§ 01 · Cover · Manifesto

Brand Guidelines · Vol. 01

«CONTEXT JAMMING»

Field notes from inside the context window.

Context Jamming is a first-person research desk for the model era. We orchestrate Claude and Gemini as a mixture of expert agents, publish what we find, and treat the workflow as the artifact. This page is how we look, sound, and show up.

§ 02 · The Brand In One Paragraph

The elevator paragraph (deck-ready).

Context Jamming is the editorial publication of ACRA Insight LLC — a single-operator research desk run by Bret Kerr that orchestrates Claude and Gemini as a mixture of expert agents and publishes the results as long-form dispatches, founder profiles, and field notes. The work is first-person, materially specific, and built on primary sources. Sections include Dispatches, Founder Files, Stupid LLM Tricks, the GemClaw Debate Society, and the Transformed by Agentic Intelligence podcast.

Drop-in copy for partner decks and press intros. For shorter or longer variants, see § Press & Boilerplate below.

§ 04 · Color System

The palette.

Five named tokens carry the full system. Each token gets a full OKLCH-derived 50–900 tonal scale on confirmation; the placeholder swatches below use prior-work approximations.

Signal Amber

#c47a4a

Primary · CTAs · cube glyph · accent rules

[CONFIRM] Prior work referenced amber/orange primary; swatch shown is the current --color-copper. Confirm hex / RGB / HSL / OKLCH / Pantone-nearest.

Circuit Blue

#5b8faf

Secondary · waveform glyph · link states

[CONFIRM] Prior work referenced a mid-blue secondary. Confirm hex and full system.

Terminal Graphite

#0d0d0d

Primary background · body containers (dark theme)

[CONFIRM] Currently --color-chalkboard. Confirm exact value + named role.

Paper

#f5f0e8

Light surfaces · print · light-theme fallback

[CONFIRM] Currently --color-chalk / --color-cream. Confirm canonical name + value.

Caution Yellow

#000000

Stupid LLM Tricks accent only — do not mix with Signal Amber

[CONFIRM] No prior swatch. Suggest a true workshop yellow (e.g. #f4c11a) — confirm or override.

Success / Live

#000000

LIVE indicator · status pills

[CONFIRM] No prior swatch. Suggest a muted phosphor green. Confirm.

Pairings that work

  • Signal Amber on Terminal Graphite — primary CTA on dark.
  • Paper on Terminal Graphite — body text on dark.
  • Circuit Blue accents on Paper — link states on light.

Pairings that don’t

  • Signal Amber on Paper for body text — fails AA contrast.
  • Circuit Blue on Signal Amber — vibrating, hard to read.
  • Caution Yellow on Paper without a Terminal Graphite container — reads as warning chrome.

Accessibility

[CONFIRM · color-a11y] AA / AAA contrast ratios for every text-on-background combination once final hexes are confirmed.The page will render a contrast matrix table here once the palette is locked.

§ 05 · Typography

The type stack.

Display / wordmark

Helvetica Neue Black(system) with Arial Black, system-ui, sans-serif fallbacks. Used only for the «CONTEXT JAMMING» wordmark and hero headlines — not for body, never for nav.
[CONFIRM · display-face] Switch to Archivo Black via @next/font/google for cross-platform consistency?Currently a system stack. Archivo Black would lock the rendering on Windows/Linux but adds a font load.

Editorial hero serif

Playfair Display (variable, via next/font). Used for italic taglines, the colophon lead, and editorial page H1s.

Body

Fraunces (variable · opsz + SOFT axes, via next/font) for long-form body. Playfair Displayalso serves italic body emphasis.

Mono

IBM Plex Mono (next/font) for editorial metadata, the math overlays (H(X) = -Σ p(x) log p(x), I(X;Y), D_KL), and section kickers. Geist Mono(next/font) for utility / code blocks.

UI / nav

IBM Plex Mono uppercase, tracking 0.3emfor nav rails and section labels. The contrast between the heavy sans-serif wordmark and the tracked-out mono nav is intentional — it gives the masthead its “research notebook” feel.

Caveat (marginalia)

Caveat (next/font) for hand-lettered annotations and grease-pencil corrections. Use sparingly.

