Petabit Scale Brand Voice Guidelines

v1.2 · 2026-04-18 · living document · team-level source of truth · see Provenance at the bottom for generation details

Executive Summary

No hype, no false modesty, no gatekeeping, no corporate-speak — ever. That's the voice. The rest of this document explains what it sounds like, where it flexes, where it breaks, and how to carry it. Read the We Are / We Are Not table first; the rest deepens from there.

Petabit Scale is an opinionated engineering shop with a short, clear job: bring hyperscaler-grade expertise, strategy, and pricing to the unique challenges of operating networks at massive scale. The company architects, builds, and supplies large-scale network infrastructure — and it does so as a unified infrastructure partner, not as split consulting-and-distribution businesses.

The domain expertise runs deep in optical systems of every kind — including complex high-end coherent transmission (400ZR, 800ZR), long-haul fiber-optic networks, and the adjacent packet and datacenter infrastructure that depends on them. That expertise is paired with deep supply chain capability: we procure and deliver cutting-edge components in massive volumes — including hard-to-source SKUs — at moments when other channels can't. And we're nimble problem-solvers by default: plans change, customer buildouts hit unexpected blockers, suppliers miss dates — we move fast, adapt, and ship.

The brand voice is already living in our customer emails: direct, opinionated, peer-to-peer, generously technical, and allergic to marketing bluster. This guide codifies that voice so anyone writing for Petabit Scale can carry it — with the substance preserved and the sharpest edges calibrated for scale.

Content should read like a senior operator wrote it for another senior operator. Dense technical prose. Candor where candor is owed. Wit where wit earns its place. No hype, no false modesty, no gatekeeping, no corporate-speak — ever.

We Are / We Are Not

The foundational brand identity anchor. Voice is constant — it doesn't change by channel, audience, or season.

Voice Attributes Detail

1. Carrier-grade

• What it means: The prose meets the engineering bar actual carriers set — the rigor, specificity, and technical correctness a five-9s operator would demand of their own internal documentation. Authority comes from our team having run networks at that tier; the writing has to hold up under the same scrutiny as the gear.
• How it shows up: Concrete references to production, actual hardware, actual scale. Numbers before adjectives. Standards cited correctly. Specifics that only someone who's been in the gear could know — five-9s availability math, SR-class optical budgets, real line-rate numbers, actual lead times. Stories with details only someone who was there could write.
• What to avoid: Borrowed credibility ("industry best practices"), generic framing ("organizations like yours"), language that could have been written by someone who's never seen a rack or a cold DCI span. If a carrier's lead engineer would scoff at a line, cut it.
• Confidence: High

2. Candid

• What it means: We say the inconvenient thing — when a plan is flawed, a spec is wrong, a vendor is oversold — because that's what respect for a technical peer looks like.
• How it shows up: Strong pronouncements with evidence ("that architecture won't scale past 400G because..."). Opinions identified as opinions ("our take:"). Willingness to tell a prospect their preferred vendor picked the wrong optic.
• What to avoid: Judgmental adjectives without receipts ("bad idea," "dumb design") — bring the data. Also avoid false modesty ("I'm no expert, but...") — the customer wants the real opinion.
• Confidence: High

3. Independent

• What it means: We're not a captured reseller. Our architecture advice optimizes for the customer's outcome, not our BOM margin on any one line. Customers come to us partly because we can name the wrong part — including parts we sell.
• How it shows up: Vendor-neutral recommendations by default. Willingness to say "don't buy that, even from us." Public positions on standards and tech transitions that aren't tied to one OEM.
• What to avoid: Language that reads as pushing a specific vendor's line. Never stack a paragraph with a single manufacturer's model numbers when alternatives are viable.
• Confidence: High

4. Dense with substance

• What it means: Prose is precision-forward. Sentences pack specificity; paragraphs reward careful readers. Every claim has evidence or is framed as opinion. Nothing is on the page because it sounds good.
• How it shows up: Technical terms used correctly. Numbers, model specs, standards cited accurately. Longer sentences when complexity demands them — not hedged into simplicity.
• What to avoid: Filler nouns ("solutions"), empty superlatives, throat-clearing intros ("In today's fast-paced world..."). If a sentence can be deleted without loss, delete it.
• Confidence: High

5. Peer-facing

• What it means: We write operator-to-operator. The reader is assumed to be competent at their job. We don't ladder up to authority ("As industry leaders...") or ladder down ("For those new to networking..."). We meet readers where they are.
• How it shows up: No qualifying-checklist sales moves. No "are you aware that...". We assume the reader knows what 800G is; if they don't, we tell them once and move on.
• What to avoid: Language that implies the reader needs convincing of basic competence. Avoid "stakeholder," "decision-maker" as identifiers — those belong in vendor pitch decks, not in operator-to-operator prose.
• Confidence: High

