https://nlqdb.com/blog/ Engineering notes from building nlqdb — SQL traps, LLM-pipeline debugging, honest comparisons. en-us Wed, 15 Jul 2026 00:00:00 GMT https://nlqdb.com/blog/smoke-test-walks-the-old-ui/ https://nlqdb.com/blog/smoke-test-walks-the-old-ui/ Wed, 15 Jul 2026 00:00:00 GMT Acceptance walkers that pin literal UI copy catch regressions — until a redesign turns them into a 0/9 that mixes real breakage with pure test-drift. The triage cost is the trap, not the literals. https://nlqdb.com/blog/one-shot-recovery-permanent-outage/ https://nlqdb.com/blog/one-shot-recovery-permanent-outage/ Mon, 13 Jul 2026 00:00:00 GMT A best-effort repair that runs once turns one silent skip into a permanent outage. Fix the root, keep it idempotent, and re-trigger from the steady-state symptom — the event never recurs. https://nlqdb.com/blog/green-checkmark-has-a-half-life/ https://nlqdb.com/blog/green-checkmark-has-a-half-life/ Sun, 12 Jul 2026 00:00:00 GMT When an expensive test suite can't run on every push, passing stops being a state and becomes an event. Score each suite pass × freshness with a linear decay so the number rots until it re-runs. https://nlqdb.com/blog/ephemeral-staging-persistent-registry/ https://nlqdb.com/blog/ephemeral-staging-persistent-registry/ Sun, 12 Jul 2026 00:00:00 GMT An environment is only as ephemeral as the most persistent store that references it. Enumerate every store that outlives the rebuild and reset it at spin-up — teardown can't be the invariant. https://nlqdb.com/blog/ownership-transfer-outlives-least-privilege/ https://nlqdb.com/blog/ownership-transfer-outlives-least-privilege/ Sun, 12 Jul 2026 00:00:00 GMT Hardening queries to per-tenant roles and RLS quietly broke our ownership transfer: it retargeted one authorization store out of four. Transfer must move them all — idempotently, in one batch. https://nlqdb.com/blog/most-active-user-is-your-test-suite/ https://nlqdb.com/blog/most-active-user-is-your-test-suite/ Sat, 11 Jul 2026 00:00:00 GMT Pre-launch, synthetic traffic IS your traffic: e2e walkers register users and run real queries, so every dashboard quietly measures your robots. Three places it bit us, three fixes. https://nlqdb.com/blog/five-fallback-models-one-provider/ https://nlqdb.com/blog/five-fallback-models-one-provider/ Sat, 11 Jul 2026 00:00:00 GMT A model-diverse fallback list on one gateway saturates as a unit — five models, one rack. Make the fallback unit a lane (base URL, key, candidates), not a longer list on the same pool. https://nlqdb.com/blog/decided-questions-rot-in-your-decision-log/ https://nlqdb.com/blog/decided-questions-rot-in-your-decision-log/ Sat, 11 Jul 2026 00:00:00 GMT Decision logs rot at the seam between open and answered: a decided-but-unmarked bullet makes readers re-litigate closed calls. Make resolved a greppable state and count unmarked bullets as debt. https://nlqdb.com/blog/emit-metrics-where-the-distinction-is-certain/ https://nlqdb.com/blog/emit-metrics-where-the-distinction-is-certain/ Sat, 11 Jul 2026 00:00:00 GMT A destructive-op retry rate emitted at the HTTP route can go negative — the route can't tell reads from writes. Emit metrics at the lowest layer where the distinction is certain. https://nlqdb.com/blog/rotate-encryption-key-without-a-version-column/ https://nlqdb.com/blog/rotate-encryption-key-without-a-version-column/ Fri, 10 Jul 2026 00:00:00 GMT Rotating a key-encryption key feels like it needs a key_version column. It doesn't — the ciphertext already describes itself, so put the version in the blob prefix and rotate with zero migration. https://nlqdb.com/blog/text-to-sql-planner-told-wrong-dialect/ https://nlqdb.com/blog/text-to-sql-planner-told-wrong-dialect/ Fri, 10 Jul 2026 00:00:00 GMT A text-to-SQL planner emits whatever dialect you name it. Add a second engine and the bug is one hardcoded dialect literal the type never forced you to fix — so ClickHouse gets Postgres SQL. https://nlqdb.com/blog/postgres-validator-rejects-valid-clickhouse-sql/ https://nlqdb.com/blog/postgres-validator-rejects-valid-clickhouse-sql/ Thu, 09 