Posts tagged Architecture
Ordering security patches by entry point, privilege and evidence
Node.js, PHP2Shell and the kernel needed patches in October 2026. I weigh entry points, privilege and evidence to set the order, and I look past the CVSS Base score.
Config fields that quietly turn protection off: four fixes from gortexa
Empty values, "*" and custom lists can silently disable security checks. Four fixes from gortexa PR #57 are reproduced in Go.
A wallet opened between reconciliation cuts gets flagged: zero-fill or skip the missing snapshot?
When a new wallet has no snapshot, NULL can cause a reconciliation mismatch. This PostgreSQL example sets out the assumptions behind zero-filling.
The 900-line PR Anthropic turned down: what a performance gain costs to maintain
A 900-line PR saved 2 ms per message but added a build plugin to maintain, so Anthropic said no. How to measure a speed-up and decide whether it stays.
How Qwenputer turns DeltaNet state into RAM: exact reads and writes, and what they do not promise
Qwenputer fixes the gate and write strength at 1, using one-hot keys for exact RAM access. A numpy example separates this from inference reproducibility.
Who tells you that you are affected: disclosure paths in four incidents
Zeabur, Railway, 91APP and Europe's cold-chain attacks disclosed scope in different units through different channels. What a downstream user can do with each.
Reading Binance's initial report: the intrusion path and the EIP-7702 authorisations
Binance's initial report links compromised work computers to EIP-7702 authorisations. The intrusion path and deployment token still need independent review.
What should an exchange publish so outsiders can keep checking assets, liabilities and delivery?
Proof of reserves covers one snapshot. Here is what an exchange should publish so outsiders can keep checking assets, liabilities and withdrawals.
What Background Agent Status Summaries Cost: Attributing Model Calls by query_source
One report puts Claude Code summaries at 16.4% of estimated cost. Grouping api_request events shows why workload and denominators matter.
Why multisig can still fail at one point: how Ronin's 5 of 9 came together
A threshold should count independent controllers. How Ronin's 5 of 9 was assembled through one company, with acceptance checks for multisig designs.
Secret store and database write ordering: what a failure leaves behind, and how a retry resumes
Without a shared transaction, write order affects the state left after failure. SQLite examples cover secret overwrites, pending rows and retries.
Who can press the stop button? Freezes, chain halts and hard forks act at different layers
Binance's 2026 freezes came from several parties at different layers. This post separates exchange pauses, issuer blacklists, validator halts and hard forks.
Once the key is safe, who gets to ask for a signature? Custody lessons from Binance in 2019
Test whether a compromised withdrawal service can obtain an unwanted signature, using a custody model, a Python example and explicit acceptance checks.
Valid signatures, wrong transaction: the Bybit and Safe signing supply chain
Bybit's 2025 loss involved a tampered Safe front end. Independent SafeTx hash checks can detect substituted fields when signing and display disagree.
Estimating AI Prototypes in Two Stages: From Zero to Demo, From Demo to Production
AI makes a demo cheap. Verification, concurrency, payment handling and observability haven't become much cheaper, so I estimate and quote the two stages separately.
Protobuf as the Single Source of Truth: Serving gRPC and HTTP/JSON from One Go Port
One .proto contract generates types and entry points for gRPC, HTTP/JSON and MCP. A Go example serves gRPC and HTTP/JSON from one h2c port and shows where shared timeouts break streams.
Batching ClickHouse Inserts: A Queue and Go Worker, Async Inserts and Deduplicated Retries
I prefer a queue and Go worker to control ClickHouse batching. Async inserts suit clients that cannot batch; retries depend on deduplication and its limits.
Shared Memory for Several AI Tools: SQLite Holds the Truth, Qdrant Only Searches
I use SQLite as the canonical store and Qdrant for retrieval through MCP. Sharing AI memory means deciding who may write and how the stores stay in sync.
After RFC 10017, the First Reason for a BFF Is Keeping Tokens Out of the Browser
RFC 10017 ranks the BFF first for browser OAuth security because tokens stay on the server. Settle token storage before working on aggregation.
Millisecond Cursors Repeat Pages: timestamptz Precision and Composite Keyset Pagination
A millisecond cursor over microsecond timestamptz repeats rows; microseconds still skip or repeat rows written in one transaction. A (created_at, id) keyset fixes both, tested on PostgreSQL 18.