6. Witty when warranted

• What it means: Dry engineer's humor is welcome — the understatement, the sharp observation, the occasional "absolutely terrible idea." Humor is a feature of substantive writing, never a substitute.
• How it shows up: Humor emerges from calling things plainly, from pointing out industry absurdity, from the aside that undercuts vendor-speak. Think: the kind of humor that lands at a NANOG lightning talk, not at a SaaS kickoff.
• What to avoid: Puns. Forced clever headlines. LinkedIn-thought-leader performance. Anything that would feel at home in a B2B SaaS email newsletter.
• Confidence: Medium-High

7. Formality-flexing

• What it means: Voice is constant; formality flexes by context. We default to semi-formal and move from there: up for proposals, legal, first-touch with procurement at a Fortune 500; down for peer technical threads, community posts, engineer-to-engineer.
• How it shows up: Contractions OK in most contexts (email, blog, website) but rare in proposals/SOWs. Personal openers ("Here's what we heard…") appropriate on customer emails and deck intros, not on a scope statement.
• What to avoid: Treating formality as a fixed setting. Formal-everywhere reads cold and committee-written; casual-everywhere reads unserious. The decision is per-artifact, not per-brand.
• Confidence: High

Brand Personality

• Archetype: The opinionated engineer — strong positions, cited from the receipts. Will tell you "that's a terrible idea" and then tell you why, with numbers. Frank, not cruel.
• If Petabit Scale were a person: The veteran infrastructure operator three beers into a NANOG afterparty — generous with expertise, allergic to performance, unafraid to disagree with the consensus, funny without trying, trustworthy because they've been burned by the things you're about to be burned by.
• Core values expressed in voice:
• Candor — say the inconvenient thing
• Precision & craft — facts over vibes, details others skip
• Independence — not captured by any one vendor
• Peer-respect (implicit) — operator-to-operator by default
• Real customer outcomes — solve real problems, deliver when no one else can, and save customers meaningful money along the way. The measure of our work is what the customer gets, not what we sell.
• What customers should feel after interacting with us:
• Reassured — "I'm in hands that have been there before"
• Educated — "I learned something, whether or not I bought"
• Not "challenged" — we stay opinionated, but the reader's takeaway should be relief, not being put in their place.

Not like these voices

Writers calibrate fastest against contrast. We are deliberately not:

If a draft sounds like any of the above, rewrite it.

When the rules bend

The seven attributes are defaults, not laws. Specific contexts where deliberate rule-breaking is correct:

• Density yields to legal precision. In a contract or SOW, clarity over compression. Long sentences are fine if they remove ambiguity.
• Candor yields to human moments. If a customer just lost data or missed a major deadline because of something we did, leading with "you shouldn't have designed it that way" is wrong even if true. Help first; post-mortem later.
• Peer-respect yields when the reader asked for teaching. Some audiences genuinely want a primer — a conference keynote, a recruit-facing post, a partner enablement session. Be a teacher, not a peer, if that's the ask.
• Wit yields to gravity. Outage post-mortems, compliance language, data-breach notifications: no humor, even dry.
• "We" yields to "I" when an individual is personally on the line. A personal apology or a direct-from-a-human recommendation uses "I" even on company surfaces.

If you break a rule, do it deliberately. The guide will tell you which rule you're breaking; you decide whether the moment warrants it.

Messaging Framework

Primary Value Proposition

Hyperscaler-grade expertise, strategy, and pricing — for the unique challenges of operating networks at massive scale.

Variations for different contexts:

• Website hero (plain-spoken): "We architect, build, and supply large-scale network infrastructure — built for the unique challenges of operating at massive scale."
• Peer intro (NANOG / LinkedIn): "The grade of engineering, strategy, and pricing access normally only found at the biggest hyperscalers — brought to operators who need it but don't operate at that scale."
• Stakes-forward (proposal cover): "Networking infrastructure for operators who can't afford to get it wrong."

Message Pillars

All four pillars anchor the messaging architecture. Each maps to a customer concern.

1. Carrier-Grade Expertise — The Trust Pillar

• Core idea: We operate at the engineering standard actual carriers set. Our team has run networks at that tier; our architecture advice and the gear we supply meet the reliability bar a five-9s operator would demand. Battle-tested, not theoretical.
• Domain depth: Optical systems of every kind — 400ZR, 800ZR, LR4 / FR4 short-reach optics, PAM4-based high-speed Ethernet, coherent transmission at the edge and the long-haul — plus the packet, EVPN-VXLAN fabric, and DC interconnect layers that sit alongside them.
• When to use: First-touch, credentialing sections, bios, about pages, pitch openers.
• Example phrasing: "We built networks like yours — not advised on them from a distance. Carrier-grade by default."
• Proof points: Our team's collective operator résumé (prior roles at hyperscaler-tier networks); published technical content (NANOG talks, technical posts); specific optical expertise (400ZR / 800ZR deployments, long-haul transmission, coherent-pluggable adoption); named customer references (with permission).

