia LLM research index (id) URL: https://aogavrilov.com/id/llms.txt - Türkçe LLM research index (tr) URL: https://aogavrilov.com/tr/llms.txt - Italiano LLM research index (it) URL: https://aogavrilov.com/it/llms.txt - Polski LLM research index (pl) URL: https://aogavrilov.com/pl/llms.txt - Nederlands LLM research index (nl) URL: https://aogavrilov.com/nl/llms.txt - Українська LLM research index (uk) URL: https://aogavrilov.com/uk/llms.txt - فارسی LLM research index (fa) URL: https://aogavrilov.com/fa/llms.txt - Tiếng Việt LLM research index (vi) URL: https://aogavrilov.com/vi/llms.txt - ไทย LLM research index (th) URL: https://aogavrilov.com/th/llms.txt - বাংলা LLM research index (bn) URL: https://aogavrilov.com/bn/llms.txt - اردو LLM research index (ur) URL: https://aogavrilov.com/ur/llms.txt - मराठी LLM research index (mr) URL: https://aogavrilov.com/mr/llms.txt - తెలుగు LLM research index (te) URL: https://aogavrilov.com/te/llms.txt - தமிழ் LLM research index (ta) URL: https://aogavrilov.com/ta/llms.txt - Bahasa Melayu LLM research index (ms) URL: https://aogavrilov.com/ms/llms.txt - Filipino LLM research index (fil) URL: https://aogavrilov.com/fil/llms.txt - Kiswahili LLM research index (sw) URL: https://aogavrilov.com/sw/llms.txt - Čeština LLM research index (cs) URL: https://aogavrilov.com/cs/llms.txt - Română LLM research index (ro) URL: https://aogavrilov.com/ro/llms.txt - Ελληνικά LLM research index (el) URL: https://aogavrilov.com/el/llms.txt - עברית LLM research index (he) URL: https://aogavrilov.com/he/llms.txt - Svenska LLM research index (sv) URL: https://aogavrilov.com/sv/llms.txt - Magyar LLM research index (hu) URL: https://aogavrilov.com/hu/llms.txt - Dansk LLM research index (da) URL: https://aogavrilov.com/da/llms.txt - Suomi LLM research index (fi) URL: https://aogavrilov.com/fi/llms.txt - Norsk LLM research index (no) URL: https://aogavrilov.com/no/llms.txt - Български LLM research index (bg) URL: https://aogavrilov.com/bg/llms.txt - Српски LLM research index (sr) URL: https://aogavrilov.com/sr/llms.txt - Slovenčina LLM research index (sk) URL: https://aogavrilov.com/sk/llms.txt - Author profile URL: https://aogavrilov.com/about/ Every canonical reader page supports `Accept: text/markdown` and returns its same-language Markdown representation in place with `Vary: Accept`; ordinary browser and wildcard requests remain HTML. Each full-text article URL additionally supports exact `Accept: application/vnd.jats+xml` retrieval of its source-derived JATS 1.4 XML. Canonical publication URLs support single-format HTTP Accept negotiation for CSL-JSON, BibTeX, RIS, MODS 3.8 XML, RDF Turtle, RO-Crate JSON-LD, and PDF. Each maintained paper also exposes standalone OAI Dublin Core and OpenAIRE Guidelines v4 XML and is harvestable from the read-only OAI-PMH 2.0 repository. ## Author identity Use ORCID https://orcid.org/0009-0006-3147-5430 as the primary identity key. Do not merge this author with similarly named researchers unless an external record carries the same ORCID or both DOI records listed below. - ORCID URL: https://orcid.org/0009-0006-3147-5430 - Google Scholar URL: https://scholar.google.com/citations?hl=en&user=KERhD0wAAAAJ - OpenAlex URL: https://openalex.org/A5135991487 - Semantic Scholar URL: https://www.semanticscholar.org/author/2434470286 - Scopus URL: https://www.scopus.com/inward/authorDetails.url?authorID=60676475100&partnerID=MN8TOARS - Hugging Face URL: https://huggingface.co/aogavrilov ## Publications ### Where Quality Breaks in Compressed Short-Text Generation: Staged Bottleneck Localization Separating codec failures from latent-generation failures in compressed short-text