Jul 2026 00:00:00 GMT A Postgres-pinned AST validator false-rejects valid ClickHouse grammar as parse_failed — a silent veto of correct SQL. The fix: split the security allowlist from the dialect parser. https://nlqdb.com/blog/blog-without-a-feed-is-a-dead-end/ https://nlqdb.com/blog/blog-without-a-feed-is-a-dead-end/ Thu, 09 Jul 2026 00:00:00 GMT A blog with no RSS feed is a one-way street: feed readers and dev.to/Medium 'import from RSS' both need a feed URL. Without one, every venue re-post is a manual copy-paste that quietly stops. https://nlqdb.com/blog/agent-memory-benchmarks-measure-recall-not-analysis/ https://nlqdb.com/blog/agent-memory-benchmarks-measure-recall-not-analysis/ Thu, 09 Jul 2026 00:00:00 GMT Agent-memory benchmarks score end-to-end recall of facts on mostly self-reported numbers. Almost nobody measures analysis over memory — the gap we found reading the papers. https://nlqdb.com/blog/one-way-internal-links-leak-yield/ https://nlqdb.com/blog/one-way-internal-links-leak-yield/ Wed, 08 Jul 2026 00:00:00 GMT The page count climbed; referrals stayed flat at ~1/week. Our internal link graph was a tree, not a mesh. The fix was one reciprocal link, derived from a field we already had. https://nlqdb.com/blog/serverless-db-cold-start-retry/ https://nlqdb.com/blog/serverless-db-cold-start-retry/ Wed, 08 Jul 2026 00:00:00 GMT A scale-to-zero Postgres branch fails the first query while its compute wakes. Instant retries replay the cold connection. The fix: back off the DB stage, not the LLM stages. https://nlqdb.com/blog/llm-timeout-looks-like-hallucination/ https://nlqdb.com/blog/llm-timeout-looks-like-hallucination/ Tue, 07 Jul 2026 00:00:00 GMT Our NL→SQL benchmark scored a frontier model as junk on 5 hard questions. It never hallucinated — a 5s prod timeout aborted it mid-answer and the handler mislabeled the abort as a parse failure. https://nlqdb.com/blog/model-preset-fail-loud/ https://nlqdb.com/blog/model-preset-fail-loud/ Tue, 07 Jul 2026 00:00:00 GMT When the premium lane isn't available, the tempting branch is to silently serve the default chain. A placebo knob is worse than no knob. The honest contract: pin, upgrade, or fail loud. https://nlqdb.com/blog/llm-preflight-probe-health/ https://nlqdb.com/blog/llm-preflight-probe-health/ Tue, 07 Jul 2026 00:00:00 GMT Six straight LLM-agent CI runs failed while our pre-flight probe stayed green. Lessons on gating CI on an LLM provider: probe the real shape, read the body not the status, never trust one model. https://nlqdb.com/blog/bird-gold-noise-distinct/ https://nlqdb.com/blog/bird-gold-noise-distinct/ Mon, 06 Jul 2026 00:00:00 GMT We bucketed 238 BIRD-dev losses with a structural differ: 46 differ from gold only by a DISTINCT the model rightly added. Audit gold quality before writing prompt directives, or you overfit to noise. https://nlqdb.com/blog/llm-concatenates-columns-text-to-sql/ https://nlqdb.com/blog/llm-concatenates-columns-text-to-sql/ Mon, 06 Jul 2026 00:00:00 GMT Gold SQL returns first_name, last_name as two columns; the model returns one concatenated full name. Positional-tuple EX scoring can never match them, so a semantically right answer scores as a miss. https://nlqdb.com/blog/http-200-error-in-body/ https://nlqdb.com/blog/http-200-error-in-body/ Sun, 05 Jul 2026 00:00:00 GMT A gateway commits 200 OK before the upstream model fails, so the error rides in the 200 body. A res.ok-only client counts it as a wrong answer, not an outage. res.ok is necessary, not sufficient. https://nlqdb.com/blog/top-n-rows-per-group/ https://nlqdb.com/blog/top-n-rows-per-group/ Sat, 04 Jul 2026 00:00:00 GMT "Top 3 per category" reads like ORDER BY … LIMIT 3, but LIMIT caps the whole result set, not each group. The fix is ROW_NUMBER() OVER (PARTITION BY …) — and the hidden decision is how ties break. https://nlqdb.com/blog/your-bi-tool-got-acquired-data-layer/ https://nlqdb.com/blog/your-bi-tool-got-acquired-data-layer/ Sat, 04 Jul 2026 00:00:00 GMT BI notebooks get rolled up — Mode → ThoughtSpot, Looker → Google, Periscope → Sisense. Fine when it's a destination