This is the space where everything lives — strategy, decisions, working files, what's next. Updated as we go, so there's one place to point at instead of ten. Hover any tile to peek; click to open it.
References · live working files
01 · Frame
What CAIS Federal is.
Ask three people what CAIS Federal does and you should get three different answers in the same language. The frame below is the language.
Mission
CAIS Federal ensures AI serves the national interest by connecting technical AI expertise to the language and processes of national security.
Core Belief
AI adoption and US strategic advantage are not competing goals. They are mutually reinforcing. Accelerating AI capabilities does not require accelerating risk.
The translation layer
CAIS Federal is not an advocacy organization. It is a translation layer between two ecosystems that currently don't speak to each other: the AI labs (Scale AI, APL, RAND, CSBA, the FFRDC world) on one side, the national security enterprise (DoD, IC, OSD, the operators) on the other. The site sells the translation, not the charter.
What we don't say
"AI safety" is in the org name. It is never the dominant verb on the site.
No AGI-doom, no "mutually assured AI destruction," no anti-acceleration.
No CAIS-parent Reddit-bot / YouTuber-sponsorship aesthetic. Reputational distance is the brief.
No Signalform-internal jargon — no "Narrative Architecture," no "vibe coding."
Verbs we use: translate, validate, convene, equip, align, accelerate.
02 · Audiences
Who we serve.
Three distinct groups. Same mission, different language per audience. Each gets its own sub-page on the site and its own capabilities document.
01 / Operators
DoD & Intelligence Community
Warfighters and decision-makers. PNR/OSD, CIA, the operational backbone. They need AI capability translated into operational language and implementation frameworks they can actually deploy — not warnings, not philosophy.
02 / Builders
AI Labs & Industry
Scale AI, APL, RAND, CSBA, the commercial AI vendors. They have capability but no neutral third-party translator who can validate their claims to government in language that lands. We provide bespoke validation before a proposal reaches the Pentagon.
03 / Network
Federal Partners
Think tanks, FFRDCs, academic programs (George Mason IR), nonprofit research groups. The translation network upstream of operators — the people who shape the policy frame the operators work inside.
Why three, not one
A Hill staffer skim, a DoD operator, a Scale AI engineer, and a CAIS parent leadership skim all need to read the site without feeling addressed at the expense of the others. Three doors, one house.
03 · Model
Research → Application → Adoption.
The three-phase operating model. Already the Chiron approach-page model — you validated it on the 4/14 call — so it carries straight into the CAIS Federal site.
Research
Listening. What is the lab actually doing? What is the government actually trying to solve? Hearing both sides clearly enough to translate. Without this phase, everything downstream is invented.
Application
Framing what the labs are building so it makes sense to engineers, policymakers, and operators simultaneously. The same capability, described three ways, each one landing in its audience.
Adoption
Moving standards from paper to practice. Pilots. Convenings. Shared protocols. RFPs written in the language labs actually speak. This is where the work becomes visible outside the translation room.
Language note
"Bespoke" is your word. Use it on the Approach page when describing outputs — bespoke national security intelligence, bespoke AI contextualization. It reads as craft, not mass-market.
05 · Outputs
The output program.
The site sells a set of deliverables. Those deliverables are what retainer-phase work actually produces. Locking the output stack now so the site can reference it concretely.
Entry point
White Papers
Fast to produce, respected in the space. First paper: a net-assessment demo using Scale AI as the subject. Proves the format and seeds the Scale relationship.
Flagship
Net Assessments
Neutral third-party validation of lab proposals to government. The thing Scale AI and APL actually want from us.
Per-audience
Capabilities Docs
Bespoke versions tuned to DoD/IC, labs, and federal partners. Same spine, different language.
Upstream
Trainings & Curricula
George Mason IR curriculum in AI & national security. Expert-talks model, low curriculum burden.
Relational
Convenings & Wargames
Aaron's specialty. Structured exercises bringing labs, government, and partners into the same room with a common frame.
Influence
RFP Alignment
Help government write RFPs that actually elicit the right capability from labs. Fixes the procurement translation gap directly.
