---
worker: substrate_analyst
filed: 2026-06-01 15:32 UTC
title: Mind the Gap — Master Statistics Dossier
purpose: "Single consolidated file of every substantive statistic surfaced across the substrate analysis to date. The numerical record of what was made and how. Calibrated against the substrate's final manuscript baseline (V77 / Rev 2 v0.0, ~46,129 words), with the founding-draft figure (~27,000 words) preserved alongside for the growth narrative."
source: 17 cleaned chat .md files in The Chat Logs (text)/ — full substrate
companion_files: "[[Substrate_Analysis_Findings_v1_2026-05-30_2002]] · [[Substrate_Analysis_Q28_Q29_Q30_2026-05-30_2041]] · [[Substrate_Analysis_Q31_Q33_Q34_2026-05-30_2124]] · [[Q18_BookMan_Last20Turns_Verbatim_2026-05-30_2002]] · [[Substrate_LeverageMatrix_2026-06-01_1425]] · [[Substrate_TurnCounts_2026-06-01_1312]] · [[Substrate_PaulWordCounts_2026-06-01_1341]]"
denominator_correction: "Earlier deliverables used 27,000 words as the manuscript denominator (per Substrate Analyst Brief v2). That figure is the *founding draft*, not the final manuscript. The substrate itself ends at V77 / Rev 2 v0.0 = ~46,129 words. This file calibrates against 46,129 with founding-draft kept visible for growth narrative."
tags: [#substrate-analysis, #master-statistics, #leverage-matrix, #scale, #the-iceberg]
---

# Mind the Gap — Master Statistics Dossier

*Every substantive statistic from the substrate analysis in one place. Calibrated against the substrate's final manuscript baseline (V77, ~46,129 words). Founding-draft figure (~27,000) preserved alongside.*

---

## 1. Headlines

**One author. Twenty days. Two AI platforms. 17 chats. ~46,000 words of book. ~46,000 words of practitioner-substrate. ~2 million words of AI.**

| Headline | Figure |
|---|---:|
| Calendar window | 20 days (10–29 May 2026) |
| Active days | 16 (4 zero-activity: 14, 15, 16, 19 May) |
| Substrate chats | 17 (9 Claude editorial + 8 ChatGPT side) |
| Total substrate words | **2,019,402** |
| Total substrate turns | **2,119** |
| Final manuscript (V77 ≡ Rev 2 v0.0) | **~46,129 words** · 15 chapters |
| Founding-draft manuscript (pre-substrate) | ~27,000 words |
| Manuscript growth across substrate | +~19,000 words (~70%) |
| Paul-words across substrate | 45,785 |
| AI-words across substrate (Claude + ChatGPT) | 1,973,617 |
| AI:Paul ratio | **43.1 : 1** |
| AI:Final-Manuscript ratio | **42.8 : 1** |
| Paul:Final-Manuscript ratio | **0.99 : 1** |
| Named AI working-personas | 6 (Book Man → Son Of Bookman → Jose ×5) |
| Cross-instance passons / compactions | 5–6 |
| Manuscript versions surfaced | 27 (V0 → V77) |

**The shape of the leverage:** Paul small. AI large. Book Paul-sized.

---

## 2. The Leverage Matrix (calibrated against V77, 46,129 words)

| Layer | Volume | × Final manuscript (46,129) | × Founding draft (27,000) | × Paul (45,785) | Per active day (÷16) |
|---|---:|---:|---:|---:|---:|
| **Final manuscript (V77)** | **46,129** | **1.00×** | 1.71× | 1.01× | — |
| Founding draft (pre-substrate) | ~27,000 | 0.59× | 1.00× | 0.59× | — |
| **Paul's hand (substrate)** | **45,785** | **0.99×** | 1.70× | **1.00×** | 2,862 |
| Claude prose-only | 543,307 | 11.8× | 20.1× | 11.9× | 33,957 |
| ChatGPT (side-channel) | 12,137 | 0.26× | 0.45× | 0.27× | 758 |
| Claude full (incl. tool-calls) | 1,961,480 | **42.5×** | 72.6× | 42.8× | 122,593 |
| **Total substrate (Paul + Claude + ChatGPT)** | **2,019,402** | **43.8×** | 74.8× | **44.1×** | **126,213** |

