The OpenSentience Library: series plan

*Seven books beside Unboxed Patterns. Drafted 2026-09-24. Unboxed Patterns is the catalog: 32 patterns, each with a witness. These books are the narratives: they walk one reader through one layer of the stack from the beginning, and each one doubles as a video.*

Why these seven

Unboxed Patterns already covers the ideas at pattern size: locus and agency, refusal as an outcome, content addressing and replay, progress scheduling, evidence discipline. What it does not cover, and what nobody can currently learn anywhere:

• the full eight-rung arithmetic ladder

• PULSE loops and PRISM measurement

• the OS-001…012 map

• how TRVM, WRL and films actually work end to end

• Super and the authority runtime

• the World Cloud / locu.systems business

• the method itself: how this stack decides what it may claim

Each book below fills one of those gaps and cites Patterns chapters rather than re-teaching them.

Suggested recording order: 0 → 2 → 1 → 5 → 3 → 4 → 6. Book 0 is the trailer. Book 2 is the MVP (R97). Then comes what Super runs on (1), what it becomes (5), and the theory (3, 4). The method book (6) is last because it is best told with the other six as its examples.

The shape every book shares

• Length: 6–9 chapters of 800–1,500 words. A book is one evening's read.

• Each chapter: a hook question → the idea in plain words → one concrete picture from the tree

→ what this does not establish → the rung, stated → links into Patterns.

• Each chapter ends in a slide block (6–10 slides with narration). In AFFiNE Edgeless, one

frame per slide becomes one video. A book's first chapter is the 3–5 minute homepage video; the rest are YouTube chapters.

• Every number carries its source and date. Before publishing, re-derive it and strip the

Sources (internal) list (the private-repo rule).

Book 0: The Stack from Above

Status: fully drafted in `book-0-the-stack-from-above/`. Audience: anyone landing on a homepage. No prior knowledge. Hook: "Any order. Any machine. Same answer. And nothing gets power just by existing."

1. The question the stack answers

2. A world you can address (WRL + TRVM)

3. Allowed, not just possible (box-and-box + WEK)

4. The workshop (Super, ampd, bots)

5. Keeping the world (World Cloud, locu.systems)

6. How far it has got (the ladder, honestly)

7. What to read next

Book 1: Any Order, Same Answer

Status: fully drafted in `book-1-any-order-same-answer/` (2026-09-24). The chapter list below is the original brief; the book's own contents are authoritative. The substrate: WRL, TRVM, films, compiled backends. Audience: engineers; the curious who liked Book 0 ch. 2. Hook: "Most computers promise to be fast. This one promises you'll get the same answer, and can prove it."

1. Why order is the enemy. Races, locks and coordinators; what "coordination-free" buys.

2. The interaction calculus in one picture. Nets, rewrites, and why confluence means the order

doesn't matter. Keep it visual, with no formalism past one diagram.

3. Meaning is a hash. WRL worlds, sealing, sem-…; the same meaning gets the same address on

every host. (Patterns: meaning-is-a-hash, identity-is-a-seal)

4. The memory spine. canonicalize → content-address → union → reduce; ~18 % work saved *at toy

scale*.

5. Films. Every run is a byte-exact receipt; replay as evidence. (Patterns: exact-replay)

6. Three judges. TRAAVIIS checks each run by reference reducer, native reducer and oracle:

WRLM proposes, WRL seals, TRVM reduces, TRAAVIIS admits.

7. Compiling a world. One C step function per world, admitted only if it agrees with the

reference everywhere; cbknd- identity moves with the code, never with the meaning.

8. What the numbers say, and don't. 480/480 byte-identical, no speedup claimed; 1.1× vs

HVM4 on one workload; 7.1× through Super, not 439×.

Must not claim: parallel speedup; external reproduction; that WRLM has a model arm; any throughput beyond the named workloads. Sources (internal): TRVM/README.md, TRVM/FINDINGS.md, TRVM/site/index.html, TRVM/compiled/README.md, WRL/README.md, TRAAVIIS/README.md, lab/BENCH.md, wek/b2/trvm/RESIDENT_TIMING.md.

Book 2: Nothing Gets Power by Existing

Status: fully drafted in `book-2-nothing-gets-power-by-existing/` (2026-09-24). The chapter list below is the original brief; the book's own contents are authoritative. Super, ampd, WEK, bots, lanes, carriers: the MVP. Audience: people who run AI agents on real code and have been burned. Hook: "Not a bot, not a machine, not a green checkmark. In this workshop, nothing has authority until it's granted, and every use leaves a receipt."

