The Skeleton Assembly: Building Paper Skeleton v1

90 min

Overview

You have spent four afternoons manufacturing parts: a value-migration argument and research questions (The Day After), problem statements and a discovery (Prototyping & Informal Benchmarking), a Positioning Memo, a build notebook full of trials and a hunch, a Proposed Study, and — as of lunch — a proof-of-concept tally. Today you stop making parts and assemble the machine: a paper skeleton with a working title, an abstract, a real introduction, and every section pointed at an artifact you already built. Then another team reads it the way a reviewer would. Tomorrow you refine it; Friday afternoon you present it. Nothing new gets invented today — that's the discipline. If a section has no artifact behind it, it goes on the gap list, and the gap list is tomorrow's work plan.

One note on authorship: you built and tried in pairs, but the papers are individual — each of you assembles your own skeleton today. Share the evidence freely; the framing, the introduction, and the claims are yours.

Assembly, not authorship

Every sentence in the skeleton must trace to something your team already produced this week. Writing new claims today is how skeletons drift away from their evidence.

The exception is the introduction — that you write fresh, because it is the one part that only exists once the parts are together.

The session at a glance

1
Lay out the six artifacts and map each to its section.
2
Write the introduction — one paragraph, four sentences that carry it.
3
Write the working title and the 150-word abstract.
4
Outline every section: what it will say, which artifact feeds it.
5
Write the gap list — asserted but not yet evidenced.
6
Swap skeletons: reviewer pass by another team, three questions.

Part 1 · The parts on the table

Map artifacts to sections. This is mechanical — do it fast:

The mapping

Introduction & MotivationThe Day After slides — the capacity unlocked, where value migrates in your domain, why this direction

Background — how it's done todaythe Positioning Memo — current practice, the evidence numbers, borrow/reinvent/invent

Problem statementthe problem statements and discovery slide from the Prototyping session — sharpened by what the week taught

Approachthe agent — its action language and training document, the builds you tried, and the hunch about what matters, with the notebook lines as the design rationale

Proposed evaluationthe Proposed Study — measures, baselines, mini-benchmark, judging protocol — clearly framed as the study you will run after this week

Preliminary explorationsthe proof-of-concept tally — the runs, the repeat, the variation, the excerpts, and the honest summary

Future workempty for now — tomorrow's refinement session fills it

Part 2 · The introduction

One paragraph, and every sentence earns its place. Write these four, in order, then smooth:

The four sentences

1
The unlocked capacity: what becomes cheap and scalable in your domain when cognitive work is near zero cost — your field's version of "when information exploded, organization and search became the inventions that mattered."
2
The problem and the evidence: the specific problem you chose, with the strongest number from your Positioning Memo.
3
The proposal: the agent you built and the question you are asking about it — and what's in your "invent" column that doesn't exist today.
4
How you'll know: your headline measure and baseline from the Proposed Study, in one clause each — and what the rough proof of concept already suggests.

Read it aloud. If a listener could not repeat back what you're doing and why it matters, revise before moving on.

Part 3 · Title and abstract

The title test

say what it contributesa method, a system, a technique, an assessment — not just an area. "Toward X" and "A Study of Y" are warnings, not titles

name the domainyour paper is specific to your field on purpose — let the title say so

Then the abstract: 150 words, and it is the introduction compressed — capacity, problem + evidence, proposal, method, what the proof of concept showed. Write it last-sentence-first if you're stuck: "We present…" forces the contribution into the open.

Part 4 · Section outlines and the gap list

For every section: two or three bullet points of what it will say, plus the pointer to its artifact. Where a bullet has no artifact behind it, don't delete the bullet — flag it:

The gap list

A gap is a claim the paper needs that this week has not yet evidenced: the benchmark case not yet written, the baseline not yet run, the measure not yet validated, the related work not yet read.

List gaps explicitly, ranked by how much the paper depends on them. Tomorrow's refinement session works this list top-down — a good skeleton with an honest gap list is exactly on schedule.

Part 5 · The reviewer exchange

Swap skeletons with another team — a different team than yesterday's evaluation exchange, if the room allows. Read theirs the way this week taught you to read: actor and critic are different jobs, and right now you are the critic.

The reviewer pass — twenty minutes

1
Read the abstract and introduction only. Write down, in one sentence, what you believe they are doing and why. Hand that sentence over first — if it doesn't match their intent, that's the finding.
2
Check three joints: Does the introduction promise only what the proof-of-concept tally can support? Do the measures measure the claim? Is the baseline fair?
3
Write THREE reviewer questions the skeleton cannot yet answer — specific enough that the team knows what work would answer each one.

Return the skeletons. The three questions you received go onto your gap list — usually near the top.

The deliverable: Paper Skeleton v1

Paper Skeleton v1

user prompt

Title: working title Abstract: 150 words Introduction: the paragraph — four sentences, smoothed Sections: each section: 2–3 bullets of content + the artifact it draws from Gap list: ranked — including the three reviewer questions you received

Before we regroup

💾 Capture

Bring Paper Skeleton v1.

Write down:

1The skeleton — title through gap list.
2The one-sentence readback the other team wrote of your abstract — and whether it matched what you meant.
3The three reviewer questions you received.

Be ready to discuss:

Tomorrow morning is multi-agent architectures; tomorrow's refinement session asks whether your approach needs them — and closes your top gap. Friday at 14:00, this skeleton is what you present. It exists now. Everything from here is improvement.