A family-heirloom book + podcast doesn't have a list price. Here's what putting one together looks like in dollars, hours, and tokens.
Three real categories (services, materials, labor) plus a couple of small line items.
| Item | Detail | Confidence | Cost |
|---|---|---|---|
| Neighborhood kid · scanning | ~263 letters × ~2 min each = ~9 hours of someone-else-time flat fee you've already committed to · Alex covered 37 himself (3 self-scans + a 104-photo batch put through ChatGPT-zip splitter) | confirmed | $180 |
| ElevenLabs · PVC + generation | ~250–325K characters total after pivot: ~15-20 curated letters (~50K chars) + narrative audiobook Part 1 (~150-200K chars) + Claude-authored podcast script (~50-75K chars). Most economical path: Pro 1 mo @ $99 covers it; or Creator 3-4 mo @ $22 = $66-88. PVC training already done at no separate charge · 8 sample audios already produced under prior plan · per-letter audio paused 5/23 | good | $90 |
| ChatGPT Plus · cursive transcription | $20/mo × ~6 mo of active project time. Could be $0 incremental if you already pay for it. cleared the cursive bottleneck — biggest single workflow unlock | depends on existing sub | $120 |
| Claude · API or Pro subscription | Pro $20/mo × ~6 mo = $120 if subscription. Pay-per-token estimate would be ~$30–60 with cache hits. heavy on file reads + edits; light on conversation length per session | depends on plan | $120 |
| Blurb · Trade Book print run | ~30 family copies × ~$20 = $600. Order 1-2 proofs first ($40-50) before the family run. paperback 6×9; the production cost peaks here | good | $640 |
| Blurb · hardcover keepsake for Mom | One hardcover printing of the same content, gift quality family decision per session 3 | good | $50 |
| Misc materials | Document protectors / sleeves, scanning props, shipping, paper for proof reads rounding bucket | rough | $50 |
| Total · midpoint estimate | $1,250 | ||
| Range: $1,000 (existing subs + smaller print) — $1,400 (full upside) · scanning line dropped by ~$20 because Alex covered the first 37 letters in-house | |||
Your hours, broken down. Scanning is mostly outsourced now; the bigger investments are processing, narrative, and Mom-interview.
| Activity | Detail | Unit Time | Total Hrs |
|---|---|---|---|
| Scanning (already done) | 37 letters scanned · 17 by Alex single-letter · 20 by Alex in one 104-photo batch (session 8) | ~2 min/ltr | ~1.5 |
| Scanning (remaining) | 263 letters · neighborhood kid · ~9 hrs of their time (you supervise) | ~5 min you/ltr supervise | ~2 |
| ChatGPT routing per letter | upload pages · paste prompt · wait · copy back to Claude · parallel-zip workflow (~20 letters per ChatGPT chat) cuts unit time materially session 8 proved 20-letter parallel batches in 6 chats hits ~5 min/letter total Alex-time | ~3-5 min/ltr | ~15 |
| ElevenLabs · curated letters (~15-20) | paste cleaned text · generate · listen · save MP3 · canonical naming 8 letters already done as sample; ~10 more after corpus complete · ~5 min/letter | ~5 min/ltr | ~1.5 |
| ElevenLabs · narrative audiobook | Part 1 of the book read in Pops's voice · chapter-by-chapter generation + edit + stitch the primary listening artifact | ~1-2 hr/ch | ~10 |
| Podcast script + ElevenLabs production | Claude writes brand-new transcript synthesizing all 300 letters · Alex reviews · ElevenLabs generates a new artifact, not letter-by-letter readings | — | ~8 |
| Mom-interview | 30 questions across 5 acts · recording + light transcription
questions ready at family-interviews/mom-questions.md |
single session | ~6 |
| Chapter outlining + drafting | Part 1 chapters across the corpus · ~6–10 chapters · dual-voice (you + AI-Pops) the biggest single Alex-time bucket | ~3–5 hrs/ch | ~30 |
| Manuscript layout | InDesign or Word · 6×9 trade book · tables of contents · drop caps wireframe is built; the typeset version is the work | — | ~25 |
| Blurb upload + proof review | PDF export · upload · order proof · review · revise · order family run | — | ~5 |
| Family review + iterations | Mom + siblings read draft · catch family-fact errors · re-record audio if needed | — | ~10 |
| Miscellaneous + admin | FOIA retry · 783rd association outreach · standing the project up · this dashboard | — | ~10 |
| Total · your hours (midpoint) | ~140 hrs | ||
| If you valued your time at $50/hr → ~$7,000 of effort. If at $100/hr → ~$14,000. (Not counted in the dollar ledger above.) Was ~125 pre-session-8; bumped by ~15hrs for the parallel-ChatGPT 20-letter batch (zip splitter, contact-sheet placer, integration of all 20 letters end-to-end including this site update). | |||
The hardest number to pin down. Claude doesn't expose a per-conversation token counter from inside the chat, so this is reasoned from session patterns.