Type scale

RoleSize (rem)Line-heightTracking
H13.751.05−0.02em
H22.51.10−0.015em
H31.8751.20−0.01em
H41.51.300
H51.251.400
H6 (small caps)11.500.01em
Body L1.1251.600
Body M11.600
Body S0.8751.500
Caption / mono0.81251.400.02em

The french-quote treatment

The wordmark is always set as « CONTEXT JAMMING » — the french-quote glyphs are part of the mark, not punctuation. Reserve this treatment for: (1) the wordmark in hero contexts, (2) editorial pull-quotes that quote the publication itself by name, (3) the lockup in partner decks. Do not use for arbitrary editorial quotation.
[CONFIRM · french-quotes-scope] Confirm: keep the «...» treatment scoped to the wordmark only, or expand to a recurring editorial device?

§ 06 · Voice & Tone

How we sound.

Context Jamming writes in the Michael Lewis × arXivlane: narrative spine of a feature writer, primary-source density of a research preprint. Three voice attributes carry the load:

Insider

Do: “Olah’s Distill thread predicted the SAE wave before SAEs existed.”

Don’t: “Researchers have made progress on interpretability.”

Generous

Do: “Here is the prompt, the chain, and the failure case — replicate it.”

Don’t: “Our proprietary method delivers results.”

Materially specific

Do: “Gemini 3.1 Pro returned 43 sources in 15 min; Claude Opus 4.7 returned 2,501 in 48 min.”

Don’t: “The two models gave different answers.”

How we handle uncertainty

We caveat. We do not hedge into mush. “The data is consistent with X; it does not establish X” is good. “X may or may not be true” is bad — that’s not a caveat, that’s a refusal to commit.

How we handle other models

We name them. We do not subtweet. If Gemini caught something Claude missed, the dispatch says so by name with a link to the run.

Forbidden moves

  • Corporate-speak: “leverage,” “solutions,” “ecosystem play.”
  • The verb delve. Ever.
  • The em-dash-as-pause overuse pattern that ChatGPT made into a tell.
  • Fake humility (“I’m just an analyst, but…”) and fake certainty (“The definitive guide to…”).
  • Listicles without a thesis. We write essays that happen to have lists in them.

§ 07 · Sub-brands

The family.

Context Jamming is the parent. Five sub-brands sit under it. Each gets its own visual register but inherits the voice and the editorial standard. Lead with the parent brand for partner decks and press; lead with the sub-brand for the editorial product itself.
[CONFIRM · slt-treatment] Treat Stupid LLM Tricks as a peer sub-brand inside this guide, or spin it off into its own /brand at /tricks/brand?Default below: peer sub-brand. Funnel-tight. Switching to standalone gives SLT independent sponsorship velocity but doubles maintenance.

Stupid LLM Tricks™

Short-form workshop column documenting funny, surprising, or revealing model behavior. Looser register than the parent; more grease-pencil, more screenshots, more yelling.

PALETTE
Caution Yellow accent + Terminal Graphite ground
WHEN
Use when the artifact is a single discovery or a single failure case.

The GemClaw Debate Society

The Oxford-style AI debate engine. Two models, three rounds, one proposition, public verdict. Inspired by Irving, Christiano & Amodei (2018).

PALETTE
Oxford red + Cream + Stone-900 (preserved from the original GemClaw chrome)
WHEN
Use for any artifact that runs through the debate arena, the chamber, or the record books.

Founder Files

Long-form editorial profiles of operators, researchers, and theorists. Format spec: ~3-5k words, 4-6 sections, hero portrait, primary-source citation density.

PALETTE
Inherits parent palette; portrait treatment varies per profile (line-art, flicker, photo, or monogram).
WHEN
Use for any profile that pushes past 1,500 words and uses primary-source interviews or document analysis.

Dispatches

The polemical column. Long-form essays on AI, media, and infrastructure. Numbered sequentially (Dispatch N°NNN). Often syndicated to a per-dispatch subdomain.

PALETTE
Inherits parent; per-dispatch can introduce a feature color.
WHEN
Use for argument-driven pieces — pieces with a thesis the editor would defend in print.

Transformed by Agentic Intelligence (Podcast)

Long-form video conversations on what changes when the unit of work becomes a multi-model loop. Channel: Context Jamming.