1. The question Super answers. *What did the work establish, what was it allowed to change,

where did it run, and why may the result be committed?*

2. A screen that can't lie. The rendered world is a function of the last frame; rows disappear

only when the runtime says so.

3. One door. seal → grant → placement → approval → claim → receipt, walked through with a real

trvm.reduce.

4. Bots know a projection, not the world. Creating a bot confers zero authority; STALE is not

UNESTABLISHED. (Patterns: state-does-not-grant-authority, established-not-known)

5. Lanes and carriers. Parallel attacks on a goal; the machine is never the identity. (Patterns:

locus-is-not-its-carrier, carrier-confinement)

6. The write boundary. WEK's B2 contract, and a verifier that is a library and judges every

trace.

7. Super builds Super. The thesis, the gate, and the honest cliffhanger: one bot-proposed change

ran the whole gate on 2026-09-23 on a build that isn't installed; the reviewer ruled the workflow demonstrated, not the task. Still spec.

Must not claim: that Super is deployed or has an installer; that box-and-box runs inside Super (design intent only); that "Super builds Super" is achieved. Sources (internal): super/README.md, super/ampd/README.md, super/site/AGENT_SUPER_APP_BLUEPRINT.md, SUPER_BUILDS_LANE.md, wek/README.md, wek/r3/HANDOFF.md, SUPER_B2_WRITE_BOUNDARY_DESIGN.md, AGENCY.md, REVISION_REGISTER.md (R97).

Book 3: Floor, Then Gradient

Status: fully drafted in `book-3-floor-then-gradient/` (2026-09-24). The chapter list below is the original brief; the book's own contents are authoritative. The eight-rung arithmetic ladder: box-and-box and the bridge. Audience: people who design policy, safety or scoring systems; the mathematically curious. Hook: "An action that fails the floor scores zero, and no amount of cleverness multiplies zero back to life."

1. Why one number is never enough. Weighted scores let a big upside buy back a forbidden

action. The fix is order: feasible ▸ permitted ▸ best.

2. Rungs 1–2: what can happen, how to rank it. Invariant arithmetic.

3. Rung 3: what is allowed. Deontic arithmetic: obligations, prohibitions, the veto.

4. Rung 4: safe over time. Temporal arithmetic and supervision.

5. Rung 5: may the rules change? Reflexive arithmetic and the entrenched ring-0 core.

6. Rungs 6–7: do we know enough; who can ensure it. Epistemic and strategic.

7. Rung 8: can we afford it. Resource arithmetic and the affine ledger.

8. The bridge and its certificate. Every verdict ships its reason; a brick of bricks is a brick.

9. Three laws that fail on purpose. CP5–CP7 print FALSIFIED in red, and the build breaks if one

starts passing. Why declaring your open gaps is a feature.

Must not claim: a 1:1 mapping between rungs and PULSE phases (they are orthogonal); that a production governance engine is deployed; any law count not derived from LAW_MANIFEST.json. Video note: the arithmetic pages and playground.html are the B-roll. Press the playground's button on camera, since it runs the kernel suite live. Sources (internal): AmpersandBoxDesign/box-and-box/README.md, LAW_MANIFEST.json, opensentience.org/*-arithmetic.html, opensentience.org/_rebuild/data/rungs.json, PULSE/docs/ARITHMETIC_INTEGRATION.md.

Book 4: Compose, Pulse, Measure

Status: fully drafted in `book-4-compose-pulse-measure/` (2026-09-24). The chapter list below is the original brief; the book's own contents are authoritative. [&] Protocol, PULSE, PRISM, Graphonomous, and the OS-001…012 map. Audience: agent builders. Hook: "[&] composes agents, PULSE gives them a heartbeat, PRISM measures their effect."

1. Declaring an agent. ampersand.json: remember, reason, time, space, body, govern; and how

[&] sits beside MCP and A2A.

2. From one agent to a coalition. CC1 → CC2; lift; what's built (CC1) and what's a draft

(CC2).

3. Every loop has five beats. retrieve · route · act · learn · consolidate, and why that's the

same shape as Graphonomous's machines.

4. Loops inside loops. Nesting, cadence, and the six signals (TopologyContext … SurpriseSignal).

5. Memory that learns. Graphonomous v0.4 → v2: from a knowledge graph to an evidence graph that

relates facts without owning them.

6. Does it actually get better? PRISM's nine dimensions and why self-measured improvement isn't

external.

7. κ: when knowledge has cycles. Topological routing, the exhaustive proof over 1.9 M systems,

and the benefit that is still untested.