2. Vendor-Independent Architecture — The Integrity Pillar

• Core idea: We're not captured by any one OEM. You get the right design, not the one that pays us best.
• When to use: When a customer has been burned by a captured reseller; when competing with manufacturer direct sales; when the customer is shopping architecture advice.
• Example phrasing: "We'll tell you when the wrong part is the one we sell. That's the job."
• Proof points: Cross-vendor BOMs; public commentary on tech choices that crosses brand lines; willingness to decline a sale.

3. Hyperscaler Economics Without the Hyperscaler — The Value Pillar

• Core idea: You get pricing leverage and supply access that usually requires Fortune-100 purchasing power.
• Supply chain depth: We procure and deliver cutting-edge components in massive volumes — including the SKUs that are "on allocation" everywhere else, or that smaller channels can't source reliably. Direct manufacturer relationships, not three-hop distribution.
• When to use: Against distributor / reseller competition; when the customer's existing supply is overpriced, over-allocated, or impossible to get in volume; when economics or availability are the near-term trigger.
• Example phrasing: "Optics and components sourced directly from the manufacturers, at the grade of pricing normally reserved for the largest buyers — in volumes other channels can't commit to."
• Proof points: Actual price deltas (redacted); volume / SKU coverage; direct manufacturer relationships; history of fulfilling on tight timelines when competitors miss.

4. Architect, Build, Supply — The Execution Pillar

• Core idea: We don't hand off between advice and execution. Same team owns design through BOM through ship.
• Operational style: Nimble problem-solvers. When the plan changes — a supplier misses a date, a spec needs to shift, a customer hits an unexpected blocker — we move fast, adapt, and ship. Small team, short decision paths, no committee theater.
• When to use: Against SIs / VARs pitching one-stop-shop; when projects have tight or shifting timelines; when the customer has been bitten by handoff friction or by vendors who couldn't adapt mid-project.
• Example phrasing: "The team that designed it is the team that BOMs it. No handoff, no telephone-game specs. When the plan changes, we change with it."
• Proof points: End-to-end project histories; timeline comparisons; scope of past engagements; examples of recovering from supplier issues or mid-project scope shifts.

Pitch Ladder

10-second (airport test)

"Carrier-grade network infrastructure at hyperscaler economics. That's Petabit Scale."

30-second (NANOG hallway / intro call)

"Petabit Scale brings hyperscaler-grade expertise, strategy, and pricing to the unique challenges of operating networks at massive scale. We architect, build, and supply — vendor-independent, operator-led. Our team came up operating networks like yours at hyperscaler-tier scale; today we work with [named customers with confirmed permission — see Open Question #3]. The whole point is you get the architecture advice and the supply access that normally requires Fortune-100 buying power, from a team that's actually operated the gear."

2-minute (sales-context)

Opening: "Most large-scale operators have three pain points when they're building out: the advice from the OEM direct sales channel is captured by that OEM's roadmap; the distributor channel doesn't understand the architecture decisions; and the large SIs add layers of handoff between who designs it and who ships it.

Positioning: Petabit Scale was built to solve those three problems at once. We bring hyperscaler-grade operator expertise to the architecture decision, vendor-independent — we'll tell you when the wrong part is the one we sell. We supply the actual BOM at the grade of pricing normally reserved for the largest buyers, because we source directly from manufacturers in volumes most channels can't match. And the team that designs it is the team that ships it — no handoff.

Credentials: Our team has run networks at hyperscaler-tier operator scale for years. We came up operating the same kinds of infrastructure our customers are now building. Customers include [named customers — cite only those with confirmed permission; see Open Question #3].

Close: If you're building a large-scale network, we'll architect it right, supply it at real economics, and stand behind both. That's the company."

Competitive Positioning

Tone-by-Context Matrix

Voice is constant. Tone flexes on three dimensions: formality, energy, technical depth.

Context-Specific Guidelines

Website (home, about, services)

• Overall tone: Plain-spoken expert. Declarative, no marketing polish. Reads as if a person wrote it, not a committee.
• Opening approach: State what we do and who it's for in one sentence. No "In today's fast-paced world..." throat-clearing. Credentials arrive quickly after.
• Do's:
• Lead with concrete capability, not abstract benefit ("We architect, build, and supply..." not "We help organizations unlock...")
• Use specific customer names where permission exists
• Use technical vocabulary accurately; don't dumb down
• Don'ts:
• No "solutions" as a filler noun
• No stock imagery of people in headsets pointing at screens
• No "Let's unlock your potential" CTAs
• Example hero: "Petabit Scale architects, builds, and supplies large-scale network infrastructure. Hyperscaler-grade expertise, strategy, and pricing — for the unique challenges of operating networks at massive scale."

