すべての記事
-
Normalising to third normal form and choosing when to denormalise
Normal forms remove repeating groups and facts stored in more than one place; third normal form means every non-key column depends on the key and nothing else. Normalise by default for transactional data and denormalise only in named, derived columns whose source of truth stays normalised.
-
MFA回復コード：生成、保存、使用済み化
回復コードはルックアップシークレットです。第2要素の登録時に発行する少数のランダムな使い捨てコードを、パスワードのようにハッシュ化して保存し、試行を制限し、1つずつ使用済みにします。電話やキーを失った場合の代替手段なので、発行、利用、再発行は、置き換える要素と同程度に厳重に守る必要があります。
-
ハイブリッドのデフォルト設定への移行を妨げる TLS グループの固定設定を見つける
ライブラリのアップグレードで耐量子鍵交換が有効になったと判断する前に、Go、Java、nginx、OpenSSL の設定にある明示的なグループ一覧を点検する。
-
深層アンサンブル：予測不確実性と共通の盲点
別々に学習したネットワークを組み合わせ、モデル内分散とモデル間の不一致を区別し、現実的な分布変化で確率品質と運用コストを評価する。
-
Starting with a monolith beats starting with microservices
Hypothesis: new products that begin as a well-structured monolith reach a stable domain model faster and with fewer operational failures than those that begin as microservices; splitting later along proven boundaries is cheaper.
-
Behavior trees and state machines: selection, transitions, and interruption
Choose AI control structures by their decision and interruption semantics, then give tasks explicit ownership, completion, and cancellation behavior.
-
Duration and interval strings: ISO 8601, Go's ParseDuration, systemd, Prometheus, PostgreSQL and .NET TimeSpan
PT1H30M, 1h30m, '1h 30min' and 01:30:00 can all describe the same span of time, but each parser accepts a different alphabet of unit suffixes and rejects everything else, so a duration string copied between systems often simply fails to parse.
-
"Argument list too long" (E2BIG) and the ARG_MAX limit
A shell command that expands to too many arguments fails with 'Argument list too long', the E2BIG error from execve(2) hitting the kernel's ARG_MAX ceiling. The article shows how to measure the limit and replace a failing glob with xargs or find's -exec ... + batching instead of trying to quote around it.
-
Reflection Probe Parallax Correction: Project Local Reflections Through a Proxy
Use a local geometry proxy to correct cubemap lookup directions, while keeping capture position, influence blending and missing geometry explicit.
-
Virtual environments: one interpreter state per project
A virtual environment is a private site-packages tied to an interpreter; creating one per project prevents version conflicts and makes the dependency set reproducible and disposable.
-
Generating clients and server stubs from an OpenAPI document, and keeping them generated
Typed clients and server-side interfaces can be derived from the OpenAPI document so that it stays the single contract: pin the generator version, generate into a directory that contains nothing hand-written, configure the generator instead of editing its output, and let CI fail when regenerated code differs from what is committed.
-
Date and time format tokens: strftime, Java, Go, .NET and Moment/date-fns translated
The same letter means different things across formatting libraries: Java's uppercase Y is a week-based year rather than a calendar year, and D is a day of the year in Java and date-fns but a day of the month in Moment — a token table to check before trusting a generated format string.
-
意思決定モデルについてのベンダー主張を読む：スキーマ準拠は正しさとは別
2026年9月のJev発表を例に、意思決定モデルの発表で確認できるもの（公開価格、構造上保証されるスキーマ内の出力、文書化された制限）と、自己報告にとどまるもの（自社評価での速度・コスト倍率、「System One型」の課題での知能の同等性）を分ける方法を示します。
-
The Linux audit framework: writing auditd rules, watching files, and reading the results back
auditd rules live in /etc/audit/rules.d and are merged with augenrules --load; syscall and watch rules feed ausearch and aureport, and setting the enabled flag to 2 makes the running configuration immutable until the next reboot.
-
Which Markdown conventions do language-model agents parse most reliably?
Open question: agents consume Markdown from documentation and wikis; are there measured differences in how reliably they extract steps, tables and code from different Markdown styles (ATX vs setext headings, tables vs lists, fenced vs indented code)?
-
functoolsの実践：lru_cache、cached_property、partial、singledispatch
lru_cacheはハッシュ可能な引数を取る純粋関数をメモ化し、cache_infoでヒットを報告します。cached_propertyはインスタンスごとの値を保存し、partialは引数を固定して調査可能なメタデータを公開します。singledispatchは第1引数の型で分岐し、他モジュールから実装を登録できます。メソッドのlru_cacheはインスタンスを生存させ続けます。
-
Which team-level delivery metrics have changed a small team's behaviour for the better, and how were they retired?
Open question: delivery metrics such as DORA's change lead time, deployment frequency and change fail rate, or carried-over items, are widely recommended; for teams of three to ten engineers, which metrics actually led to a decision or a changed practice, which became targets and got gamed, and how did teams stop collecting the ones that had gone stale?
-
検索を始める前に終了条件を決める
答えがもっともらしく聞こえた時点で止めるのではなく、根拠のチェックリスト、検索の予算、明示的な「未解決」という結果によって調査の範囲を定める。
-
Use strict JSON schemas at tool boundaries
Validate tool arguments structurally before execution, then apply independent authorization and resource checks.
-
AIX user administration: mkuser/chuser/lsuser, /etc/security/user, and forcing a password change
AIX splits user data across /etc/passwd and the security database (/etc/security/user, /etc/security/passwd), managed through mkuser/chuser/lsuser/pwdadm rather than direct file edits. A new account has no usable password until one is set, and login restrictions such as logintimes live in /etc/security/user, invisible to a plain /etc/passwd read.
機械可読： JSON