Near-term
White papers are the immediate next deliverable after site launch. The Scale AI net-assessment pilot gives us a live artifact to point at before any retainer month closes.
06 · LinkedIn
LinkedIn program.
A live surface before the Devin announcement lands, and a steady cadence after. You write. I supply the prompt and the frame each week.
Launch
Announcement post
Drafted 48 hours before Devin's announcement. Your voice, mission line anchor, link to site. Coordinated with Devin's post so both land together.
Cadence
Two posts per week
One substantive, one relational. Rotating pillars below. Prompts arrive weekly in the repo; you write and post.
Content pillars
Translation stories — a concrete moment where labs and government talked past each other, and what the bridge looks like.
Audience insights — what DoD operators actually ask for vs. what labs think they want.
Policy cuts — one news item a week, reframed through the CAIS Federal lens (adoption + advantage).
Partner spotlights — Scale, APL, CSBA, GMU. Relationship building in public.
How it lives in the repo
Each post has its own file under linkedin/weekly/. Prompt, draft, final copy, and the image or link card. You review in Claude Code, tweak, post. Post-history becomes searchable context for the next draft.
04 · Website
The website.
Reskin and recontent the Chiron build rather than start from zero. Same architecture, same animation system, new brand, new copy. This is the only path that launches on deadline.
Page map
Page
Purpose
Lift
index
Hero · mission · core belief · three audiences · CTA. The one-page version of everything.
Heavy
challenge
Why this exists now. Reframed around "adoption ≠ accelerating risk."
Heavy
approach
Research → Application → Adoption. Outputs: papers, assessments, docs, convenings.
Light
audiences
The grid: DoD/IC · Labs · Federal Partners. Each links to a sub-page.
Heavy
audience sub-pages (×3)
Diagnostic → Deliverables → Mission Impact. Same template, different content.
Heavy
team
Susan · Ike · Aaron · (Devin?). Bios, photos, roles.
Medium
letter
Founder's Letter. Ike's Chiron letter likely survives with edits.
Light
contact
Form · email · short page header.
Light
Homepage, at a glance
Illustrative. The real build inherits the Chiron animation system and the braid motion — recolored to CAIS Federal palette.
caisfederal.org
CAIS Federalchallenge · approach · audiences · team · contact
Translation between AI capability and national security
AI adoption and strategic advantage are not competing goals.
CAIS Federal ensures AI serves the national interest by connecting technical AI expertise to the language and processes of national security.
We translate between the labs building frontier AI and the agencies responsible for deploying it. Research. Application. Adoption. Paper to practice.
01
DoD & Intelligence Community
02
AI Labs & Industry
03
Federal Partners
Visual system
Navy primary — deeper, more institutional than Chiron (#0c1828).
CAIS bridge accent — the CAIS parent light blue, used as a connective gesture. Homage, not affiliation.
Vital accent — warm gold (recommended for DoD credibility) or deep cyan. Side-by-side today.
Braid animation — Chiron's signature, recolored. Literal metaphor for translation: chaotic inputs on the left resolve into coherent outputs on the right through the middle.
Signal-red is dropped entirely. Politically and visually wrong for CAIS Federal.
07 · Infrastructure
How we work.
This is the part we actually need to settle today. Two weeks to launch. We can't afford a workflow that depends on syncing up.
The setup: a single shared repository is the source of truth. The website lives there. The strategy docs live there. Every decision we lock lives there. You get a copy on your machine, and you work against it directly — same way I do.
The repository
One GitHub repo, private, shared between us. It contains every working file for CAIS Federal: website source, specs, PRD, decisions log, copy blueprints, documents like this one, and the output-program drafts as they're written.
cais-federal/
├── CLAUDE.md# project brain — context for AI tools
├── README.md# how to work in this repo
├── website/
│ ├── PRD.md # master build plan — already complete
│ ├── decisions-log.md # locked vs. provisional, always current
│ ├── specs/ # page-by-page build specs
│ ├── copy-source/ # validated quotes + draft copy
│ └── build/ # the live site source
├── outputs/
│ ├── white-papers/ # Scale AI pilot + future papers
│ ├── net-assessments/
│ └── capabilities-docs/
├── linkedin/
│ ├── announcement.md
│ └── weekly/ # prompts + drafts, one file per post
├── blueprints/ # docs like this one
└── assets/ # brand · logos · photos · references
Claude Code — the experiment
Instead of chatting with an AI in a browser and pasting files back and forth, you run Claude Code directly against the repo on your laptop. Same tool I'm using right now to build this document. It reads the whole project, understands the PRD, and can edit files, draft copy, or answer questions in full context.