PALETTE
Inherits parent; podcast cover art uses a tighter chalk-and-copper composition.
WHEN
Use for any audio/video conversation released through the channel.

§ 08 · Visual System

The motifs.

The math overlay

The chalkboard hero pattern — H(X), I(X;Y), D_KL, argmax, ∂L/∂θ, “AI Safety via Debate — Irving 2018” — is an intentional motif, not decoration. It signals that this is a research-desk publication, not a magazine. Use sparingly: the home masthead, brand-guide cover, and any piece whose argument hinges on the math itself.
[CONFIRM · math-overlay-set] Lock the formula set, or rotate per dispatch?Currently rotates: each masthead picks formulas from a fixed pool. Could be tightened to a canonical seven.

Infographic conventions

  • 9:16 vertical as the canonical infographic format (Story / Twitter share-card friendly).
  • Richard Scarry density — many small things, all individually labeled, no whitespace anxiety.
  • Paula Scher typography — oversized type that fills its container; tight tracking; no cowardice.
  • SPY Magazine sidebars — numbered, sourced, irreverent, always footnoted.

Photography & illustration

Direction: “Anthropic meets Palantir.”Cool light, hard edges, technical primary sources rendered as objects. Avoid stock photography. Avoid AI-generated photorealism unless the dispatch is about AI-generated photorealism.

The three-glyph hero motif

The cube · waveform · robot trio is a sub-brand decorative motif — it is not part of the official logo system. Read left-to-right:

  • Cube — the context graph. The accumulating knowledge layer.
  • Waveform — the signal. What the human and the models exchange.
  • Robot — the agent. The model in the loop.

Use as a hero accent on the home page and the podcast section. Do not use in place of the wordmark in lockups, press kits, or partner decks.

§ 10 · Press & Boilerplate

Drop-in copy and headshots.

Org boilerplate · 50 words

Context Jamming is an independent AI research publication run by Bret Kerr at ACRA Insight LLC. We orchestrate Claude and Gemini as a mixture of expert agents and publish the results as long-form dispatches, founder profiles, and a podcast. Field notes from inside the context window.

Org boilerplate · 100 words

[CONFIRM · org-boilerplate-100] Approve the proposed 100-word org boilerplate.Draft should reference: ACRA Insight LLC, MoEA orchestration, Triple Transformation methodology, Anthropic + Google DeepMind ecosystem, the editorial sections (Dispatches / Founder Files / Stupid LLM Tricks / Debate Society / Podcast).

Org boilerplate · 250 words

[CONFIRM · org-boilerplate-250] Approve the proposed 250-word org boilerplate.

Founder bio (Bret Kerr) · 50 words

Bret Kerr is the founder of ACRA Insight LLC and the operator behind Context Jamming. He runs a single-operator research desk that orchestrates Claude and Gemini as a mixture of expert agents. Prior: Mimecast (through IPO and the Permira acquisition).

Founder bio · 100 words

[CONFIRM · bio-100] Approve the proposed 100-word founder bio.Reference: ACRA Insight LLC, Triple Transformation methodology, MoEA orchestration, prior Mimecast tenure (through IPO and Permira acquisition), the Context Jamming research desk. Do NOT include Depakote / diagnostic history.

Founder bio · 250 words

[CONFIRM · bio-250] Approve the proposed 250-word founder bio.

Headshots

[CONFIRM · headshots] Final headshot files (color, b&w, print resolutions).Drop into /public/brand/headshots/ once provided.

Logo pack

§ 11 · Contact

Where to send what.

Sponsorships

bret.kerr@gmail.com

[CONFIRM] Suggested: sponsors@contextjamming.com — confirm whether to create + alias to existing inbox or stand up dedicated.

Press

bret.kerr@gmail.com

[CONFIRM] Suggested: press@contextjamming.com — confirm.

Editorial / general

bret.kerr@gmail.com

[CONFIRM] Suggested: editorial@contextjamming.com or hello@contextjamming.com — confirm.

DMCA / legal

bret.kerr@gmail.com

[CONFIRM] Required if running sponsored content. Suggested: legal@contextjamming.com.

[CONFIRM · email-routing] Stand up dedicated inboxes (sponsors@, press@, editorial@, legal@) or alias them all to bret.kerr@gmail.com for now?