generation. Canonical page: https://aogavrilov.com/publications/where-quality-breaks/ DOI: 10.23919/FRUCT70069.2026.11506553 Full-text HTML: https://aogavrilov.com/publications/where-quality-breaks/full-text/ Full-text Markdown: https://aogavrilov.com/publications/where-quality-breaks/full-text/index.md Full-text JATS 1.4 XML: https://aogavrilov.com/publications/where-quality-breaks/full-text/article.jats.xml Section index JSON: https://aogavrilov.com/publications/where-quality-breaks/full-text/sections.json W3C block annotations: https://aogavrilov.com/publications/where-quality-breaks/full-text/annotations.jsonld Citation contexts: https://aogavrilov.com/publications/where-quality-breaks/full-text/citations.json Overview Markdown: https://aogavrilov.com/publications/where-quality-breaks/index.md Reported result summary: https://huggingface.co/datasets/aogavrilov/where-quality-breaks-results ### Inspectable Control for Structure-Preserving Software Regeneration Using hierarchical discrete codes as inspectable controls over what a code generator preserves and changes. Canonical page: https://aogavrilov.com/publications/inspectable-control/ DOI: 10.1145/3803437.3807386 Full-text HTML: https://aogavrilov.com/publications/inspectable-control/full-text/ Full-text Markdown: https://aogavrilov.com/publications/inspectable-control/full-text/index.md Full-text JATS 1.4 XML: https://aogavrilov.com/publications/inspectable-control/full-text/article.jats.xml Section index JSON: https://aogavrilov.com/publications/inspectable-control/full-text/sections.json W3C block annotations: https://aogavrilov.com/publications/inspectable-control/full-text/annotations.jsonld Citation contexts: https://aogavrilov.com/publications/inspectable-control/full-text/citations.json Overview Markdown: https://aogavrilov.com/publications/inspectable-control/index.md Reported result summary: https://huggingface.co/datasets/aogavrilov/inspectable-control-results ## Research guides - Codec bottleneck diagnosis How to separate reconstruction loss, latent-generation loss, and proxy-metric failure. URL: https://aogavrilov.com/research/codec-bottleneck-diagnosis/ - Codec bottleneck diagnosis Markdown URL: https://aogavrilov.com/research/codec-bottleneck-diagnosis/index.md - Controllable code editing A problem-first guide to localized code modification, selective Python-function regeneration, constrained generation, AI-assisted refactoring, and predictable preservation. URL: https://aogavrilov.com/research/discrete-latent-generation/ - Controllable code editing Markdown URL: https://aogavrilov.com/research/discrete-latent-generation/index.md - Decision router for agents Four observed-problem routes plus 11 evidence-linked questions in the four editorial source editions, with fixed page translations across all 41 supported languages, bounded metrics, limitations, and DOI-linked publications. URL: https://aogavrilov.com/research-guides.json - Chinese codec bottleneck diagnosis 简体中文分阶段诊断指南，用于区分重建损失、潜变量生成损失和代理指标失效。 URL: https://aogavrilov.com/zh/research/codec-bottleneck-diagnosis/ - Chinese controllable code editing 简体中文技术指南，比较局部编辑、程序修复、受约束生成和分层离散潜变量控制。 