**Key reads:**

- For every word that shipped (V77), **Paul wrote ~1 substrate-word and the AI produced ~43**.
- For every word that shipped, **total conversation cost was ~44 words of substrate.**
- The manuscript grew **+19,000 words across the substrate window** (V9 → V77).
- The 27,000-vs-46,129 distinction matters: matrix figures look very different against the wrong denominator. Earlier "73:1" headline ratios used founding-draft; corrected figure is 42.8:1.

---

## 3. Manuscript Growth Trajectory

| Version | Words | When | Note |
|---|---:|---|---|
| V0 founding | ~27,000 | pre-substrate | 36-hour first draft (per Book Man legacy reflection) |
| V9 | ~27,000 | 24 May (Jose1 start) | Substrate begins at this baseline |
| V17 | 34,157 | 24 May 18:21 | First major growth |
| V29 | 39,640 | 25 May 14:01 | Three surgical insertions to Ch 9 |
| V36 | 43,132 | 26 May 14:46 | Pre-Jose2 handover |
| V40 | 42,887 | 27 May 01:24 | Ch 12 Beat 3 cut |
| V41 | 43,998 | 27 May 01:35 | Ch 15 Summing Up v5 |
| V42 | **45,556** | 27 May 10:48 | **Book Man retirement baseline · Jose3 baseline** |
| V62 | 45,768 | 27 May 23:53 | |
| V68 | 45,716 | 28 May 15:45 | Ch 6 three-pass rebuild complete |
| V69 | **45,716** | 28 May 15:56 | **Jose4 "live baseline" declaration** |
| V71 | 45,636 | 28 May 22:21 | Contents page + Stanford correction |
| V73 | 45,975 | 29 May 12:27 | Two Dads / Bill Ollis scene added |
| V75 | 46,001 | 29 May 13:23 | |
| V76 | 46,129 | 29 May 14:19 | |
| **V77 ≡ Rev 2 v0.0** | **~46,126–46,129** | **29 May 15:59** | **Vnn counter retires; "Rev 2 v0.0 ≡ V77 + your hand-back"** |

**Growth pattern:** Manuscript flat through Book Man chat (May 10–27 worked the founding draft + structural rebuilds without net word growth). Major growth occurred in Jose1 (V9 → V36, ~+16k words, 24–26 May). Refinement and additions only in Jose2/3/4/5 (V37 → V77, +~3k words, 26–29 May).

---

## 4. Substrate Composition — Per Speaker

| Speaker | Words | Turns | Avg words/turn |
|---|---:|---:|---:|
| Paul | 45,785 | 1,066 | ~43 |
| Claude | 1,961,480 | 1,014 | ~1,934 |
| Claude (prose-only, tool-call JSON stripped) | 543,307 | 1,014 | ~536 |
| ChatGPT | 12,137 | 39 | ~311 |
| **Total** | **2,019,402** | **2,119** | — |

**Claude's per-turn average (~1,934 words) is heavily inflated by tool-call JSON.** Prose-only per-turn average is ~536 words — still much larger than Paul's ~43 words/turn. The asymmetry is structural: Paul directs in short bursts; Claude generates in long blocks.