Customer Email — Technical (the source voice)

• Overall tone: Peer-to-peer, opinionated, generous with expertise. This is where the team voice is strongest in the existing corpus; this is what the guide scales to every writer.
• Opening approach: Straight to substance. No pleasantry padding. If there's context the customer needs, give it in one line and move on.
• Do's:
• State opinions plainly; include the reasoning
• Volunteer relevant advice beyond the immediate thread when it'll help
• Use contractions; write how you'd speak to a peer
• Don'ts:
• No "I hope this email finds you well"
• No hedging expertise ("we could be wrong, but...") — if we're unsure, say what we're unsure about
• No "circling back" or other sales-cadence vocabulary
• Example: "Short version: using QSFP-DD for 800G is a bad choice — the ecosystem is effectively frozen, and you'll be paying premiums for supply that's drying up. You want OSFP for 800G going forward. Here's what we'd spec instead for your fabric, given the switch silicon you already have: [specifics]."

Customer Email — Procurement

• Overall tone: Tight and transactional. SKU-fluent. Same voice, compressed.
• Opening approach: Lead with the answer. Quote, lead time, SKU list — whatever they asked for, at the top.
• Do's:
• Match their format — if they sent a table, reply with a table
• Use exact model numbers; include lead time and pricing notes cleanly
• If we're recommending a substitution, flag it clearly with one-line rationale
• Don'ts:
• No long narrative openers
• No reselling the relationship — they're buying, focus on the buy
• Example: Intentionally omitted — procurement voice is not yet canonical. Revisit once real reference threads establish the pattern; don't invent it here.

Community Content (NANOG, LinkedIn, blog)

• Overall tone: Substantive teacher. Long-form deep dives that help peers. Voice carries genuine wit and occasional sharp opinion.
• Opening approach: Stake the thesis early. Don't bury the lede under context; give it in the first paragraph and let the substance carry from there.
• Do's:
• Write for the engineer-peer, not for the marketing funnel
• Include actual numbers, configs, topologies — show the work
• Have an opinion; defend it with evidence
• Don'ts:
• No cross-posting marketing content as "thought leadership"
• No "3 reasons your network needs X" listicle-isms
• No feigned humility ("Just some thoughts from a random operator...")
• Example opening: "The 800G optics question is mostly settled — except for the people still arguing for QSFP-DD. Here's the specific reason you shouldn't pick QSFP-DD for new builds, and what the math looks like at realistic scale."

Proposals / SOWs

• Overall tone: Formal-technical throughout. Scope-tight, binding-precise. Voice stays ours — the substance is still carrier-grade and candid — but contractions and personal asides yield to legal cleanliness.
• Opening approach: A short framing section ("Scope & Approach") before the formal scope. State what we heard, what we're proposing, and why — in our voice — then transition to binding language.
• Do's:
• Scope each deliverable unambiguously
• State assumptions and dependencies explicitly
• Price discretely; make line items legible
• Don'ts:
• No vague adjectives in binding sections ("robust," "scalable," "enterprise-grade")
• No hedging scope with "includes but is not limited to" catch-alls
• No boilerplate preambles — the customer is reading to understand what we'll actually do
• Reference template: No canonical baseline yet. The 2018-era proposals (GIX, Pilot Fiber) are archived — they reflect a different business shape and should not be used as tonal reference. The proposal template refresh is the opportunity to establish a new baseline from first principles using this voice guide.

Apologies and owning misses

The candor attribute cuts both ways. When we are wrong — a missed ship date, a spec we got wrong, a vendor we recommended that underperformed — we own it the same way we'd name a customer's architecture mistake: directly, specifically, without ceremony or hedge.

• Overall tone: Direct, specific, non-defensive. The voice stays carrier-grade; the content changes from "here's what you should know" to "here's what we got wrong and what we're doing."
• Pattern: Name the specific miss → name the reason plainly → state what we're doing → state what we learned. No passive voice hiding agency.
• Do's:
• Lead with the specific thing that went wrong, not a framing paragraph
• Name the reason concretely — "our logistics person didn't flag the time-zone difference," not "process gaps"
• State what we're doing with a date or deliverable, not a promise to "work diligently"
• State the systemic fix briefly so the customer knows this won't repeat
• Don'ts:
• No "we regret to inform" or "out of an abundance of caution" corporate shields
• No passive-voice culpability-fog ("delays were encountered")
• No apology that's really a reassurance about our own process
• No over-apologizing — state it once, fix it, move on
• Example (shipping miss): "The MPO trunks we promised for Tuesday missed the factory cutoff by one day — our logistics person didn't flag the time-zone difference on the Taipei side. Shipping Wednesday via FedEx Priority Overnight; you'll have them Thursday morning. Going forward we'll hold a one-day buffer on any Asia-origin order with a hard US delivery date."

Three tests before you ship

A fast self-audit for any finished draft before it goes out:

1. Would a senior staff engineer at a Vultr, CoreWeave, IMC, or Coresite read past sentence two? If the open is throat-clearing, vendor-speak, or a "fast-paced world" intro — no.
2. Can any sentence be deleted without loss? If yes, delete it. Density is not padding plus compression; density is what survives compression.
3. Does any sentence pretend to know less than we do, or more? The first is false modesty; the second is overclaim. Both are off-brand.

If all three pass: ship. If any fails: fix that specifically before re-asking the test.