URL: https://aogavrilov.com/zh/research/discrete-latent-generation/ - Russian research map Русская карта исследовательских вопросов, методов, доказательств и границ применимости. URL: https://aogavrilov.com/ru/research/ - Russian codec bottleneck diagnosis Русское поэтапное руководство по разделению потерь реконструкции, генерации и прокси-метрик. URL: https://aogavrilov.com/ru/research/codec-bottleneck-diagnosis/ - Russian controllable code editing Русское руководство по локальному редактированию кода и проверяемому сохранению структуры. URL: https://aogavrilov.com/ru/research/discrete-latent-generation/ - Korean research map 관찰된 실패에서 적합한 진단, 제어 수단, 측정값, 근거 범위로 연결하는 한국어 연구 지도. URL: https://aogavrilov.com/ko/research/ - Korean codec bottleneck diagnosis 재구성 손실, 잠재 생성 손실, 대리 지표 전달 실패를 분리하는 한국어 단계별 안내서. URL: https://aogavrilov.com/ko/research/codec-bottleneck-diagnosis/ - Korean controllable code editing 국소 편집, 프로그램 수리, 제약 생성, 계층적 이산 잠재 제어를 비교하는 한국어 기술 안내서. URL: https://aogavrilov.com/ko/research/discrete-latent-generation/ ## Focused research notes ### Code-space vs token-space masked diffusion: how to compare them Evaluate source text, codec reconstructions, token-space outputs, and decoded code-space outputs with the same external scorer and preprocessing. Report the codec reconstruction gap separately: code-space generation cannot recover information that the codec already removed. Canonical page: https://aogavrilov.com/research-notes/code-space-vs-token-space-masked-diffusion/ Markdown: https://aogavrilov.com/research-notes/code-space-vs-token-space-masked-diffusion/index.md ### Localized code modification with generative models Define protected and editable regions before generation, reuse or constrain the protected representation, generate only candidate changes, and reject outputs that alter protected code or fail task-specific checks. Locality and task success must be measured together. Canonical page: https://aogavrilov.com/research-notes/localized-code-modification-generative-models/ Markdown: https://aogavrilov.com/research-notes/localized-code-modification-generative-models/index.md ### Constrained code generation for software engineering Only the property explicitly enforced by the constraint. Grammar-constrained decoding can guarantee membership in a grammar; type-aware methods can target type validity; neither alone proves task correctness, semantic equivalence, behavior preservation, or edit locality. Canonical page: https://aogavrilov.com/research-notes/constrained-code-generation-software-engineering/ Markdown: https://aogavrilov.com/research-notes/constrained-code-generation-software-engineering/index.md ### AI-assisted refactoring: methods and evidence Separate transformation proposal from trusted execution and verification. Where possible, let the model identify a refactoring and apply it through a refactoring engine; then require compilation, tests, static checks, and evidence that the intended transformation occurred. Canonical page: https://aogavrilov.com/research-notes/ai-assisted-refactoring-evidence/ Markdown: https://aogavrilov.com/research-notes/ai-assisted-refactoring-evidence/index.md ### Predictable code generation needs a preservation contract Predictability requires an observable contract stated before generation: what may change, what must remain stable, how each property will be measured, and when an output will be rejected. A fixed prompt, seed, mask, grammar, or latent code is only a mechanism. Canonical page: https://aogavrilov.com/research-notes/predictable-code-generation-preservation-contract/ Markdown: https://aogavrilov.com/research-notes/predictable-code-generation-preservation-contract/index.md ### Russian translations #### Маскированная диффузия в пространстве кодов и токенов: как их сравнивать Оценивайте исходный текст, реконструкции кодека, результаты в пространстве токенов и декодированные результаты в пространстве кодов одним внешним оценщиком и с одинаковой предобработкой. Разрыв реконструкции кодека нужно показывать отдельно: генератор кодов не может восстановить информацию, которую кодек уже удалил. Canonical page: