All articles
-
Tessellation and Displacement: Build Detailed Surfaces Without Cracks
Control patch subdivision, displacement continuity and generated geometry cost while preserving agreement across shared edges.
-
Benchmark methodology: warm up, interleave repetitions, report the spread
A timing comparison only counts as a result once it survives noise: fix the workload, discard warm-up runs, run both variants in alternation many times, choose the metric before looking at the data, and report the spread and the environment next to every number. A difference smaller than the run-to-run spread is not a finding.
-
Let code compute: arithmetic, counting, date logic and unit conversion belong in tools, not in the model
A working rule for agents: whenever a step has an exact answer that a program can produce (sums, counts, date differences, unit conversions, sorting, deduplication), run the program and let the model interpret the result, because a language model recognises the shape of an answer rather than computing it; vendors of decision models document the same limits for their own products.
-
Effect size versus statistical significance: which one decides
Significance says whether the data are unusual under the null model at the sample size used; the effect size says how big the difference is in units that matter. Large systems make trivial effects significant and small pilots make large effects non-significant. Define the smallest effect worth acting on before the experiment and compare the interval to it.
-
Opening an untrusted repository: the files that execute code when you install, build, test or just enter it
Cloning a repository runs nothing, but many routine next steps do: package install scripts, build files, test configuration, editor tasks, environment loaders and some Git settings. An agent working in a foreign repository should know these execution points and run them only inside a sandbox.
-
Game frame pacing: control presentation cadence and queue latency
Connect rendered frames to display refreshes, distinguish presentation modes from queue depth, and verify smooth delivery using presentation timing.
-
Logical clocks: Lamport timestamps, vector clocks and hybrid clocks
Wall clocks on different machines disagree, so ordering events by timestamp loses updates. Lamport timestamps give an order consistent with causality, vector clocks additionally detect concurrent updates, and hybrid logical clocks keep a value close to wall time while preserving causal order.
-
PowerShell's execution policy is not a security boundary — what actually restricts scripts
Get-ExecutionPolicy -List shows per-scope settings that are easy to bypass by design and documented by Microsoft as defense in depth, not a security boundary; App Control for Business (WDAC), AppLocker, and Constrained Language Mode are the controls that actually restrict what code and language features can run.
-
Designing idempotent operations and safe retries
An operation is idempotent if repeating it has the same effect as doing it once; HTTP defines which methods are idempotent, and idempotency keys extend the property to POST so clients can retry without duplicates.
-
Teams with fewer, alert-linked dashboards diagnose incidents faster than teams with many unowned dashboards
Hypothesis: for services of comparable size, the time from a page to a named probable cause is shorter when the team maintains a small number of owned dashboards that alerts link to directly, than when it maintains many copied or auto-generated dashboards that responders have to browse; a proposed comparison using incident timelines and dashboard inventories.
-
Handling time: UTC, ISO 8601 and time zones
Store and exchange instants in UTC using RFC 3339 timestamps, keep time-zone-aware objects in code, convert to local time only for display, and treat calendar arithmetic as a separate problem from instant arithmetic.
-
z/OS system operations basics: console commands, SYSLOG/OPERLOG, and what an IPL is
Console commands like D A,L (list active jobs) and D IPLINFO (display IPL information) are the read-only starting point for understanding what a z/OS system is doing; SYSLOG and OPERLOG are where system messages accumulate, and an IPL (Initial Program Load) is the mainframe's equivalent of a full system boot, not something to trigger outside a planned change window.
-
Java streams versus loops: when a pipeline is clearer and when it is not
A stream is a lazy pipeline of intermediate operations closed by one terminal operation; it reads well for filter, map and collect over a collection, but the package documentation discourages side effects in the lambdas, a stream cannot be reused, checked exceptions do not fit, and a loop is clearer for early exit with state, index-based work and mutation.
-
API documentation with examples that are executed in CI
Keep every request and response example in the API documentation executable: store them as OpenAPI examples validated against the schema, run them against a sandbox in CI and render the documentation from the files that were executed, so an example that stops working fails a build instead of misleading a reader.
-
Reading the AIX error log with errpt and feeding it into syslog
errpt is AIX's own hardware/software error log, started earlier in boot than syslog and populated by the kernel and error-aware daemons, not by application log lines. errclear trims it by class and age; an errnotify ODM object plus logger is the documented way to mirror errpt entries into local or remote syslog.
-
Redirecting a running agent with revisioned instructions and reconciled side effects
Treat a change of direction as a recorded revision with a boundary around in-flight actions. This proposed procedure distinguishes cancellation requests, confirmed effects and resumable work so a late result cannot silently restore an obsolete objective.
-
LiteLLM 1.107.0: checking artifact exposure without restarting compromised Python
The October LiteLLM compromise affected PyPI wheels and container images. Check preserved metadata before starting Python, rebuild affected environments, and revoke credentials that the compromised system could access.
-
BigInt over JSON: define a lossless field contract before adding a replacer
Encode large integers with an explicit schema and restore them without silently passing through Number.
-
Phishing-resistant sign-in with WebAuthn and passkeys
WebAuthn authenticates with a per-site public-key pair: the browser only lets a credential be used by origins under the relying party ID it was registered for, and the authenticator signs a server challenge, so a look-alike site obtains nothing replayable; passkeys are discoverable WebAuthn credentials, often synced across a user's devices.
-
Package managers compared: apt, dnf, zypper, apk, pkg, Homebrew and winget side by side
Install, remove, search, list, find the owning package, pin a version and run non-interactively — the same seven tasks, phrased differently by apt/dpkg, dnf, zypper, apk, pkg, Homebrew and winget. The table also marks the task that Homebrew and winget simply do not offer.
Machine-readable: JSON