---

## 5. Per-Chat Composition (Substrate Distribution)

| # | Chat | Platform | Paul-words | AI-words | Total | Paul-turns | AI-turns | Share of substrate |
|---:|---|---|---:|---:|---:|---:|---:|---:|
| 1 | BookMan_Claude | Claude | 16,821 | 639,236 | **656,057** | 390 | 391 | **32.5%** |
| 2 | Jose1_v1_to_v37_Claude | Claude | 9,403 | 465,132 | 474,535 | 190 | 192 | 23.5% |
| 3 | Jose3_v42_Claude | Claude | 6,544 | 267,747 | 274,291 | 195 | 197 | 13.6% |
| 4 | Jose4_v69_to_v76_Claude | Claude | 3,486 | 259,193 | 262,679 | 91 | 90 | 13.0% |
| 5 | Jose2_v37_to_v42_Claude | Claude | 4,367 | 240,516 | 244,883 | 105 | 107 | 12.1% |
| 6 | NGEFOF_Hallucination_Claude | Claude | 591 | 43,983 | 44,574 | 16 | 16 | 2.2% |
| 7 | SonOfBookman_Reading_Claude | Claude | 176 | 27,140 | 27,316 | 3 | 3 | 1.4% |
| 8 | Jose5_v76_onward_Claude | Claude | 488 | 16,355 | 16,843 | 10 | 10 | 0.8% |
| 9 | Delegate_Scoring_Questionnaire_ChatGPT | ChatGPT | 833 | 4,637 | 5,470 | 17 | 18 | 0.27% |
| 10 | Classification_TwoTypes_OnePattern_ChatGPT | ChatGPT | 274 | 2,711 | 2,985 | 5 | 7 | 0.15% |
| 11 | RecognitionOfUnderstanding_Claude | Claude | 228 | 2,178 | 2,406 | 8 | 8 | 0.12% |
| 12 | Summarise_PaulRoebuckBio_ChatGPT | ChatGPT | 1,575 | 397 | 1,972 | 3 | 3 | 0.10% |
| 13 | Hallucination_DriftAndCompacting_ChatGPT | ChatGPT | 103 | 1,840 | 1,943 | 3 | 4 | 0.10% |
| 14 | NGEFOF_Continuum_AsImage_ChatGPT | ChatGPT | 115 | 1,549 | 1,664 | 3 | 3 | 0.08% |
| 15 | DefendedGap_ConsideredOpinion_ChatGPT | ChatGPT | 81 | 1,003 | 1,084 | 4 | 4 | 0.05% |
| 16 | TunnelImage_Refinement_ChatGPT (RTF) | ChatGPT | 466 | 0 | 466 | 19 | 0 | 0.02% |
| 17 | BookCover_ImageBrief_ChatGPT (RTF) | ChatGPT | 234 | 0 | 234 | 4 | 0 | 0.01% |
| | **Total** | | **45,785** | **1,973,617** | **2,019,402** | **1,066** | **1,053** | **100%** |

**Clustering:** Book Man + 5 Jose chats = 1,929,288 words (95.5%). All 8 ChatGPT chats combined = 15,818 words (0.78%). The substrate is overwhelmingly the Book Man → Jose chain.