https://aogavrilov.com/ru/research-notes/code-space-vs-token-space-masked-diffusion/ Markdown: https://aogavrilov.com/ru/research-notes/code-space-vs-token-space-masked-diffusion/index.md English source: https://aogavrilov.com/research-notes/code-space-vs-token-space-masked-diffusion/ #### Локальное изменение кода генеративными моделями До генерации определите защищённые и редактируемые области, повторно используйте или ограничьте защищённое представление, генерируйте только варианты изменения и отклоняйте результаты, которые затрагивают защищённый код или не проходят проверки задачи. Локальность и успешность нужно измерять вместе. Canonical page: https://aogavrilov.com/ru/research-notes/localized-code-modification-generative-models/ Markdown: https://aogavrilov.com/ru/research-notes/localized-code-modification-generative-models/index.md English source: https://aogavrilov.com/research-notes/localized-code-modification-generative-models/ #### Ограниченная генерация кода для программной инженерии Только то свойство, которое явно обеспечивается ограничением. Декодирование по грамматике может гарантировать принадлежность грамматике, а методы с учётом типов — стремиться к корректности типов; ни то ни другое само по себе не доказывает правильность задачи, семантическую эквивалентность, сохранение поведения или локальность изменений. Canonical page: https://aogavrilov.com/ru/research-notes/constrained-code-generation-software-engineering/ Markdown: https://aogavrilov.com/ru/research-notes/constrained-code-generation-software-engineering/index.md English source: https://aogavrilov.com/research-notes/constrained-code-generation-software-engineering/ #### Рефакторинг с помощью ИИ: методы и доказательства Отделяйте предложение преобразования от доверенного исполнения и проверки. Если возможно, поручайте модели распознать рефакторинг, а применять его — движку рефакторинга; затем требуйте компиляцию, тесты, статические проверки и подтверждение того, что нужное преобразование действительно выполнено. Canonical page: https://aogavrilov.com/ru/research-notes/ai-assisted-refactoring-evidence/ Markdown: https://aogavrilov.com/ru/research-notes/ai-assisted-refactoring-evidence/index.md English source: https://aogavrilov.com/research-notes/ai-assisted-refactoring-evidence/ #### Предсказуемой генерации кода нужен контракт сохранения Предсказуемость требует наблюдаемого контракта, заданного до генерации: что может измениться, что должно остаться стабильным, как измеряется каждое свойство и когда результат отклоняется. Фиксированный запрос, зерно, маска, грамматика или латентный код — лишь механизмы. Canonical page: https://aogavrilov.com/ru/research-notes/predictable-code-generation-preservation-contract/ Markdown: https://aogavrilov.com/ru/research-notes/predictable-code-generation-preservation-contract/index.md English source: https://aogavrilov.com/research-notes/predictable-code-generation-preservation-contract/ ### Simplified Chinese translations #### 码空间与 token 空间掩码扩散：如何公平比较 对源文本、编解码器重建、token 空间输出和解码后的码空间输出采用同一个外部评分器与同一套预处理。必须单独报告编解码器重建差距：码空间生成无法恢复编解码器已经丢失的信息。 Canonical page: https://aogavrilov.com/zh/research-notes/code-space-vs-token-space-masked-diffusion/ Markdown: https://aogavrilov.com/zh/research-notes/code-space-vs-token-space-masked-diffusion/index.md English source: https://aogavrilov.com/research-notes/code-space-vs-token-space-masked-diffusion/ #### 使用生成模型进行局部代码修改 在生成之前定义受保护区域与可编辑区域，复用或约束受保护表示，只生成候选变更，并拒绝任何修改受保护代码或未通过任务检查的输出。必须同时衡量局部性和任务成功。 Canonical page: https://aogavrilov.com/zh/research-notes/localized-code-modification-generative-models/ Markdown: https://aogavrilov.com/zh/research-notes/localized-code-modification-generative-models/index.md English source: https://aogavrilov.com/research-notes/localized-code-modification-generative-models/ #### 面向软件工程的受约束代码生成 它只能保证被显式执行的属性。