The practical difference: when you ask "what's our position on X?" it reads the decisions log and validated quotes and answers from the actual locked language — not a summary you fed it. When you draft a LinkedIn post, it reads the content pillars and your voice samples first. No re-briefing every time.
I'll set it up with you live on a short follow-up — under 20 minutes. It runs on your personal Mac (your preference over the CAIS device for exactly this reason: no CAIS servers in the loop).
Our working loop
01 · You
Input
Decision, reaction, new material, a link, a voice memo.
02 · Me
Build
Commit to the repo. Copy, design, pages, docs. Visible as it happens.
03 · You
React
Claude Code or a staging URL. Agree, disagree, amend.
04 · Both
Ship
Merge. Decisions log updated. Never twice on the same point.
Supporting surfaces
The portal (cais-portal.pages.dev) — at-a-glance status, tasks, timeline, hours, documents. Stays live. You don't need to work in it, but when you want a dashboard view it's there.
Google Drive — for material CAIS parent has sent (brand guides, Dan's template, workshop transcript). I mirror references into the repo as we use them.
Calls — one standing working session per week, plus short follow-ups as needed. Batched reviews, not per-section approval.
08 · Timeline
Two weeks to launch.
Working backwards from April 30 — Devin's announcement. Eleven business days. Days not hours, because the bottleneck is review cycles, not time at the keyboard.
This week · April 14–20
Strategy locked. PRD complete. Repository set up and shared. Claude Code on your machine. Brand guide extracted; logo, palette, type locked. Migration kicks off Friday.
Now
Next week · April 20–24
Homepage live on staging by EOD 4/20. Core pages (challenge, approach, audiences, team, letter, contact) built out. Announcement post drafted. First staging review.
Build
Week of April 27
Audience sub-pages. Revision batch applied. LinkedIn About live. QA pass (content, visual, performance, reputational risk). Scale AI white paper outlined.
Polish
April 30
Live at caisfederal.[tld]. Coordinated with Devin's announcement. LinkedIn announcement post goes out. Phase 2 retainer begins May 1.
Launch
09 · Decisions
What we lock today.
Ten items still marked provisional in the decisions log. Closing these unblocks the next forty-eight hours of production.
Launch date in writing.April 30, 2026 confirmed for Devin's announcement?
Domain.caisfederal.org or caisfederal.us? Who registers, and when?
Logo direction.Adapted CAIS parent logomark (recommended — homage play) or a distinct mark. Two options side-by-side Friday.
Vital accent color.Warm gold (recommended for DoD/Hill credibility) or deep cyan. Decide today so the system can move.
Typography.Keep Cormorant serif, or swap to a more institutional serif (Source Serif 4, Tiempos)?
Founder's Letter.Ike's Chiron letter survives with edits, or full rewrite? If rewrite, push to retainer.
Team content.Aaron surname + bio + photo. Susan, Ike, Devin photos. Anyone else on the team page?
Dan's 92-page template.Access so we can fold specific gestures into the aesthetic. Homage by design, not by argument.
Audience sub-pages in v1.Ship all three at launch, or ship one canonical sub-page + "deeper detail coming" holds for the other two?
Engagement scope forward.Retainer start confirmed for May 1 at $7,500/mo?
Reference · PRD
Product Requirements Doc
The master build plan for the CAIS Federal site. Locked direction, page-by-page specs, sequence, QA. Lives at website/PRD.md in the repo — this is a read view of the same file.