| Driver | Detail | Est. tokens | If paid by API |
|---|---|---|---|
| Per-session context loading | Memory files + codebase context + transcripts when relevant ~50–100K tokens/session input · mostly cache hits after first read | ~550K | ~$1.65 |
| File reads during sessions | Transcripts, HTML, CSV — session 8 alone read all 20 new transcripts back for letters.html + draft.html rebuild ~100–400K tokens/session input | ~2.0M | ~$6.00 |
| Tool calls + Claude responses | File edits, writes, bash, search results · session 8 had ~2,000 lines of HTML authored for the site rebuild ~30–150K tokens/session output | ~600K | ~$9.00 |
| Web fetches + research | 783rd unit history, ElevenLabs docs, Wikipedia for the wireframe Interlude irregular; spiked in session 4 | ~300K | ~$1.00 |
| ChatGPT cursive passes | 37 letters × ~10K vision tokens + text response · billed via Plus subscription not a Claude cost; included in ChatGPT Plus line above · session 8 added 20 in parallel zips | ~370K | included |
| Estimated total Claude tokens · 20 sessions to date | ~20M | ~$100* | |
| *Re-estimated at session 20. The jump since session 8 is the multi-agent Workflows (Ch.1 draft, ancestry reconcile, podcast harvest, and the 34-agent batch QC + integration runs) — each spends ~1–2M subagent tokens. At API rates ~$60–150; on a flat-rate Max/Pro plan, effectively capped. Still a ±50% estimate. | |||
Confirmed: the $200 you're paying the neighborhood kid; ElevenLabs published pricing (Creator $22/mo · 100K credits; Pro $99/mo · 500K credits; PVC included on Creator+); Blurb Trade Book pricing in the $18–22/copy range for 6×9 paperback at small quantities; your stated ~2 min/letter scan rate.
Estimated from observation: per-letter Alex-time (~10 min for ChatGPT routing + audio gen + listen-verify) is based on the 8 letters processed so far. Chapter drafting time is rough — depends entirely on how the narrative voice settles.
Estimated honestly without a counter: Claude tokens. There's no built-in token counter for this conversation. Numbers above are reasoned from typical session shape — file reads + tool calls + memory loads — and current Anthropic pricing (~$3/M input, $15/M output for Sonnet-class, with prompt caching reducing effective cost ~70%). Real number could be ±50% off; the order of magnitude is reliable.
Not counted: the value of your time. Most folks doing a family memoir project undercount their hours — at $50/hr that's another $6,500 of effort; at $100/hr, $13,000. This is a labor of love, not a cost-recoverable enterprise. Worth being honest that the hours are real even if not invoiced.
What would change these numbers: (1) if Mom shares Pops's letters that aren't in the box, the corpus grows; (2) if FOIA on the medical/military record comes back, we add research time; (3) if family wants an audiobook compilation distributed beyond the print run, ElevenLabs costs scale linearly with re-renders.