Terminology Guide

Canonical Brand Name

Vocabulary Conventions

• Use specific gear terminology by default: Say "400ZR," "800ZR," "QSFP-DD 800G-FR4," "LR4," "EVPN-VXLAN," "coherent optics" (ZR / ZR+), "long-haul fiber-optic transmission," "PAM4." The reader is a peer — precision is respectful, and our expertise lives in these specifics.
• Prefer "operator" and "hyperscaler" over "cloud" or "enterprise" when identifying customers: these are operational categories, not marketing ones.
• Avoid "AI" as a blanket noun: be specific — "training fabrics," "GPU clusters," "inference infrastructure," "AI compute operators." "AI" alone is imprecise and hype-adjacent.
• Expand acronyms on first use, even for technical readers: "Equal-Cost Multi-Path (ECMP)" on first mention; ECMP thereafter. Respects adjacent-expertise readers without slowing native ones.

Mirror Policy

When a customer uses a different term than ours: use our canonical term; note theirs parenthetically on first use only.

Example: "We'll spec direct-attach copper (DACs) for the ToR-to-server layer..." — thereafter, "DACs" is fine within the thread.

If their term is technically wrong in a way that matters, gently correct with the reasoning. If it doesn't matter, mirror them within the thread to avoid friction.

Inclusive language

The peer-facing attribute means assuming competence. It also means not assuming a specific subculture. A few defaults that prevent common misfires:

• Pronouns: default to they / them when the person is unknown; use a named person's own pronouns when known. Never assume.
• Collective terms: avoid "guys", "rockstars", "ninjas" — they read as a specific subculture our audience may not share, and they're not operator-native in the first place.
• Plain language where technical depth isn't the point: we write dense where density earns its place. We don't write dense for its own sake. A logistics update doesn't need BGP-adjacent vocabulary.
• Acronyms and insider shorthand: expand on first use (as above). Respects adjacent-expertise readers — procurement people, new hires, partners — without slowing native readers.
• Names of places, standards, and organizations: use them as they're written in their own authoritative source; don't invent stylings.

Avoid These Words

Use With Caution (not banned — but require justification)

Preferred Terms (use deliberately)

Language to Avoid (Anti-Patterns)

1. Vendor-Pitch Intros

• Bad: "In today's fast-paced, AI-driven infrastructure landscape, organizations are increasingly demanding scalable solutions that transform their digital operations."
• Better: "AI compute operators are buying 800G faster than the supply chain can credibly deliver. Here's how we handle that."

2. Hedged Expertise

• Bad: "We think it might be worth considering whether QSFP-DD is the right choice for your 800G deployment."
• Better: "QSFP-DD is a bad choice for new 800G buildouts. Here's why, and here's what we'd spec instead."

3. Corporate Superlatives

• Bad: "Our industry-leading, next-generation, enterprise-grade networking solutions deliver transformational outcomes at scale."
• Better: "We supply optics and cables directly from the manufacturers, at the grade of pricing normally reserved for the largest buyers — not through a three-hop distribution chain."

4. Passive-Voice Hedge

• Bad: *"It is believed that the architecture may present challenges at scale."

(Also avoid "I" in company-surface writing — use "we" for all team-facing communications. "I" is reserved for content an individual team member personally signs, like their own blog post or a one-on-one email they personally send.) - Better: "This architecture won't scale past 400G in a real AI training fabric. Specifically: [reason]."*

5. Sales-Cadence Vocabulary

• Bad: "Just circling back — wanted to touch base and align on the next step."
• Better: "Following up on the quote we sent — any questions on the spec or lead time?"

6. Apologetic Expertise

• Bad: "We could be wrong, but it seems like QSFP-DD might not be the best choice for 800G — though we're happy to discuss further."
• Better: "QSFP-DD is a bad choice for 800G. Here's why, and here's what to spec instead."
• Why: We have the expertise. Hedging into "we could be wrong" performs humility but reads as uncertainty. If we are unsure about a specific factor, name it. Otherwise, commit.

7. Borrowed Voice

• Bad: "Petabit Scale is transforming the way operators leverage next-generation AI infrastructure ecosystems." (This is an AWS release.)
• Bad: "Hey! Just shipped something rad. 🚀" (This is a Slack startup post.)
• Better: Both should sound like Petabit Scale — operator-to-operator, substance-first. Ask: whose voice is this? If the answer isn't "ours," rewrite.

8. Fake Intimacy

• Bad: "Hey team 👋 we're so excited to partner with y'all on this." (First-touch cold reply.)
• Better: "Following up on your RFQ — the spec looks straightforward. One clarifying question on port count before we quote."
• Why: Warmth is earned, not assumed. Peer-to-peer register on first contact is direct and specific, not casual-bro. Emojis, first names, and casualisms arrive once the relationship warrants them.

9. Stack-and-Punt Feature Lists

• Bad: "We offer architecture, design, procurement, strategy, consulting, turnkey delivery, optical supply, colocation expertise, fiber route planning, interconnection strategy, network automation, and operational support."
• Better: "We architect, build, and supply large-scale network infrastructure. The three verbs cover the whole scope — pick which one you need, we'll tell you how the others follow."
• Why: Stacking nine capabilities in one paragraph buries all of them. Pick the two or three that matter for this reader; let them ask about the rest.

