Dear Pop — Cost & Time Ledger
What this project actually costs · honest estimates with methodology
Updated 2026-06-16 · session 23
Estimates · dollars · hours · tokens

The Ledger

A family-heirloom book + podcast doesn't have a list price. Here's what putting one together looks like in dollars, hours, and tokens.

Audio strategy pivot, 2026-05-23: per-letter ElevenLabs paused after letter-08. ElevenLabs is now reserved for ~15-20 curated letters + the narrative audiobook (Part 1 of the book) + a Claude-authored podcast script. The numbers below have been re-cut to reflect this — the ElevenLabs line shrinks substantially; the per-letter cash figure now reflects transcription + book costs only, with audio called out as a separate bucket.
About these numbers: some are confirmed (the $200 you're paying the neighborhood kid, ElevenLabs published pricing), some are observed (per-letter processing time so far), and some are honest estimates (Claude tokens — there isn't a built-in counter for this conversation). Each row in the ledger flags its confidence level. Range estimates assume the project finishes the full 300-letter corpus.
~$1,250
Total Project Cost
midpoint estimate · range $1.0K–$1.4K · scanning -$20 (Alex covered first 37) · + paid Gemini 3.1 Pro API OCR ~$2.40 (session 20)
~175hrs
Your Time Investment
across ~8 months · +~30 hrs sessions 9–20 (narrative drafting, ancestry records, podcast, batch integration)
300
Letters in the Corpus
111 transcribed · 189 to go · 37.0% · latest 34 via Gemini 3.1 Pro OCR (~$2.40)
~20M
Claude Tokens (Est.)
research, edits + heavy multi-agent workflows across 20 sessions

Where the Money Goes

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

Where the Time Goes

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).

Project Unit Economics

Per-letter for the transcription path · audio is now a separate bucket sized by deliverable, not by letter count
Cash / letter (transcription + print)
~$3.90
$1,180 ÷ 300 letters · scanning + ChatGPT + Claude + Blurb prints. Audio not included (decoupled).
Your time / letter
~6 min
scan supervision + ChatGPT routing + Alex spot-check. Per-letter audio is no longer in the loop.
Audio (project-level, not per-letter)
~$90
~15-20 curated letters + narrative audiobook + podcast — ~4-6 hrs of total runtime.
Family member who hears it
priceless
but the dollar denominator only works because they exist

What's Spent · What's Left

Already Spent

to date · 2026-05-26
  • ElevenLabs (PVC + 8 sample audios)~$55
  • ChatGPT Plus (~3 mo active)~$60
  • Claude Pro / API (~3 mo)~$60
  • Scanning labor$0
  • Print costs$0
  • Your time (~40 hrs)~40 hrs
Project so far ~$175

Still To Spend

projected · ~263 letters + 3 audio deliverables
  • Neighborhood kid · scanning$180
  • ElevenLabs (curated + narrative + podcast)~$35
  • ChatGPT + Claude (~3 more mo)~$120
  • Blurb print run (proofs + family)~$640
  • Hardcover for Mom~$50
  • Misc materials~$50
  • Your time (~100 hrs left)~100 hrs
Remaining ~$1,075

Tokens & Compute · Honest Estimate

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.
"~$1,250 to typeset 300 letters, print the book, and give Pops his voice back where it matters most."
A family heirloom of this scope, professionally done, would run $15K–25K. Doing it yourself with these tools: under $1,300 + your time. Per-letter audio gave way to a curated audiobook + podcast — fewer hours of listening, more weight to each. Session 8's parallel-ChatGPT batch proved the workflow scales: 20 letters end-to-end in roughly the time it used to take to do 5.

Methodology & Caveats

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.