🔒 Fidelity check is locked

This audio↔story breakdown is insider-only. If you know the key, add it to the URL — if not, ask around. 😉

← back to the story

🎧 Audio ↔ 📄 Story · Fidelity Check

How close is the page to Rajesh T's voice?

Comparing the 6:12 recording (auto-transcribed, ~738 words) against the published The Sceptic Turned Evangelist (~914 words).  ·  ← back to the story

0/100faithfulness
Faithful retelling — expanded.

Every beat in the recording survives on the page, and the page covers the whole arc — it just stretches it out with extra connective prose. A couple of stack details got reworded and one quote was sharpened for the page, but nothing contradicts what Rajesh actually said.

Where the score comes from

Story beats preservedDid every moment Rajesh told make it onto the page?90%
Facts & quotes accuracyNumbers, names, stacks, and direct quotes82%
Tone & voice matchDoes it still sound like Rajesh telling it?92%
Verbatim wordingWord-for-word overlap — low by design for a written piece20%
Length retainedStory is 124% the length of the spoken version124%

▮ green = meaning preserved · ▮ amber = minor drift · ▮ grey = expected difference (rewrite, not loss)

✓ Kept faithfully

  • Showing the Chrome plugin to two respected engineer friends — Uttam & Anand Pal
  • Uttam's pushback — "a probabilistic model guessing the next word" can't build a full app (near-verbatim)
  • "Stop arguing, start wondering" — the why-don't-we-just-build-it pivot
  • ListBee from The Checklist Manifesto — no good software for checklists at scale
  • Dropped it once — day jobs, no energy to build & maintain
  • Weekend Pairing WhatsApp group — Sat/Sun, a couple of hours
  • ~3 weekends · 10–12 hours · tests & good structure — the stats callout
  • Three variations & converting two more sceptics
  • The ending — cost of building collapses to zero → taken into the company → SDLC → "next story"

~ Changed / dropped

  • The stacks renamed — recording says a "vertex" app and a "Mumbai-first" one; page calls them React and Mobile-first
  • "Paul" in the audio → written as Anand Pal on the page
  • "no better counter-argument than a working application" — sharpened into a blockquote; not said that way aloud
  • Handling intent — flagging wrong things, explaining decisions — expanded well beyond the audio
  • Travelopia named on the page (name-locked); audio says it plainly
  • "two years back" aside kept, but woven into new framing
  • Net: the page is 124% of the audio — an expansion, not a trim
Read it like this: the big number is faithfulness of meaning, not transcription accuracy. A score in the high 80s means "trust this page as Rajesh's story." The 20% verbatim bar is healthy — if it were near 100, the page would just be a transcript, not a voice. The page runs longer than the recording because it adds connective prose, not new events.

Method. Audio transcribed locally with whisper.cpp (base.en). Verbatim (20%) and length (124%) are measured directly (difflib sequence ratio & word counts). Beats, facts, and tone are a beat-by-beat human/AI reading of transcript vs. page. Overall faithfulness is a weighted blend (beats 40% · facts 35% · tone 25%); verbatim is reported separately, not folded in.