What an HMAC Request Signature Should Cover: Method, Path, Query and the Raw Body
Sign only the body and the path or query can be swapped freely; re-serialise JSON before verifying and honest requests fail. Cover the method, path, sorted query and raw body bytes.
Treat Unset as Zero: Hard Caps, Approval Thresholds and Which Way a Kill Switch Fails
Defaults decide what a money system does when a setting is missing or unreadable: reject out-of-range amounts, treat an unset approval threshold as zero, and fail closed on writes only.
Mini Apps Inside Messaging Apps: WeChat, Telegram and How to Design the Host
Messaging apps can host third-party code because they already have the social graph, payments and sharing. Recent data on WeChat, Telegram and LINE, and how to design the host.
From Messaging to Community and Short Video: Storage, Recommendations, Mini Apps and Accounts
Public posts, video, ranking and mini apps put the server back in charge: storage after MinIO, a Qdrant recommender, mini app identity, and Taiwan and EU rules.
Building a Messaging App from the Device Out: Local-first, E2EE and On-Device Search
Once the device owns the chat history and the server only relays, identity keys, encryption, ordering, new-device history and search all change. Checked against Signal, WhatsApp and LINE.
Only 86 Million Ways to Shuffle: What Game RNG Certification Checks
Game RNG certification checks the seed, state recovery and mapping bias; statistical tests aren't enough. From a 1999 poker shuffle to Lineage M's odds dispute.
Counting the Cost of AI Code Review: Bugs Found, False Alarms and People's Time
AI code review's main cost is human triage time, not tokens. Whether it pays depends on the precision and severity of findings, as curl and an ICSE 2025 study show.
The Same Payment Sent Twice: Idempotency and Retries, by Way of Santander's Christmas
Safe payment retries need an idempotency key both sides agree on: when to create it, what goes in it, and what to do if payloads differ or a refund arrives first.
The Six Things We Didn't Build: What the Framework Teaches
Six things gortex deliberately left unbuilt, from a full graceful-shutdown chain to OpenAPI generation, and the principles the whole series teaches.
Why Build a Kitchen-Sink Framework
Why gortex is a kitchen-sink framework: what a framework layer trades away once the goal shifts from one lean microservice to consistency across services.
Next.js Server Boundaries: What Stays in the Framework, What Leaves
SSR, SSG, ISR, and RSC each solve a different problem. Sort them out first, then decide what stays in Next and what moves to a standalone BFF.
Svelte as a Mirror for the Frontend, Not an Idol
Svelte holds DOM intuition, HTML-first structure and an explicit server boundary together, which makes it a reference for judging front-end tools, not an idol.
The Framework Server Layer Is Not Your BFF
Keep the framework server layer to browser-facing aggregation; service auth, trust boundaries and risk belong in a standalone BFF, whatever Next or Nuxt can run.
Where Does the HTML Come From: SSR, CSR, SSG, ISR on One Path
Stop memorising acronyms. Ask one question instead: when and where was the HTML the user just received actually generated.
Two Ways to Wire a BFF: htmx Returns Partials, SvelteKit Goes JSON
One call chain, two front ends: htmx partials for admin tables where state lives on the server, JSON over SvelteKit when the client needs to hold state.
HTML-first or SPA: Reset the Default, Skip the Holy War
The test is where state lives and whether interaction swaps content or drives local state; default to HTML-first and upgrade to the client only when you must.
Why BFF Still Earns Its Keep in 2025
A BFF is not a fashionable layer. It is where you make plain who faces the browser, who faces the backend, and who carries the risk.
Markup First: Describe the Structure, Then the Behaviour
Rather than starting from framework components, describe the structure in HTML first and let behaviour be a thin layer on top.
Semantics Lost at the Boundary, and Why the Monolith Keeps Pulling
When a large system is hard to compose, the components are usually fine. What broke is the meaning, flattened into bytes as it crossed a service boundary.
One Responsibility per Layer, and the Discipline It Takes
Each layer should own one job you can state plainly; blur it and debugging becomes guesswork. On over-abstraction, under-abstraction, and when not to abstract.