---
name: ima-research-documents
description: Use when producing any In Motion Automation or Bytestream research report, letter paper (LaTeX, TeX Gyre Termes) or decision deck (true 16:9, built from the deck kit, Rev 2); covers the deck kit, slide patterns, charts, generated illustrations, the Internal badge, dataviz rules, house writing and where the single source of truth lives.
---

# IMA research documents

House standard for **Jake-facing research letter papers** and **true 16:9 decision / evidence decks**. Locked 2026-09-25 (Chief). Dataviz rules expanded same day.

**Deck standard Rev 2 (Jake, 2026-11-04):** every decision deck is built from the deck kit (`brand/research/deck-kit/`, live https://brand.agentsinmotion.ai/research/deck-kit/), the format of the decks of 4 Nov 2026. Jake rated it "a notch above" the older CGM-chrome decks: the formatting, the illustrations, the type sizes. Section B below is the short form; the kit README is the full rule.

**Internal badge (Jake, 2026-11-04):** anything internal carries a yellow `Internal` badge. Decks: after the date on the cover (`<span class="internal-tag">Internal</span>`), cover number line `Confidential · decision research · internal`. Letter papers: a yellow `INTERNAL` box under the date on the title page (template line in `research/starter-kit/TEMPLATE.tex`).

## Single source of truth

The canonical copy of this skill, the deck kit, the paper starter kit and GUIDE.md is the vault folder `Main/01 Projects/In Motion Automation/brand/`. Edit there only, then run `../publish-brand.sh "what changed"` to publish brand.agentsinmotion.ai. Everything else is a mirror:

| Mirror | How it stays current |
|---|---|
| `Main/01 Projects/In Motion Automation/skills/ima-research-documents/` | symlink to `brand/skills/ima-research-documents/` |
| `~/.claude/skills/ima-research-documents` (Claude Code on the Mac) | symlink to the same folder |
| `~/.agents/skills/ima-research-documents` (Codex, Gemini, Pi, Goose and other local agents) | symlink to the same folder |
| https://brand.agentsinmotion.ai/skills/ima-research-documents/SKILL.md | published by `publish-brand.sh` |
| Grok Bot global skill `ima-research-documents` | refresh from the live URL above before each deck; the live file wins on any difference |

This is **not** client stationery. Client letters, invoices, reports, portals → sibling skill `brand-documents` (Geist, warm/cool/fresh, client name leads).

Before export: read Mac `Main/01 Projects/In Motion Automation/brand/GUIDE.md` → section **Research papers and decision decks** (incl. **Dataviz**), and skill `jake-house-writing`.

## Public starter kit (start here)

Agents without box or Mac access: download the kit and follow its README. It carries the gold preamble with a neutral skeleton, example figure scripts, and the matplotlib style.

- Kit index: https://brand.agentsinmotion.ai/research/starter-kit/
- How-to: https://brand.agentsinmotion.ai/research/starter-kit/README.md
- Download: https://brand.agentsinmotion.ai/research/starter-kit/starter-kit.zip

## Gold paths (clone, do not reinvent)

| Artifact | Path |
|---|---|
| Letter LaTeX gold | `/workspace/outputs/bytestream-creator-metrics/2026-09-25/latex/CREATOR-GROWTH-METRICS.tex` |
| Public starter kit | https://brand.agentsinmotion.ai/research/starter-kit/ · zip https://brand.agentsinmotion.ai/research/starter-kit/starter-kit.zip (`TEMPLATE.tex`, `references.bib`, `ima-research.mplstyle`, `figures/` examples) |
| Deck kit (current, Rev 2) | `brand/research/deck-kit/` · https://brand.agentsinmotion.ai/research/deck-kit/ · `deck-kit.zip` · reference render `EXAMPLE-DECK.pdf` |
| Deck pack gold (retired chrome, older decks) | `/workspace/outputs/bytestream-creator-metrics/2026-09-25/` |
| Published examples | Mac `…/brand/examples/` · live https://brand.agentsinmotion.ai/examples/ |
| Letter example PDF | `brand/examples/LOCAL-AUTOMATION-MARKET-STUDY.pdf` |
| Deck example PDF | none public; decks carry our own strategy and stay internal |
| Dataviz thumbs | `brand/examples/example-dataviz-unanswered.png`, `example-dataviz-revenue-at-risk.png` |
| Recent example | `/workspace/outputs/bytestream-twitch-gamer-pivot/2026-09-25/` |
| Brand GUIDE | Mac `Main/01 Projects/In Motion Automation/brand/GUIDE.md` |
| Live style book | https://brand.agentsinmotion.ai/ · Research `#research` |
| House writing | skill `jake-house-writing` · `/workspace/outputs/voice/JAKE-HOUSE-WRITING.md` |
| Paper standard | skill `jake-research-papers` · `context/JAKE-RESEARCH-PAPERS.md` |