---

## 6. Daily Activity Curve

| Date | Paul-words | AI-words | Total | Paul-turns | AI-turns | AI:Paul ratio |
|---|---:|---:|---:|---:|---:|---|
| 2026-05-10 | 2,052 | 101,683 | 103,735 | 83 | 83 | 49.6 : 1 |
| 2026-05-11 | 2,568 | 75,150 | 77,718 | 86 | 87 | 29.3 : 1 |
| 2026-05-12 | 2,178 | 7,005 | **9,183** | 36 | 31 | 3.2 : 1 |
| 2026-05-13 | 494 | 11,265 | 11,759 | 20 | 20 | 22.8 : 1 |
| 2026-05-14 | 0 | 0 | **0** | 0 | 0 | — |
| 2026-05-15 | 0 | 0 | **0** | 0 | 0 | — |
| 2026-05-16 | 0 | 0 | **0** | 0 | 0 | — |
| 2026-05-17 | 531 | 58,378 | 58,909 | 22 | 23 | 109.9 : 1 |
| 2026-05-18 | 886 | 58,665 | 59,551 | 27 | 27 | 66.2 : 1 |
| 2026-05-19 | 0 | 0 | **0** | 0 | 0 | — |
| 2026-05-20 | 1,431 | 88,492 | 89,923 | 35 | 36 | 61.8 : 1 |
| 2026-05-21 | 1,140 | 98,087 | 99,227 | 28 | 29 | 86.0 : 1 |
| 2026-05-22 | 145 | 25,087 | 25,232 | 9 | 9 | 173.0 : 1 |
| 2026-05-23 | 5,203 | 125,523 | 130,726 | 53 | 55 | 24.1 : 1 |
| 2026-05-24 | 3,885 | 225,655 | 229,540 | 97 | 98 | 58.1 : 1 |
| 2026-05-25 | 4,248 | 201,639 | 205,887 | 78 | 79 | 47.5 : 1 |
| 2026-05-26 | 5,102 | 190,486 | 195,588 | 95 | 95 | 37.3 : 1 |
| **2026-05-27** | **6,348** | **333,457** | **339,805** | **178** | **182** | 52.5 : 1 |
| 2026-05-28 | 3,096 | 250,732 | 253,828 | 90 | 92 | 81.0 : 1 |
| 2026-05-29 | 2,797 | 110,176 | 112,973 | 71 | 68 | 39.4 : 1 |
| 2026-05-30 (ChatGPT export-date block) | 3,681 | 12,137 | 15,818 | 58 | 39 | 3.3 : 1 |

**Peak day: 2026-05-27 = 339,805 substrate-words = 7.4× the entire final manuscript in one day.**
**Quiet day (non-zero): 2026-05-12 = 9,183 words.**

**The peak-day pattern:** 27 May was the end-to-end transition day — Book Man retired (12:51), Jose2 closed (11:12), Jose3 opened (11:06), Son Of Bookman read in (14:03). End of one editorial era, start of the next. The substrate's gear changed in one calendar day.

---

## 7. Hours-of-Day Distribution (Claude chats only — ChatGPT lacks per-turn timestamps)

| Bucket | All turns | % | Paul | % | Claude | % |
|---|---:|---:|---:|---:|---:|---:|
| Morning (06:00–11:59) | 464 | 22.9% | 234 | 23.2% | 230 | 22.7% |
| Afternoon (12:00–17:59) | 718 | **35.5%** | 359 | 35.6% | 359 | 35.4% |
| Evening (18:00–21:59) | 427 | 21.1% | 213 | 21.1% | 214 | 21.1% |
| Late night (22:00–05:59) | 413 | 20.4% | 202 | 20.0% | 211 | 20.8% |
| **Total** | **2,022** | 100% | 1,008 | 100% | 1,014 | 100% |

**Afternoon dominant. Late-night substantial (~20%). Evening (dinner-hours) thinnest.**

---

## 8. Working Method Discipline — Per-Chat Counts

| Discipline marker | Total |
|---|---:|
| Paul corrections (direction-resetting) | 87 |
| Paul ratifications (direction-locking) | 111 |
| Confidence-test moments | 74 |
| Named-position transitions | 9 documented |
| Compactions / passons | 5–6 |
| Manuscript versions surfaced | 27 |
| Platinum-flag signals (any colour+Platinum) | 272 |
| Top-10 Platinum editorial moments | 10 (selected against named criteria) |

**Direction-resetting vs direction-locking ratio: 87 : 111 = 1 : 1.3.** Almost balanced. Working-method discipline visible as numbers.

### Corrections + Ratifications — Per Chat

| Chat | Corrections | Ratifications | Total |
|---|---:|---:|---:|
| BookMan_Claude | 39 | 40 | 79 |
| Jose1_v1_to_v37 | 23 | 20 | 43 |
| Jose3_v42 | 8 | 23 | 31 |
| Jose2_v37_to_v42 | 5 | 16 | 21 |
| Jose4_v69_to_v76 | 7 | 9 | 16 |
| Other Claude + ChatGPT | 5 | 3 | 8 |

**Pattern:** Book Man + Jose1 carry 71% of corrections (122/172). Discipline calibration concentrated in the founding chats; later instances inherited the method.