8. The protocol map. OS-001…012 with each one's honest status.

Must not claim: "shipped" for OS-006/009/010/011 (protocols.json disagrees with STACK_COMPLETION.md); that κ routing improves outcomes; Graphonomous MCP as current (it is slated for teardown and Travis reports it broken). Sources (internal): AmpersandBoxDesign/README.md, docs/UMBRELLA.md, PULSE/README.md, PULSE/manifests/, PRISM/README.md, graphonomous/docs/spec/README.md, graphonomous/v2/handoff/STATUS.md, opensentience.org/_rebuild/data/protocols.json.

Book 5: Keep the World

Status: fully drafted in `book-5-keep-the-world/` (2026-09-24). The chapter list below is the original brief; the book's own contents are authoritative. T&R, RRABBIT, World Cloud, studbook, locu.systems: the business. Audience: prospective users and investors. Hook: "Your whole machine is the product. Back up the world, not just the files, and pay only for the work it actually did."

1. Files are the wrong unit. What a backup forgets: the running world.

2. Three sizes of world. WORLD, WORLD VERSION, WRL GRAPH (R11), and the error of confusing them.

(Patterns: three-sizes-of-world, orthogonal-persistence)

3. An OS that ships its own verifier. T&R on FreeBSD; 364 MB server image; about 55 MB written

on first boot per the live page's own measurement (the README's 51 MB is stale).

4. The road. RRABBIT: apps as signs on a road, flying in at 1:1.

5. Restore is a grant. Moving a world to a new machine is a fresh permission that can be

refused; bytes go straight to R2 and never through the Worker.

6. Data that must prove where it came from. studbook, and the one unanswered question

(confidentiality under content addressing). (Patterns: confidentiality-under-content-addressing)

7. Billing for progress. locu.systems: loci, carriers, admitted transitions; balances that can

never move between accounts (R89); the floor never checks a budget (R90).

8. The order it ships in. R97: Super → components → World Cloud → locu.systems, and why "nothing

built" at stage 3 is on schedule.

Must not claim: World Cloud or locu.systems deployed; any price as chargeable (planned tiers are published as "planned and not yet chargeable"; quote them only with that label); studbook as a Supabase replacement today. Sources (internal): CLOUD_V1.md, WORLD_SIZE.md, STACK_HUB.md, studbook/docs/spec/README.md, PUBLISH_PATH.md, locu.systems/docs/spec/90_LOCUS_HOSTING.md, 91_SETTLEMENT.md, computedriven/docs/settlement-boundary.md, travel-and-rrabbit/README.md, RRABBIT/README.md, REVISION_REGISTER.md (R4, R7, R11, R89–R97).

Book 6: Rungs, Not Adjectives

Status: fully drafted in `book-6-rungs-not-adjectives/` (2026-09-24). The chapter list below is the original brief; the book's own contents are authoritative. The method: how this stack decides what it's allowed to say. Audience: builders, reviewers, anyone tired of AI hype. This is the channel's thesis video. Hook: "The most credible number in this whole stack is zero: the count of things anyone outside has reproduced. Here's why we say it first."

1. Five rungs. spec · in_tree · live_local · live_deployed · external; *when in doubt,

downgrade.*

2. A number in two places. Why every count is derived, and the stories of law counts that

drifted (116 → 117 → 210; "64 of 116").

3. Refusal is an outcome. Gates that refuse, and the two gates that could never fire until

someone probed them.

4. Retractions in public. Banded, never rewritten: the RETRACTED lines in CLAUDE.md, the

investor pages' retractions band.

5. Measure the effect, not the input. 7.1× not 439×; a cached module that read "103 passing"; a

harness that serialized its own benchmark.

6. Rulings. The revision register: who decides, and why a ruling beats a recommendation.

7. Built with the thing it builds. R97: work done beside Super is not evidence about Super.

Must not claim: anything as external. Sources (internal): DOCTRINE.md, STACK_COMPLETION.md §8, REVISION_REGISTER.md, OPENSENTIENCE_SURFACE.md, CLAIM_LEDGER.json, lab/BENCH.md, and the Patterns anti-patterns (A Number in Two Places, The Quiet Correction).

Series Two (drafted 2026-09-24)

Chosen from the coverage survey in 02-gaps-and-series-two.md, which also holds four more proposals (The Floor, The Table That Closed Zero Cells, What an Agent Is, Your First Rung) and 20 verified untold stories.

Appendix: portfolio sites and their honest status (2026-09-24)