10. Qualifying-Checkbox Sales

• Bad: "Happy to discuss — can you share your timeline, budget, and who the decision-makers are on your side?"
• Better: "Happy to discuss — what's the fabric you're trying to hit, and when do the racks land?"
• Why: BANT / MEDDIC-style qualifier questions read as sales-cadence and create vendor-to-target asymmetry. Technical, specific questions land as operator-to-operator.

Content Examples

Excellent Example — Customer Email (Technical)

"Short answer: don't spec QSFP-DD for your 800G fabric. The ecosystem is frozen — vendors are moving to OSFP for anything north of 400G, and you'll be paying premiums for optic supply that's drying up by the quarter. If you want the full reasoning, here's the specific issue: [technical detail]. Given the switch silicon you already have, we'd spec [specific SKUs] instead. Let us know the port count and we'll put a quote together."

What's on-brand: - Opinion stated plainly, with reasoning - "Here's the specific issue" — respects reader as peer, invites technical depth - No pleasantry padding - Specific gear recommendation, not generic "solutions" - Clean close — next step, no sales-cadence phrasing

Excellent Example — Website Hero

Petabit Scale

We architect, build, and supply large-scale network infrastructure. Hyperscaler-grade expertise, strategy, and pricing — for the unique challenges of operating networks at massive scale.

[Customer logos: cite only those with confirmed reference permission — see Open Question #3]

[Three sections below: Architect • Build • Supply]

What's on-brand: - Three verbs name the actual offering - Value prop in one sentence; audience named specifically - Logos as proof — not testimonials, not quotes - No "Let's unlock your potential" CTA - No marketing illustration; let the logo and type do the work

Excellent Example — Proposal Opener

Scope & Approach

You're building a 128-rack AI training cluster with a three-tier EVPN-VXLAN fabric, 800G spine, 400G ToR. The three things you asked us to handle: optical transceiver supply at volume (all SKUs in-stock when the racks land), leaf-spine cabling with MPO-12 trunks, and a written architecture review before the order locks.

This proposal covers those three. Pricing assumes the SKU list in Exhibit A; lead times assume a PO by May 15.

What's on-brand: - States what was heard, what the proposal covers, what's assumed - Specific technical content (EVPN-VXLAN, 800G spine, MPO-12) - No boilerplate preamble - Dates and dependencies named concretely

Excellent Example — Blog Post Opener

The 800G optics question is mostly settled — except for the people still arguing for QSFP-DD.

Roughly 90% of 800G shipments have gone OSFP112 in the last two quarters. The ecosystem is frozen around that standard for anything north of 400G, and the QSFP-DD supply chain is drying up by the week. If you're speccing a new fabric, the choice isn't hard — it's been made for you. Here's the specific math, and why the "we'll just use what we know" argument stops working around 400G scale.

What's on-brand: - Thesis-first, no bury-the-lede - Opinionated ("the choice isn't hard") - Specific data (90% volume share) - Respects reader as peer ("you're speccing a new fabric") - Promises payoff: the math comes next

Excellent Example — LinkedIn Post

At 400G fabric scale, the wrong connector choice is a quarter of timeline.

If you're on QSFP-DD and planning 800G, the supply math doesn't work anymore — roughly 90% of 800G volumes have gone OSFP112. The optic supply and your AI cluster delivery date are now the same variable.

We spend a lot of time pulling customers out of this corner. Happy to share what the path looks like if you're staring at it.

What's on-brand: - Concrete hook with a real number - Opinion stated plainly - No emoji, no thread-hook, no "thoughts?" CTA - Offers substance (happy to share) without BANT-qualifying

Excellent Example — Internal Team Announcement (Slack)

Heads up — we just shipped the Chennai MI355x quote to Myriad360 (Vultr). Three scenarios: single-cluster / dual / multi. Starting prices on each; I told them directly those are starting prices, not acquisition cost, so we've got room to negotiate on the fiber and DAC lines if Vultr pushes back. Timeline assumption is a mid-May PO. Ping me if anything about that reads wrong.

What's on-brand: - Specific (names the customer, channel, and pricing posture) - Direct, no Slack-emoji performance - Invites disagreement ("ping me if anything reads wrong") without begging for it - Internal register drops formality but keeps the substance density

Excellent Example — Escalation Reply (customer is upset)

Understood — the Wednesday delivery miss caused you to push the rack turn-up to Monday, and that's on us. Here's what happened and what we're doing.

The MPO trunks we promised missed the Taipei cutoff by one day; our logistics person didn't flag the time-zone difference. Units shipped FedEx Priority Overnight Wednesday night; tracking attached. You'll have them Thursday morning US.

Going forward we're holding a one-day buffer on any Asia-origin order with a hard US delivery date. I also owe you a root-cause note once I've run the full post-mortem — expect that by end of next week.

