RedFlag

What RedFlag saw in the 1999 Lance Armstrong doping denial

← All articles · By the RedFlag team · September 22, 2026

We ran a well-known clip through RedFlag's analyzer: a 1999 interview, filmed just after Armstrong's first Tour de France win, in which he denies ever using performance-enhancing drugs. RedFlag is not a lie detector — it measures where a speaker's voice and face diverge from their own baseline and leaves the interpretation to you. What makes this clip useful is that the interpretation is no longer in dispute: in 2012 the U.S. Anti-Doping Agency documented the program, and in January 2013 Armstrong confessed to Oprah Winfrey that he had doped in all seven of his Tour wins.

So this is a rare case where the documented record is settled. That lets us ask a narrow, honest question: on a statement we now know was false, what did the measurable signals actually do? The answer is more modest than a TV lie-detector would claim — and more interesting.

The analyzed video

This is RedFlag's own output — the analyzer running on the clip. The screen turns red during the 6 flagged moments, the gauge swings toward red, and the counter climbs as each red flag is caught.

This is RedFlag's actual output for this clip — the analyzer's live overlay (red tint, gauge, flag counter, captions) captured straight from the app. Run it yourself on any interview at the analyzer. Original clip: Trans World Sport on YouTube ↗.

What RedFlag flagged

Across the 6-minute clip, RedFlag flagged 6 moments totalling 1 minute 25 seconds — about 23% of the footage. Each flagged episode lasts at least 3 seconds (shorter blips are dropped as measurement noise). The flagged windows:

#TimestampLength
10:28 – 0:4315s
20:59 – 1:2223s
32:30 – 2:4717s
43:00 – 3:1718s
55:08 – 5:179s
65:41 – 5:465s

What actually drove the flags

This is the part a lie-detector app would hide. The flags here are almost entirely a pitch-monotony story, not a facial one. RedFlag's verdict is pitch-anchored: it fires when a speaker's pitch variability flattens below the range typical of engaged conversation — roughly 2.3 to 4.5 semitones of variation. In the flagged windows above, Armstrong's delivery goes flat and stays flat; his blink rate also dips, consistent with the cognitive-load literature (concentration suppresses blinking), though in this clip the blink channel was secondary and pitch carried the episodes.

The exact thresholds — the semitone band, the person-relative blink and voicing knees, the 3-second episode rule — are all public in the scoring method. Nothing here is a black box.

What this does and does not mean

It does not mean the analyzer "detected a lie." Flattened pitch has many causes: rehearsed delivery, fatigue, deliberate control, a difficult question. Plenty of truthful people go monotone. The honest reading is narrower: at these specific moments, this speaker's vocal delivery deviated from his own baseline in a measurable, repeatable way. What that meant is a judgment — and in this particular case, history has already made it.

That gap — between "a measurable signal" and "a conclusion" — is the whole point of RedFlag. It is analytics, not a verdict machine. The full scoring method is public, thresholds and all.

Run your own. Paste any public YouTube interview with a single speaker into the analyzer — the face and voice analysis runs in your browser, free, no signup. RedFlag charts the signals; what they mean stays your call.

Open the analyzer

Test your own judgment · Why no tool can detect lies

Sources