Evaluating LLM hallucinations in 30 seconds Silent video: no voiceover, music or sound effects. This transcript repeats every on-screen word and describes each scene. The film includes the topic label "LLM hallucination evaluation", the note "Illustrative diagram", the TecnoStack logo and progress numbers 1, 2, 3, 4, 5, 6. Elements enter and leave during transitions; in portrait, the closing sign-off replaces the footer. [0:00–0:05] Beat 1 of 6 (intro) Kicker: LLM hallucination evaluation Headline: Is the answer supported? Support: Test the complete application Diagram (flow): Real question; Retrieved passages; Answer and citations Connections: Real question → Retrieved passages; Retrieved passages → Answer and citations Highlighted: Answer and citations Scene: Flow: Real question, Retrieved passages, Answer and citations. Connections: Real question to Retrieved passages; Retrieved passages to Answer and citations. Emphasis: Answer and citations. [0:05–0:10] Beat 2 of 6 (step1) Kicker: 01 Headline: Use real questions Support: Include unanswerable cases Diagram (hub and spokes): Test set; Real user questions; Cases to decline; Arabic and English Connections: Test set → Real user questions; Test set → Cases to decline; Test set → Arabic and English Highlighted: Test set Scene: Network: Test set, Real user questions, Cases to decline, Arabic and English. Connections: Test set to Real user questions; Test set to Cases to decline; Test set to Arabic and English. Emphasis: Test set. [0:10–0:15] Beat 3 of 6 (step2) Kicker: 02 Headline: Agree on labeling rules Support: Compare independent reviewers Diagram (flow): Test case; Labeling guide; Two reviewers; Agreed label Connections: Test case → Labeling guide; Labeling guide → Two reviewers; Two reviewers → Agreed label Highlighted: Labeling guide Scene: Flow: Test case, Labeling guide, Two reviewers, Agreed label. Connections: Test case to Labeling guide; Labeling guide to Two reviewers; Two reviewers to Agreed label. Emphasis: Labeling guide. [0:15–0:20] Beat 4 of 6 (step3) Kicker: 03 Headline: Score each stage separately Support: Identify what needs fixing Diagram (evidence checklist): Right passages found; Claims supported; Citations accurate; Correct declines Highlighted: Claims supported Scene: Comparison without scores: Right passages found, Claims supported, Citations accurate, Correct declines. Emphasis: Claims supported. [0:20–0:25] Beat 5 of 6 (step4) Kicker: 04 Headline: Look beyond the average Support: Report case counts per slice Diagram (decision paths): All test results; One average; Results by slice Connections: All test results → One average; All test results → Results by slice Highlighted: Results by slice Scene: Branching diagram: All test results, One average, Results by slice. Connections: All test results to One average; All test results to Results by slice. Emphasis: Results by slice. [0:25–0:30] Beat 6 of 6 (close) Kicker: TECNOSTACK Headline: Turn failures into test cases Support: Read the full TecnoStack guide Diagram (timeline): Test set; New failure; Add the case; Re-run after changes Connections: Test set → New failure; New failure → Add the case; Add the case → Re-run after changes Highlighted: Add the case Sign-off: TecnoStack logo and tecnostack.com Scene: Sequence: Test set, New failure, Add the case, Re-run after changes. Connections: Test set to New failure; New failure to Add the case; Add the case to Re-run after changes. Emphasis: Add the case.