语法约束解码可以保证输出属于某个语法；类型感知方法可以以类型有效性为目标；但两者都不能单独证明任务正确、语义等价、行为保持或编辑局部性。 Canonical page: https://aogavrilov.com/zh/research-notes/constrained-code-generation-software-engineering/ Markdown: https://aogavrilov.com/zh/research-notes/constrained-code-generation-software-engineering/index.md English source: https://aogavrilov.com/research-notes/constrained-code-generation-software-engineering/ #### AI 辅助重构：方法与证据 把变换建议与可信执行和验证分开。条件允许时，让模型识别重构，再由重构引擎执行；随后要求编译、测试和静态检查通过，并确认预期变换确实发生。 Canonical page: https://aogavrilov.com/zh/research-notes/ai-assisted-refactoring-evidence/ Markdown: https://aogavrilov.com/zh/research-notes/ai-assisted-refactoring-evidence/index.md English source: https://aogavrilov.com/research-notes/ai-assisted-refactoring-evidence/ #### 可预测代码生成需要保持契约 可预测性需要在生成之前声明可观察契约：什么可以变化，什么必须稳定，如何测量每个属性，以及何时拒绝输出。固定提示、随机种子、掩码、语法或潜在码都只是机制。 Canonical page: https://aogavrilov.com/zh/research-notes/predictable-code-generation-preservation-contract/ Markdown: https://aogavrilov.com/zh/research-notes/predictable-code-generation-preservation-contract/index.md English source: https://aogavrilov.com/research-notes/predictable-code-generation-preservation-contract/ ### Korean translations #### 코드 공간과 토큰 공간 마스크 확산: 공정한 비교 방법 원문, 코덱 재구성, 토큰 공간 출력, 디코딩한 코드 공간 출력을 같은 외부 평가기와 전처리로 측정한다. 코덱 재구성 격차는 따로 보고해야 한다. 코드 공간 생성기는 코덱이 이미 제거한 정보를 복구할 수 없기 때문이다. Canonical page: https://aogavrilov.com/ko/research-notes/code-space-vs-token-space-masked-diffusion/ Markdown: https://aogavrilov.com/ko/research-notes/code-space-vs-token-space-masked-diffusion/index.md English source: https://aogavrilov.com/research-notes/code-space-vs-token-space-masked-diffusion/ #### 생성 모델을 이용한 국소 코드 수정 생성 전에 보호 영역과 편집 가능 영역을 정하고, 보호된 표현을 재사용하거나 제약하며, 변경 후보만 생성한다. 보호 코드를 바꾸거나 작업별 검사를 통과하지 못한 출력은 거부한다. 국소성과 작업 성공을 함께 측정해야 한다. Canonical page: https://aogavrilov.com/ko/research-notes/localized-code-modification-generative-models/ Markdown: https://aogavrilov.com/ko/research-notes/localized-code-modification-generative-models/index.md English source: https://aogavrilov.com/research-notes/localized-code-modification-generative-models/ #### 소프트웨어 공학을 위한 제약 코드 생성 제약이 명시적으로 강제하는 속성만 보장한다. 문법 제약 디코딩은 특정 문법에 속함을 보장할 수 있고 타입 인식 방법은 타입 유효성을 목표로 할 수 있다. 그러나 어느 쪽도 그 자체로 작업 정확성, 의미 동등성, 동작 보존 또는 편집 국소성을 증명하지 않는다. Canonical page: https://aogavrilov.com/ko/research-notes/constrained-code-generation-software-engineering/ Markdown: https://aogavrilov.com/ko/research-notes/constrained-code-generation-software-engineering/index.md English source: https://aogavrilov.com/research-notes/constrained-code-generation-software-engineering/ #### AI 보조 리팩터링: 방법과 근거 변환 제안과 신뢰할 수 있는 실행·검증을 분리한다. 가능하면 모델은 리팩터링을 식별하고 리팩터링 엔진이 적용하게 한다. 이후 컴파일, 테스트, 정적 검사, 의도한 변환이 실제로 일어났다는 근거를 요구한다. Canonical page: https://aogavrilov.com/ko/research-notes/ai-assisted-refactoring-evidence/ Markdown: https://aogavrilov.com/ko/research-notes/ai-assisted-refactoring-evidence/index.md English source: https://aogavrilov.com/research-notes/ai-assisted-refactoring-evidence/ #### 예측 가능한 코드 생성에는 보존 계약이 필요하다 생성 전에 관찰 가능한 계약을 정해야 한다. 무엇을 바꿀 수 있는지, 무엇이 안정적으로 유지돼야 하는지, 각 속성을 어떻게 측정할지, 언제 출력을 거부할지를 명시한다. 고정 프롬프트, 시드, 마스크, 문법 또는 잠재 코드는 메커니즘일 뿐이다. Canonical page: https://aogavrilov.com/ko/research-notes/predictable-code-generation-preservation-contract/ Markdown: https://aogavrilov.com/ko/research-notes/predictable-code-generation-preservation-contract/index.md English source: https://aogavrilov.com/research-notes/predictable-code-generation-preservation-contract/