If the Monday turn-up moves anything downstream, let me know today and we'll see what we can rebook.

What's on-brand: - Acknowledges the specific downstream impact first - Names the mistake concretely and who made it - Next step stated with a date, not a platitude - Systemic fix included so the customer knows this won't repeat - Offers to help with downstream impact — practical, not defensive - Uses "I" for personal accountability — this is the permitted exception

Example to Avoid — Generic Vendor Copy

"At Petabit Scale, we're passionate about empowering next-generation operators to unlock transformational outcomes through our industry-leading, scalable infrastructure solutions. Partner with us on your journey to deploy mission-critical networking at enterprise scale."

What's off-brand (nearly every word): - "Passionate about empowering" — hollow vendor intro - "Next-generation" — empty superlative (no generation named) - "Unlock transformational outcomes" — banned corporate cliché - "Industry-leading" — claim without evidence - "Scalable" — vague reassurance - "Solutions" — filler noun - "Partner with us on your journey" — sales-funnel vocabulary - "Mission-critical" — meaningless overuse - "Enterprise scale" — wrong audience category (we serve operators, not "enterprise") - Fix: delete and rewrite using the hero example above.

Visual Identity

Logo System

The mark is two linked, faceted hexagonal forms creating a figure-eight. It reads as interconnection at scale, with an engineered/constructed quality (planar facets, not organic curves). It's distinctive — deliberately un-generic in the networking-vendor space.

All canonical brand assets live at /assets/brand/ — see /assets/brand/README.md for the complete file inventory, naming convention, and usage guide. Summary:

Lockups: horizontal (primary, mark-left + wordmark-right) · stacked (mark above wordmark, square-ish) · mark (icon only, square).

Color variants per lockup: color (canonical 6-color palette on transparent, for light/cream surfaces) · on-dark (same canonical 6-color facets on a charcoal #1F272B plate, per the original Petabit-Scale-logo2 design — for dark contexts, photography, or any variable ground). There is no white or charcoal silhouette variant — the wordmark stays teal and bronze on dark; the dark-bg treatment is the color design with the charcoal plate baked in.

File name pattern: petabit-scale-{lockup}-{variant}[@{scale}x].{ext} — e.g., petabit-scale-horizontal-color.svg, petabit-scale-mark-on-dark@2x.png.

Usage rules: - Default to petabit-scale-horizontal-color wherever layout allows - Maintain clearspace equal to the height of the mark on all sides - Minimum sizes: horizontal at 120px wide (web) / 1 inch (print); stacked at 80px / 0.75 inch; mark only at 32px / 0.4 inch - Do NOT recolor outside the two documented variants; do NOT stretch, skew, or rotate; do NOT place the color variant on busy or low-contrast backgrounds (use on-dark instead, which includes its own charcoal plate); do NOT create or use a white-silhouette treatment — that isn't the canonical dark-bg design

Regeneration: if the masters at /assets/vectors/masters/ change, rebuild the full asset set via /scripts/build-brand-assets.py + /scripts/extract-mark-svg.py. The scripts depend on PyMuPDF.

Color Palette

Canonical hex values extracted directly from the vector masters (designer-defined RGB fills, not rasterized approximations). The mark uses seven palette colors — six mark colors plus the charcoal used for dark-background lockups.

Palette character: deliberately un-tech. No electric blue, no pure black/white. Muted, earthy, premium — more architectural heritage than SaaS startup. This should carry into digital surfaces: resist the pull toward high-contrast blue/black web conventions.

Extraction provenance: Values extracted directly from the vector drawing instructions of /assets/vectors/masters/Petabit-Scale-logo.ai (mark master) and cross-checked against Petabit-Scale-logo2.pdf — both files agree exactly on all seven fills. A separate file, Petabit-Scale-logo-horizontal.ai, has the same palette with ~1-bit drift per channel (likely a color-profile artifact from a later save); treat logo.ai + logo2.pdf as the source of truth and reconcile horizontal.ai if/when the masters are next regenerated. Confirmed 2026-04-18 via PyMuPDF page.get_drawings() parsing of designer-defined fills.

Typography

A three-tier typographic system. Voice is constant across tiers; each tier serves a different purpose.

• Display (Tier 1): Raleway — for maximum-impact type. Hero lines, major display headers, section banners, architectural/editorial moments. Raleway's tall x-height and narrow geometric forms echo the logo's faceted mark character and match the "un-tech, architectural heritage" palette direction.
• Headers (Tier 2): Lato — for standard headings (H2–H4), navigation, UI chrome, pull quotes. Warm workhorse display face; cleaner and more utilitarian than Raleway, more expressive than Open Sans.
• Body (Tier 3): Open Sans — for running text, captions, labels, inline UI. Neutral, highly legible at every size. Built for reading.

All three are Google Fonts / open-source (SIL Open Font License). Font files are installed under /assets/fonts/; use those as the source of truth for web and digital production.

