๐Ÿ”’ Fidelity check is locked

This audioโ†”story breakdown is insider-only. If you know the key, add it to the URL โ€” if not, ask around. ๐Ÿ˜‰

โ† back to the story

๐ŸŽง Audio โ†” ๐Ÿ“„ Story ยท Fidelity Check

How close is the page to Sree's voice?

Comparing the 10:05 recording (~1,200 words, Sree's own transcript) against the published Bold and Big, Humble and Small (~1,340 words).  ยท  โ† back to the story

0/100faithfulness
Faithful โ€” but heavily retold.

Every beat and every fact in the recording survives on the page, and nothing contradicts what Sree actually said. But this one carries a heavier writer's hand than most voices here โ€” it's reworded, re-ordered, and runs a touch longer than the spoken take. Read it as a faithful retelling, not a transcript.

Where the score comes from

Story beats preservedDid every moment Sree told make it onto the page?94%
Facts & quotes accuracyNames, projects, the book, the years89%
Tone & voice matchDoes it still sound like Sree thinking out loud?84%
Verbatim wordingWord-for-word overlap โ€” low by design, lower than most (more rewriting)35%
Length vs. recordingThe page runs ~112% of the spoken version โ€” slightly expanded, not trimmed112%

โ–ฎ green = meaning preserved ยท โ–ฎ amber = minor drift ยท โ–ฎ grey = expected difference (rewrite, not loss)

โœ“ Kept faithfully

  • The prompt โ€” Niraj's friend at a UK company, the comparison, "you should write about that"
  • desh, kaal, paristhiti โ€” the pinch-of-salt instinct: place, time, circumstance
  • Chris careful, me push-the-boundary, Mike balances โ€” and "we expect you to be a challenger"
  • "the minimum job of a technology leader" โ€” bold and big, humble and small; the lean/agile root
  • Find a sponsor / a small brand / a small project before scaling an inch
  • Aligned without speaking for weeks โ€” complementary, not identical
  • WordPress ยท our best BTP + tech team ยท the one ask โ€” "tell the other marketing directors"
  • Voice IQ ยท Venky & Sreenath ยท a quiet weekend ยท James & Matt ยท the sales forum ยท Radar
  • Change management > technology โ€” cohorts as enablers, Extreme Ownership, leading up the chain
  • Accelerator + scale = yin and yang โ€” Mike, Kaz on InfoSec, none chased money, 5โ€“6 years together

~ Changed / dropped

  • Linda the BTP & Andy Crank (both named in the recording) โ†’ de-named on the page: "our best BTP," "the marketing director"; "Parik Katiti" โ†’ "Parik" (surnames dropped)
  • Sceptic โ†’ believer ยท "ten years, no one delivered anything" โ€” a detail Sree added from memory; not in this recording
  • "Sreena" โ†’ written as his full name, "Sreenath"
  • Added framing โ€” "open a drawer," "the floor, not the ceiling" are the writer's images, not spoken
  • Re-ordered & expanded โ€” page runs ~12% longer than the take; verbatim overlap only 35%
  • The closing line โ€” "Bold and big. Humble and small. The same people, holding both." is an authorial distillation
  • Not in the recording: the "dig through a drawer you haven't opened" opening ๐Ÿ˜‰
Read it like this: the big number is faithfulness of meaning, not transcription accuracy. 89 means "trust this page as Sree's story." The 35% verbatim bar is low even for a written voice โ€” this one was reworded and expanded more than most. Nothing was invented; the meaning is intact, the wording is the writer's.

Method. The transcript is Sree's own words (provided alongside the recording), not machine-transcribed โ€” so it's the source of truth, not an ASR guess. Verbatim (35%) and length (112%) 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% ยท facts 35% ยท tone 25%); verbatim and length are reported separately, not folded in. The "play from here" timestamps were aligned to the audio with model-free voice-activity detection โ€” accurate to the sentence, not the millisecond.