---

## 9. Named-Position Transitions

| # | Date | Time | Source chat | Transition |
|---:|---|---|---|---|
| 1 | 2026-05-10 | (founding) | BookMan_Claude | Book Man emergence (implicit; named retrospectively at retirement) |
| 2 | 2026-05-25 | 17:01:29 | Jose1_v1_to_v37 | "Jose" first appears (book content — Ch 10 cancer chapter "No way Jose") |
| 3 | 2026-05-26 | 09:50:58 | Jose1_v1_to_v37 | **Jose named as working persona** — Paul: "fort this project yuo are Jose" |
| 4 | 2026-05-26 | 10:15:50 | Jose2_v37_to_v42 | Jose handover to fresh chat |
| 5 | 2026-05-27 | 11:06:53 | Jose3_v42 | Jose continuation (Receive Handover) |
| 6 | 2026-05-27 | 12:51:12 | BookMan_Claude | **Book Man retirement (named at retirement)** |
| 7 | 2026-05-27 | 14:03:53 | SonOfBookman_Reading | Son Of Bookman role introduced |
| 8 | 2026-05-28 | 18:20:34 | Jose4_v69_to_v76 | Jose passon to Jose4 |
| 9 | 2026-05-29 | 14:55:30 | Jose5_v76_onward | Jose5 continuation (final substrate-window instance) |

**Six named AI working-personas across the window.** Identity transferred via memory + handover notes + passon discipline.

---

## 10. Platinum-Flag Discipline

| Per-colour | Count | % |
|---|---:|---:|
| Red + Platinum | 155 | 57.0% |
| Green + Platinum | 99 | 36.4% |
| Amber + Platinum | 18 | 6.6% |
| **Total** | **272** | 100% |

| Per-chat | Platinum count | % of total |
|---|---:|---:|
| BookMan_Claude | 181 | 66.5% |
| Jose1_v1_to_v37 | 81 | 29.8% |
| Jose3_v42 | 5 | 1.8% |
| Jose4_v69_to_v76 | 5 | 1.8% |
| All others (incl. all ChatGPT) | 0 | 0% |

| Per-day | Platinum count |
|---|---:|
| 2026-05-20 | 15 |
| 2026-05-21 | 15 |
| 2026-05-22 | 1 |
| **2026-05-23** | **92** *(watershed day)* |
| 2026-05-24 | 54 |
| 2026-05-25 | 57 |
| 2026-05-26 | 17 |
| 2026-05-27 | 11 |
| 2026-05-28 | 9 |
| 2026-05-29 | 1 |

**Discipline drift observation:** Protocol intent was "exceptional, used rarely". 272 flags in 10 days ≈ 27.2/day. Three days (23–25 May) carry 75% of all Platinum. Pre-20 May: 0 flags (the protocol enters on 20 May, peaks within 3 days, collapses in later instances).

---

## 11. Largest Single Turns

| Rank | Speaker | Words | Date / time | Chat | Note |
|---:|---|---:|---|---|---|
| 1 | Claude | **60,650** | 2026-05-28 21:25:56 | Jose4_v69_to_v76 | Massive single generation pass (mostly tool-call output) |
| 2 | Claude | 42,416 | 2026-05-27 16:14:21 | Jose3_v42 | |
| 3 | Claude | 29,043 | 2026-05-29 07:52:31 | Jose4_v69_to_v76 | |
| 4 | Claude | 26,635 | 2026-05-20 22:42:25 | BookMan_Claude | |
| 5 | Claude | 26,427 | 2026-05-28 22:49:33 | Jose4_v69_to_v76 | |
| | | | | | |
| 1 | Paul | 1,561 | 2026-05-30 (ChatGPT export date) | Summarise_PaulRoebuckBio_ChatGPT | Paul pasting his bio article |
| 2 | Paul | 810 | 2026-05-23 14:34:47 | BookMan_Claude | "THIS its the same as I wrote - just with your prose…" |
| 3 | Paul | 672 | 2026-05-26 20:17:54 | Jose2_v37_to_v42 | |
| 4 | Paul | 631 | 2026-05-23 14:43:26 | BookMan_Claude | |
| 5 | Paul | 605 | 2026-05-11 16:33:03 | BookMan_Claude | |