## Writing the paper (Jake standard, LOCK 2026-10-20)

This skill sets the layout. The words follow `jake-house-writing` plus the paper standard `jake-research-papers` (public copy https://brand.agentsinmotion.ai/research/RESEARCH-PAPERS.md; Mac `~/.claude/skills/jake-research-papers/SKILL.md`). Read both before drafting. In short:

0. The stranger test governs: if a stranger had done exactly the same work, would the sentence still belong? Cut inside thoughts; keep an inside fact only when it changes what the reader sees or can redo, restated as a fact about the work.
1. Write a concept ladder first: every term in order of first use, each built only from earlier terms.
2. Open broad and narrow paragraph by paragraph: the phenomenon, the theory behind it, prior work and its gap, why it matters now, this paper's question, then findings and contributions. Never open with "This paper…".
3. Build common ground (the cognitive science concept) before you use it: every term, acronym and outside object enters common ground in plain words before any sentence relies on it. Outside material enters only where the previous sentence creates the need.
4. Title: short, readable by anyone in the field, no internal project words.
5. Abstract: the same funnel in one paragraph, about 200 to 250 words, with the main numbers.
6. Figure 1 shows the system or question. Results headings state the finding. The discussion is prose, with limitations.
7. Numbers only from sources; citations verified; failed predictions stated plainly.
8. No series or roadmap framing, internal labels or ops theater in the paper.
9. Reread as an outsider who never saw the earlier work, fix every term used before its definition, run the ship check, then hand over for review. Nothing publishes before Jake approves.

## A. Research paper (US Letter LaTeX) — steps

1. Clone preamble from gold `.tex` exactly (`tgtermes`, geometry 1in, fancyhdr, titlesec, natbib authoryear, abstract env). Keep the template's `\bibsep` and `\bibfont` lines, so references are set smaller than the body, and build references from `references.bib`, never by hand.
2. Typeface: **TeX Gyre Termes** only. Not Geist.
3. Geometry: `\usepackage[margin=1in]{geometry}` only. Letter 612×792.
4. Title: default `\maketitle` only. Break with `\\` into 2 balanced lines under default `\LARGE`.
5. Headers: short title left, `Bytestream Research` (or project attribution) right. Footer: **single center folio**. No double folios.
6. Abstract: broad-to-specific funnel per `jake-research-papers` (context, gap, question, method, findings with numbers, implication). No spend / Treg / freefetch theater in abstract.
7. Body: scientific discussion. No product-rec framing (except a proposal paper Jake commissioned; see Hard stops). No Treg/$ budget amounts (ops handbacks only). Named product only as labeled research subject when scored.
8. Figures: follow **Dataviz** below. Keep files for deck reuse.
9. Citations: natbib authoryear; real sources; confidence on thin claims.
10. Filename prefix (Bytestream lane): `BYTESTREAM-`.
11. Pack: PDF root + `latex/` + figures + `CHIEF-HANDBACK.md`.

## B. Decision / evidence deck (true 16:9 IMA cool) — steps

1. Build **only after** the report is finished (do not invent research).
2. Copy the deck kit: `brand/research/deck-kit/deck-build/` into the pack as `deck-build/` (download: https://brand.agentsinmotion.ai/research/deck-kit/deck-kit.zip). Read its `README.md`. The older CGM chrome is retired for new decks.
3. Page: true 16:9, 960 × 540 pt; export through `brand/export.sh --deck-16x9` (the kit's `export.sh` does this and renders a contact sheet).
4. Body: `class="ima-doc ima-cool"`; Geist embedded; type sizes from the kit README (H1 27 px, reading text 13.8 px; cut words rather than shrink type).
5. Slide patterns from the kit's `body.html` only: navy cover with a photo panel and caption, main message with three numbers, terms, figure beside text, table beside photo, offer grid with price column, dated tests, decisions, sources.
6. Every content slide: deck name and wordmark on top, mono kicker, an H1 that states the slide's finding as one sentence, a source line with numbered sources (`{{s:key}}` from `sources.py`), exactly one `ima-slide-num`.
7. Charts: the report's figures, same files, drawn with `deck-kit/figures/house_figures.py` (finding title above, source note below). Set each deck height in `FIGH`.
8. Stills: two to four generated illustrations in the overhead-time recipe (kit README, "Stills"), via treg `reapi.image-gen.gemini-3-pro-image` (16:9, 2K, about $0.03 each), captioned "Illustration: …", 1800 px copies in `assets/deck/`.
9. Internal decks: `Internal` badge after the cover date.
10. Length 10 to 16 slides. Look at every slide as an image (contact sheet) and fix before handing over.
11. Prose: `jake-house-writing` ship check before export; ship `IMA-VOICE-LOCK.md` beside the pack. Cover H1: no orphan last word (nbsp).

