🔒 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 7:40 recording (auto-transcribed, ~889 words) against the published Build First, Ask Questions Later (~780 words).  ·  ← back to the story

0/100faithfulness
Faithful, full-coverage retelling.

The recording is the whole story, start to finish — and the page follows it beat for beat. Every hypothesis, every term (the Builder, Dark Mode, V1, discovery), and the closing "massive fight" teaser all survive. It's tightened and styled for reading, not invented.

Where the score comes from

Story beats preservedDid every moment Rajesh told make it onto the page?95%
Facts & quotes accuracyNumbers, names, dates, and direct quotes93%
Tone & voice matchDoes it still sound like Rajesh telling it?92%
Verbatim wordingWord-for-word overlap — low by design for a written piece33%
Length retainedStory is 88% the length of the spoken version88%

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

✓ Kept faithfully

  • "There will be a lot of software" — the first hypothesis, with Base44, Emergent named as proof
  • Three things change — composition, nature of work, ways of working
  • "May 2026… almost two years back" — the dates land exactly as spoken
  • The Builder — collapsing PM / BA / dev / frontend / backend / QA / designer into one role that partners with AI
  • Discovery — "right thing vs. building it right", "two, three days" to "a couple of weeks"
  • The VC-subsidy asterisk — "cheapness is debatable… funded by VC money"
  • V1 + the graveyard — "a lot more abandoned projects" without the old cost
  • Dark Mode — "earlier we called it hackathon", two-to-four days of intense focus
  • The closing teaser — "it ended in a massive fight… the next one"

~ Changed / dropped

  • "Lord of software" (transcription artefact) → correctly read as "a lot of software"
  • "half these in code" → smoothed to "even half-decent code"
  • Blockquote framing — "if everybody can become a coder, what happens to existing teams?" is a page device; spoken as two flowing sentences
  • "Real usage beats a discovery workshop. Every time." — a sharp page line; the spoken version is softer ("get real-life data to see whether it is worth building more")
  • "Notifications off. World off." — added colour for Dark Mode; the recording says only "intense focus"
  • Otherwise: no facts invented — the page adds rhythm and punctuation, not events
Read it like this: the big number is faithfulness of meaning, not transcription accuracy. A score in the mid-90s means "trust this page as Rajesh's story" — the recording covers the whole arc and the page keeps all of it. The 33% verbatim bar is healthy — if it were near 100, the page would just be a transcript, not a voice.

Method. Audio transcribed locally with whisper.cpp (base.en). Verbatim (33%) and length (88%) 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.