§ · Invoice No. 001 · The Build Ledger

The Ledger.

Filed · contextjamming.com

What a conservative mid-market digital agency would have quoted for the same scope, itemized against what this site actually cost. Agency numbers are the floor — not the premium brand-studio tier.

TIME

12 weeks

2 days

~42× faster

COST

~$150,000

~$300

~500× cheaper

TEAM

5-person agency

1 human + 3 models

Same deliverable

§ Itemized — what a mid-market agency SOW would have billed

Discovery · brand positioning · workshops40–80 hr$10,000
Design system · Figma tokens · 3 rounds60–120 hr$18,000
Wavesurfer audio carousel · single-track context60–100 hr$16,000
Dual lightbox systems · focus trap · keyboard30–50 hr$8,000
LLM product flows · streaming · state machine80–160 hr$26,000
Stripe · checkout · webhooks · env hardening40–80 hr$10,000
Editorial routes · 6 sub-pages · templates60–100 hr$14,000
Accessibility pass · aria · reduced-motion40–80 hr$10,000
QA · cross-browser · mobile matrix60–100 hr$14,000
Cross-publication rebrand · masthead + IA · 2026-04-2820–40 hr$6,000
Subtotal~700 hr$126,000
Project management · 18% overhead$24,000
Agency total — conservative floor~700 hr~$150,000
Actually spent · Claude + Gemini stack~20 hr~$300

Agency figure assumes ~700 billable hours at $200/hr blended, plus ~18% PM overhead — the conservative floor of a mid-market SOW. Premium brand studios would have quoted 2–3× that. Stack: Claude Code 4.7 Max, Claude Opus 4.6, Gemini 3.1 Pro, Vercel Pro.

§   Colophon

How this site is made.

Vol. 26 · build log

Every page on contextjamming.com is the output of a real-time, three-body Mixture-of-Experts loop. One model orchestrates. Two consult. The human holds the thesis. No single model commits alone.

Orchestrator

Claude Code 4.7

1M context · Max tier

  • Primary author
  • Terminal-native, direct push to Vercel
  • Audit trail to GitHub on every commit
  • Adaptive thinking · effort: extra-high

Auditor

Claude Opus 4.6

1M context

  • Editorial critic
  • Code review before merge
  • Backup-of-record
  • Co-signs every commit

Adversary

Gemini 3.1 Pro

Cross-model MoE

  • Factual adjudication
  • Structural dissent
  • Deep Research → semantic triples
  • Caught the Donelan incident

Stack

Next.js
16.2 · App Router
React
19.2
TypeScript
5
Tailwind
v4 · @theme inline
framer-motion
transitions
wavesurfer.js
audio waveforms
marked
MD → HTML at build
fast-xml-parser
RSS + Atom

Typeset in

Fraunces
variable · opsz + SOFT
Playfair Display
debate display
IBM Plex Mono
editorial metadata
Geist Mono
utility mono
Caveat
grease-pencil marginalia
All via
next/font/google
Palette
single @theme block
No dupe tokens
ever

Infrastructure

Deploy
Vercel Edge Network
ISR
30-min revalidate · wire + notebook
Repo
github.com/BretKerrAI/founderfile
Branch
hero-redesign-library
Analytics
Google Tag Manager
Apex
contextjamming.com
Runtime
Node 24
Build tool
Turbopack
       human intent
            │
            ▼
   ┌────────────────────┐         ┌─────────────────┐
   │  Claude Code 4.7   │  ◄────► │  Claude Opus 4.6 │      ← auditor loop
   │    (orchestrator)  │         │     (auditor)   │
   └─────────┬──────────┘         └─────────────────┘
             │  ◄───────────┐
             ▼              │
       ┌──────────┐    ┌────┴───────┐
       │  Vercel  │    │ Gemini 3.1 │          ← adversarial loop
       │  (edge)  │    │    Pro     │
       └─────┬────┘    └────────────┘
             │
             ▼
       contextjamming.com
             │
             ▼
       ┌──────────────┐
       │   Git push   │         ← audit trail
       └──────────────┘
Assembled on Mac in Terminal · Filed from Franklin, MAContext Jamming · ACRA Insight LLC · MIT License · FounderFile.ai · RelationalIntelligence.xyz · Commission a Dispatch →