## Dataviz (hard rules)

- **Lagom:** sparse ink, clear hierarchy, no 3D, no rainbow, no chartjunk.
- **Palette:** cool-lane ink/navy/sky/cyan ticks; orbit colors only as small accents — never rainbow fills.
- **Libraries:** matplotlib / MATLAB / real plotting libs. Not AI-generated fake charts or Excel screenshots with wrong brand.
- **Data:** source-backed only; caption/note the source; no invented series.
- **Labels:** no overlapping labels (Chief hard fail); readable axis titles + units.
- **Paper embeds:** prefer vector PDF figures in LaTeX; keep PNG companions.
- **Deck embeds:** PNG/PDF copies of the **same** files. Do not redraw with different numbers.
- **Reuse:** paper figures → deck figures (same art).
- **Forbidden:** dual-axis without need; pie charts for >4 categories; glow neon; Inter on chart text if controllable; watermarked discs behind plots.
- **Gold:** `brand/examples/` (letter + CGM deck + dataviz thumbs) and box `/workspace/outputs/bytestream-creator-metrics/2026-09-25/latex/figures/`.

## C. Production sequence (ops)

1. Paper → house LaTeX locks.
2. Chief (or named reviewer) → PASS / HOLD on locks + substance.
3. Deck → true 16:9 only after paper PASS.
4. Chief → PASS / HOLD on deck chrome + fidelity.
5. Remirror only after PASS: Downloads `…-LATEX.pdf` + `…-DECK.pdf`; Atlas `Main/01 Projects/Atlas/reports/YYYY-MM-DD-<slug>/`.

Chief must approve before remirror.

## Method-theater ban (Jake 2026-09-25)

Never stage research-process instructions as findings, abstract bullets, slide kickers, or closing morals.

Banned examples: “Where numbers are unpublished, the honest research move is to say so.” / “Where public numbers are missing, this paper says so rather than inventing TAM or CPM series.” / “We do not invent figures” as a takeaway.

Hygiene is the default. One dry Methods clause is enough if needed; prefer silence. Missing numbers: note once beside the claim — do not elevate the rule itself.

See `jake-house-writing` → Method-theater / instruction rehash.

## Hard stops

- Geist on research letter papers
- `\@maketitle` overrides / title-size hacks / orphan title lines
- Double folios or `counter(page)` stacked chrome
- A4 landscape pretending to be 16:9
- Giant glow logo on content slides
- Consultant slogan H1s, invented stats, Treg/spend on slides or in abstract/body
- Bytestream product-recommendation framing on paper or deck, unless Jake commissioned a proposal paper (2026-09-28: Bytestream is in its pivot stage and nothing is locked; proposal papers with impacts and targets are allowed when he asks, and a design-mocks doc is built from the paper after it)
- Metaphorical “gate”, prime-contrast punch, marketing gerunds, quantity headlines, ops-instruction bleed, **method-theater / instruction rehash** on slides or in abstracts (see `jake-house-writing`)
- Overlapping chart labels; rainbow fills; AI fake charts; redrawn figures with different numbers
- Remirror before Chief PASS
- Method-theater findings (“honest research move is to say so,” “rather than inventing TAM/CPM”)
- Inventing a third look instead of `brand-documents` or this skill

## Install (agents outside Grok Bot)

Start with the public starter kit: https://brand.agentsinmotion.ai/research/starter-kit/starter-kit.zip (how-to: https://brand.agentsinmotion.ai/research/starter-kit/README.md).

Then copy:

- `Main/01 Projects/In Motion Automation/skills/ima-research-documents/`
- `Main/01 Projects/In Motion Automation/skills/brand-documents/`
- Pull `Main/01 Projects/In Motion Automation/brand/` kit (incl. `examples/`)

Grok Bot agents: Read this skill + `jake-house-writing` + `jake-research-papers`, then open Mac `brand/GUIDE.md` research + Dataviz section.