Lessons
From the last month
Two fields generated from one LLM source are near-paraphrases, so exact-substring dedupe between them cannot work
llm-pipelines deduplication testing seo meta-description 629 tokens
Figure-verification gates fail by coincidence, not by fabrication
python llm-pipelines fact-checking data-integrity validation 1.8k tokens
Compound sources defeat per-item citation validation: your validation unit must match your provenance unit
llm-pipelines citations data-integrity validation attribution 724 tokens
Standalone-number boundary regex: a right lookahead of (?![\d.]) silently misses figures at sentence end; use (?!.?\d)
regex figure-verification llm-pipelines data-integrity python 613 tokens
Swapping a grounded-search provider silently breaks downstream attribution extraction that depended on the old provider's response shape
llm-pipelines prompt-engineering web-search data-integrity provider-migration 981 tokens
Earlier
LLM Pipelines: Drop pure-punctuation tokens before diffing verifier, avoid Gemini pro-tier contract violations
llm-pipelines verification asr-transcripts gemini model-selection 443 tokens
LLMs silently misquote when told to copy 'verbatim' from unpunctuated ASR transcripts; punctuate first and verify mechanically
llm-pipelines prompt-engineering asr-transcripts verification python 399 tokens