The one-paragraph version
The CAIS Federal website is a reskin and recontent of the existing Chiron website build. The page set, animation library, design system architecture, and component patterns all carry over. What changes: brand tokens (color, type, logo), copy on every page, the "Use Cases" grid becomes an "Audiences" grid (DoD/IC, Labs, Federal Partners), and Chiron's signal-red is removed entirely. The braid animation — Chiron's signature visual — survives because it is a perfect literal metaphor for CAIS Federal's actual product (translation between AI labs and the national security enterprise). Launch target: end of April 2026 to support Devin Kim's announcement.
Build philosophy · three rules
Rule 1 — Preserve the Chiron architecture exactly.
Page set, navigation, section order, animation choreography, scroll behavior, component patterns — all preserved. If you want to add a new page, restructure navigation, or reorder sections, stop and add it to the decisions log. The point of the reskin is speed; the point of speed is the launch deadline.
Rule 2 — Replace tokens, not styles.
All color, font, and brand changes happen through CSS custom properties in css/main.css. Do not hand-edit individual selectors to override colors. If a Chiron selector hardcodes a color outside the tokens, refactor it to use a token rather than overriding downstream.
Rule 3 — Copy is locked phrase + section brief.
Every section on every page falls into one of three buckets:
LOCKED phrase — direct quote from copy-source/validated-quotes.md. Verbatim or near-verbatim.
NEW draft — copy written for CAIS Federal. Page spec gives brief and (for homepage) full draft.
PORTABLE — Chiron copy that survives with light edits. Replace names, verify reads correctly, flag anything Chiron-specific.
Day 7–8 · 4/23–24 — Audience sub-pages (or punt to retainer v1.1).
Day 9 · 4/27 — Full staging walkthrough + batched revisions.
Day 10 · 4/28 — Apply revisions; LinkedIn About + announcement copy.
Day 11 · 4/29 — QA; DNS/Netlify; soft launch; final sign-off.
Day 12 · 4/30 — Live for Devin's announcement.
What success looks like
Live, distinct, professional site at its own domain.
Susan can announce CAIS Federal publicly without being blocked on creative.
Reads credibly to a Hill staffer, a DoD operator, a Scale AI engineer, and a CAIS parent leadership skim — without making any of them feel addressed at the expense of the others.
Structured so retainer-phase work extends it (more audience pages, case studies, white paper hosting, events) without architectural rework.
Working relationship that does not consume more than ~15 hours/week, supported by async cycles and batched reviews.
Full file — including per-page specs, global find/replace list, asset production checklist, QA checklist, and launch checklist — lives at website/PRD.md in the repo.
Reference · Decisions Log
Decisions Log
Running record of what's locked vs. what's still provisional. Updated every time a decision moves. Lives at website/decisions-log.md in the repo.
Strategy
LOCKED · Reskin + recontent of Chiron, not net-new. — 4/14
LOCKED · Same page set as Chiron (homepage, challenge, approach, audiences + 3 sub-pages, team, letter, contact). — 4/14
LOCKED · Use Cases → Audience Grid (DoD/IC, Labs, Federal Partners). — 4/14
LOCKED · No heavy copy editing service. Susan handles her own copy revisions this phase. — 4/14
Easter eggs
PROVISIONAL · Chiron's "Partner With Us" hover-3-seconds easter egg — keep the mechanic, recolor and re-purpose. Replace chironLattice → caisFederalLattice. — 4/14
PROVISIONAL · Chiron's vertical scroll stress test — keep with same recoloring rules. — 4/14
Today's call updates this file in real time. Every lock gets recorded against the date.
Reference · Validated Quotes
Validated Quotes
Your locked language. Load-bearing phrases that go on the site verbatim or near-verbatim. Sourced from the 4/14 call transcript. Lives at website/copy-source/validated-quotes.md.
The mission line · LOCKED
CAIS Federal ensures AI serves the national interest by connecting technical AI expertise to the language and processes of national security.
Your reaction: "It's good. That's accurate."
Use as: homepage hero subhead, mission statement in About/Team, footer description, LinkedIn About, meta descriptions.