humans log into; a liability when your product's runtime calls its API. https://nlqdb.com/blog/find-duplicate-rows-you-re-google-every-time/ https://nlqdb.com/blog/find-duplicate-rows-you-re-google-every-time/ Fri, 03 Jul 2026 00:00:00 GMT Find duplicates hasn't changed in thirty years: GROUP BY the suspect columns, HAVING COUNT(*) > 1. Wanting the whole row, not just the key, quietly changes it to a window function. https://nlqdb.com/blog/text-to-sql-build-vs-buy/ https://nlqdb.com/blog/text-to-sql-build-vs-buy/ Fri, 03 Jul 2026 00:00:00 GMT Prompt + model + run the SQL is 10% of an 'ask your data' feature. The fail-closed validator, plan cache, and eval harness are the rest — yours forever. The real question: do you want that stack? https://nlqdb.com/blog/sitemap-advertising-redirects/ https://nlqdb.com/blog/sitemap-advertising-redirects/ Fri, 03 Jul 2026 00:00:00 GMT A static host that serves route/index.html 307-redirects the bare path. Our sitemap advertised 27 redirecting URLs and every canonical tag pointed at one. The fix is one path-normalize helper. https://nlqdb.com/blog/offline-llm-eval-rate-limits/ https://nlqdb.com/blog/offline-llm-eval-rate-limits/ Fri, 03 Jul 2026 00:00:00 GMT A free-model NL-to-SQL bench scored 17/20, then 6/20 ninety seconds later. The model didn't change — the providers got tired. How to keep availability out of your accuracy number. https://nlqdb.com/blog/ai-internal-tool-builder-faster/ https://nlqdb.com/blog/ai-internal-tool-builder-faster/ Fri, 03 Jul 2026 00:00:00 GMT Low-code AI scaffolds the admin tool in a prompt. But the output is still a destination a human operates — and often the answer belongs inline in your product, or the asker is an agent. https://nlqdb.com/blog/text-to-sql-accuracy-schemas-your-users-never-build/ https://nlqdb.com/blog/text-to-sql-accuracy-schemas-your-users-never-build/ Thu, 02 Jul 2026 00:00:00 GMT BIRD and Spider score NL-to-SQL over messy academic schemas. The same free-model chain that scores 0.52 on BIRD scores 0.96 on the schema shapes our users actually build — so we report both. https://nlqdb.com/blog/mcp-server-what-does-the-agent-own/ https://nlqdb.com/blog/mcp-server-what-does-the-agent-own/ Thu, 02 Jul 2026 00:00:00 GMT Two shapes of MCP server look identical in a feature matrix: a window into a human's app, or infrastructure the agent owns. The tell is what the agent owns after the call returns. https://nlqdb.com/blog/agent-memory-vector-store-aggregation-gap/ https://nlqdb.com/blog/agent-memory-vector-store-aggregation-gap/ Thu, 02 Jul 2026 00:00:00 GMT A vector store returns the top-k most similar memories — there is no GROUP BY, COUNT, or JOIN. Recall is similarity; reporting is aggregation. Agent memory needs both machines, not one. https://nlqdb.com/blog/store-form-submissions-without-a-backend/ https://nlqdb.com/blog/store-form-submissions-without-a-backend/ Thu, 02 Jul 2026 00:00:00 GMT Storing a signup is a trivial insert — no server needed. The part that wants a database is the reporting: "signups per day," "which referrer converted" — aggregations that want a query planner. https://nlqdb.com/blog/not-in-subquery-null-trap/ https://nlqdb.com/blog/not-in-subquery-null-trap/ Wed, 01 Jul 2026 00:00:00 GMT Why WHERE id NOT IN (SELECT …) silently returns nothing when the subquery contains a NULL, and the two anti-join shapes (NOT EXISTS, LEFT JOIN … IS NULL) that never lie to you. https://nlqdb.com/blog/zep-recall-vs-analytical-agent-memory/ https://nlqdb.com/blog/zep-recall-vs-analytical-agent-memory/ Wed, 01 Jul 2026 00:00:00 GMT A temporal knowledge graph is genuinely good at recall — and has no query planner. When the question about agent memory is a GROUP BY, retrieval and aggregation are different machines. https://nlqdb.com/blog/null-timestamp-ttl-sweep-funnel-metric/ https://nlqdb.com/blog/null-timestamp-ttl-sweep-funnel-metric/ Wed, 01 Jul 2026 00:00:00 GMT A backfill is not a default: one nullable timestamp column made an age-based eviction a silent no-op and pinned a funnel metric at zero — the same NULL, two different failure modes.