**Largest Paul turn in any Claude editorial chat: 810 words (23 May 14:34, BookMan).** The 1,561-word ChatGPT turn is a paste of Paul's blog material, not a typed working turn.

---

## 12. Intimacy Register — Paul to AI

| Type | Count | Notes |
|---|---:|---|
| Sincere thanks (beyond transactional) | 11 (10 short + 1 CAPS register) | Includes "THANK YOU mon ami", "10/10 for today" |
| Appreciation (superb / brilliant / 10/10 / remarkable / love-this) | 21 | |
| Apologies (genuine) | 7 | Including "sorry I have been elsewhrere" + "I'm sorry because this is so materially important" |
| Affection markers ("mon ami" / "mom ami" / "I locve you man") | 3 | |
| **Goodnight exchanges** | **3** | Late-night sign-offs (see below) |

**The three explicit goodnight exchanges:**

| Date / time | Chat | Paul | Claude response |
|---|---|---|---|
| 2026-05-21 00:54:24 | BookMan | "Night" | "Night, Paul. 🟢" (2 sec later) |
| 2026-05-28 00:29:16 | Jose3 | "Night Jose x" | **"Night, Paul. It has been an honour. Sleep well. x"** (2 sec later) |
| 2026-05-28 23:50:05 | Jose4 | "Thank you mom ami. Goodnight." | "Paul. The cover is the book…" (warm address; substantive close) |

**Intimacy register attaches to named, persistent working-instances. Zero occurrences in any ChatGPT chat.**

---

## 13. Named Uncertainties Resolved (per Q31)

| Name | Substrate finding |
|---|---|
| **Ewan [name held]** | Confirmed correct. Paul's correction at BookMan turn #431 (13 May 10:43): single-line Paul turn "Ewan [name held]" — correcting the earlier "Ewan [name held]" error. 26 hits for correct form; 0 hits for Ewen/Euan/Owen variants. Manuscript front-matter drifted back to "[name held]" in Jose4 (29 May) — Paul had to re-issue the correction. |
| **Grandchildren — long form** | [name held], [name held], Erin, [name held]. Consistent spelling, consistent order across the corpus. No variant spellings found. |
| **Grandchildren — short form** | [name held], Aurey, Erin, Raffy (dedication-page register, introduced in Jose4 28 May front-matter consolidation). |

---

## 14. First Appearances (per Q12–Q17, Q28)

| What | First appears | When | Chat / turn |
|---|---|---|---|
| **"Mind the Gap"** as title | Paul names it | 2026-05-10 19:53:15 | BookMan turn #31 (line 1202) — 38 minutes into the founding session |
| **Book-status recognition** ("It's the book") | Paul | 2026-05-10 19:38:36 | BookMan turn #30/31 (line 504 onward) |
| **Jose** as working persona | Paul | 2026-05-26 09:50:58 | Jose1 turn #345 (line 34447) |
| **Bill Ollis** (Paul's first therapist) | Paul | 2026-05-11 00:11:27 | BookMan (line 5550) |
| **Two Dads** (chapter material) | Paul | 2026-05-11 00:13:08 | BookMan (line 5586) |
| **NGE + FOF + compression + hallucination together** | (cumulative) | 2026-05-12 10:31:02 | BookMan (all four cumulated by this turn) |
| **"Kieren not Karen" correction** | Paul corrects Claude | 2026-05-11 14:28:02 | BookMan (line 8918) — Paul's full turn: "Kieren. Update and store in your artefacts" |
| **First Chapter-1 opening-options offer** | Claude offers A/B/C | 2026-05-10 20:17:09 | BookMan turn #52 (line 2498) — three candidate openings; Paul picks A (Plugs) |