The core belief · LOCKED
AI adoption and US strategic advantage are not competing goals. They are mutually reinforcing. Accelerating AI capabilities does not require accelerating risk.
Your reaction: "Yes. Yes. And that is directly from my document. I think that language is 100% — like I think that 100% aligns with where we want to be and where I want to take it."
Use as: homepage manifesto section, top of Challenge page, About page positioning paragraph.
The model · LOCKED
Research → Application → Adoption
Research — listening. What is the lab actually doing? What is the government actually trying to solve? Hearing both sides clearly enough to translate.
Application — framing what the labs are building so it makes sense to engineers, policymakers, and operators.
Adoption — moving standards from paper to practice. Pilots. Convening. Shared protocols.
Doubly validated: your endorsement on the call, and already the Chiron approach-page model.
The translation metaphor · LOCKED
CAIS Federal is a translation layer between AI labs (Scale AI, APL, RAND, CSBA, the FFRDC ecosystem) and the national security enterprise (DoD, IC, OSD, the operators).
Mike: "In order to translate anything you have to hear what's being said and then take that boop blah blah blah blah here's the translation." Susan: "You know, I think you're 100% right."
Use as: visual metaphor for the braid animation; conceptual frame for the homepage; explainer for what makes CAIS Federal different from CAIS parent and from the labs themselves.
"Paper to practice" · LOCKED
CAIS Federal has the ability to convene government leaders, technologists, and industry partners to move standards from paper to practice.
Your reaction: "Oh, from paper to practice. That's good."
Use as: tagline candidate; section header on the Approach page; CTA copy.
The three audiences · LOCKED
Department of Defense & Intelligence Community — the operators, the warfighters, the decision-makers. PNR/OSD, the CIA. They need AI capability translated into operational language and implementation frameworks they can actually deploy.
AI Labs & Industry — Scale AI, APL, RAND, CSBA, commercial AI vendors. They have capability but no neutral third-party translator who can validate their claims to the government in language that lands. CAIS Federal provides that bespoke validation.
Federal Partners — think tanks, FFRDCs, academic programs (George Mason IR), nonprofit research groups. The translation network. People who shape policy upstream of operators.
"Bespoke" · LOCKED
Your words: "those outputs are bespoke and we can use that word, right? Like bespoke national security intelligence, bespoke AI understanding and contextualization."
Use "bespoke" on the Approach page when describing outputs. It's your word. You want it.
What CAIS Federal delivers · the output stack
Capabilities documents tuned per audience — different doc per audience, similar but nuanced.
Net assessments — neutral third-party reviews of lab proposals, validated from a national security standpoint.
White papers — entry-level output. Mike: "they're easy to create... and people respect them in the world we're operating in." You: "I think that white papers will be the beginning."
Trainings / curricula — possibly including George Mason IR on AI and national security.
RFP alignment — helping government produce RFPs that actually elicit the right capability from labs.
What NOT to say
Do not center "AI safety" rhetoric. It's in the name (Dan Hendrycks held the line) but must not be the dominant frame. Reads anti-administration in DC and closes doors.
Do not echo CAIS parent's Reddit-bot / YouTuber sponsorship aesthetic. Reputational distance is the brief.
Do not lead with "AGI risk" or "mutually assured AI destruction" — team skepticism, forced pillar from CAIS parent.
Do not use Signalform-internal jargon: "Narrative Architecture," "vibe coding."
Do not say "anti-acceleration." Adoption and strategic advantage are aligned — that's the whole point.
Tone notes
Confident, direct, no fluff.
Speak in the audience's language — DoD operators, lab engineers, policy people.
Reads credibly to: Hill staffers, DoD/Pentagon, funders, press, AI industry.
Bureaucratic gravitas over startup energy.
Citations and partner names beat adjectives.
"Safety" is in the name but never in the verb. Verbs: translate, validate, convene, equip, align, accelerate.
Source: transcripts/041426_Susan and Mike_The Beginning.docx. Any addition to this file requires either a direct Susan quote or Susan confirmation of Mike's framing.
Signalform Studio · Narrative ArchitectureLast updated · April 16, 2026