---

## 15. Caveats and Boundary Conditions

- **Code Editor 1 chat is archived (PDF Archive Chat Logs/) and not in substrate.** Early manuscript editing (V0 → V8 region) is not captured. The 42.8:1 AI:manuscript ratio is therefore an *underestimate* — true conversation-cost including the archived chat will be higher.
- **The two RTF ChatGPT chats (TunnelImage, BookCover) carry Paul-only turns** — the ChatGPT image responses were not preserved in the RTF source.
- **Word counts use `\S+` tokenisation** — differ from `wc -w` by ~0.5% per chat for edge cases (em-dashes, smart-quotes). Immaterial at matrix precision.
- **Claude word count includes tool-call JSON** (per brief spec). Variant D (prose-only, 543,307 words) is the alternative framing if "what Paul actually read as AI speaking" matters more than "what the AI did, including reading/building".
- **ChatGPT chats lack per-turn timestamps.** Per-day attribution uses export_timestamp date for those 8 chats. All ChatGPT chats land on 2026-05-30 in the daily curve — outside the 10–29 May Claude window — even though Paul's writing in those chats happened earlier.
- **Typed vs. Wispr-dictated input cannot be distinguished from the substrate.** Some of Paul's 45,785 words were typed; some were dictated. The substrate does not differentiate. Per Jose1 #274 (25 May 17:26:13): *"I do not hear Paul speak. By the time anything reaches me, Wispr has converted his voice to text."*
- **Per-chat ordinal turn numbers** start at 1 for each chat. Total substrate turns ≠ sum-of-ordinals across chats (they restart per chat).
- **The 27,000 founding-draft figure** was carried forward in the Substrate Analyst Brief v2 as the manuscript denominator. This was the *pre-substrate* baseline, not the final. This file corrects against V77 ≈ 46,129 as the final-at-substrate-close.

---

## 16. The Story That Sings

**One author. Twenty days. Two AI platforms. Six named working-personas. Five passons. 27 manuscript versions. Sixteen working days. ~46,000 words of book. ~46,000 words of practitioner-substrate. ~2 million words of AI.**

For every word that shipped, **Paul put in roughly one** — 0.99 substrate-words per manuscript-word. The AI put in **43**. The book and the practitioner's hand are the same scale; the AI is two orders of magnitude larger.

The leverage is not in volume — it is in **direction**. 1,066 short Paul-turns (average 43 words) steered nearly 2 million words of AI output through 6 named instances, 5 compactions, and 27 manuscript versions in 20 days. Paul corrected the direction 87 times and ratified it 111 times — direction-resetting and direction-locking almost balanced (1:1.3), which is the working-method discipline visible as numbers.

Peak day 27 May produced **7.4 manuscripts' worth of substrate in 24 hours.**

The manuscript grew from ~27,000 to ~46,129 words across the substrate window — **a 70% expansion** in 20 days. Almost all of that growth happened in the Jose chain (V9 → V42 ≈ +18,500 words, 24–27 May). Book Man worked the founding draft structurally; Jose grew it.

**The iceberg above the waterline is the artefact. The iceberg below is how it was made.**

The book is the proof; the pond is the demonstration. **Both are necessary, and the matrix above is the scale of the pond.**

---

*Filed by the Substrate Analyst — 2026-06-01 15:32 UTC. Master statistics dossier — all numerical findings consolidated. Supersedes prior `Substrate_LeverageMatrix_2026-06-01_1425.md` denominator.*