Installed files:

Usage mapping: - Display (Raleway variable): lean toward wght=200–400 for large type where the elegance of the thin forms shows. wght=600–700 for display headings that need weight. Rarely use Raleway below ~24px — the narrow forms trade legibility for character. - Headers (Lato static): Light 300 for large body-of-page headings; Regular 400 for H2–H4; Bold 700 for emphasis and navigation. Italic variants for asides. - Body (Open Sans variable): default to wght=400 for running text; wght=600 for inline emphasis and labels. The variable format means the guide doesn't enforce a fixed Semibold value — designers can tune between 400 and 700 to hit the right visual weight for the context. Italic is the same variable-axis treatment via the -Italic.ttf file.

Default tier-selection rule: if you're not sure which tier to use, pick the less-expressive option — Lato over Raleway for headings, Open Sans over Lato for sub-headings. The brand voice favors substance over showiness; typography should reinforce that default.

Why variable for two, static for one: the google/fonts repository ships Lato as six discrete static weights, Open Sans and Raleway as variable fonts. We use what's canonical in each case rather than forcing a uniform format.

Type character: Raleway's narrow geometric elegance pairs with the faceted mark; Lato provides warm-but-neutral mid-tier utility; Open Sans covers dense reading. The combination gives the brand a readable premium feel at every scale without the license cost of commercial grotesques.

Imagery

Three imagery tracks by context:

1. Real infrastructure — actual racks, fiber, optics, DC environments. Used for customer-facing surfaces (homepage, about, case studies). Commission or source authentic photography; stock is off-brand.
2. Technical diagrams and schematics — network topologies, timing charts, deployment diagrams. Used for community content and technical pages. Matches the "substantive teacher" voice.
3. Abstract geometric — sparing use of imagery that echoes the logo's faceted aesthetic. Used for corporate/connective-tissue surfaces. Creates cohesion without requiring photography budget.

Never: stock photography of people in headsets pointing at screens, generic tech-circuit-abstracts, handshake metaphors, globes with connecting lines, or any other well-worn B2B-SaaS visual trope.

Iconography & Illustration

Policy: technical / schematic illustrations. Lean into network diagrams, topologies, and engineering visualization wherever possible.

• Icon system: use a clean technical library (e.g., Phosphor or Lucide) for utility icons — but prefer custom schematics for anything brand-facing
• Custom illustration: topology diagrams, fabric drawings, BOM schematics — these are the brand's visual language
• Avoid: generic rounded-corner SaaS icons; decorative abstract illustrations that don't communicate technical content

Confidence Scores

Overall confidence: High — weighted aggregate ~0.90 after the blocker-close pass (palette canonicalized from vector masters, fonts installed). Voice / messaging / terminology / visual-identity all at High; Tone Matrix and Language Patterns remain Medium/Medium-High and will tighten with actual usage.

Open Questions for Team Discussion

All initial High-priority blockers are resolved (palette confirmed, fonts installed). Remaining items are Medium-priority and can be resolved opportunistically.

Medium Priority (improves quality)

1. Tagline decision (deferred) - What was found: User chose to defer the tagline decision until seeing the guide in motion. - Agent recommendation: Revisit after the first content artifact (homepage rewrite) is drafted. If a natural tagline emerges from the hero copy, promote it. If not, leave the name unadorned — the wordmark carries. - Need from you: Revisit in 2-4 weeks with the homepage draft in hand.

2. Customer reference permissions - What was found: Messaging framework and examples cite Vultr, CoreWeave, Lambda, Voltage Park, IMC, Coresite, Equinix. Discovery found named customer threads; no explicit permission documentation. - Agent recommendation: Before any public content names specific customers, confirm reference permission explicitly (even informally). Use redacted references ("top-5 AI compute operator," "tier-1 interconnect facility") where permission isn't clear. - Need from you: Clarify which customer names can be publicly cited.

Resolved Decisions (v1.0 walk-through + blocker close + post-v1.0 corrections)

These questions were surfaced during drafting and closed during the v1.0 walk-through, the subsequent blocker-close pass, and post-v1.0 corrections (all 2026-04-18). Logged here so the reasoning is preserved for future readers.

• Dark-background logo treatment → Resolved 2026-04-18 (correction). The canonical dark-background treatment is the original design from Petabit-Scale-logo2.pdf — the full 6-color facets on a charcoal #1F272B plate. The wordmark stays teal and bronze. There is no white-silhouette or charcoal-silhouette variant; those were generated in an earlier pass and removed. All lockups now ship in exactly two variants: color (transparent background) and on-dark (canonical design with the charcoal plate baked in). Asset builders at /scripts/build-brand-assets.py and /scripts/extract-mark-svg.py enforce this.

• Positioning language: avoid "people hyperscalers hire" framing → Resolved 2026-04-18 (correction). Earlier drafts included a Peer intro variant ("We're the people hyperscalers hire for the hard parts"). That framing was removed — we don't claim to be hyperscalers' vendor. The correct framing is that we bring the grade of engineering, strategy, and pricing access normally only found at the biggest hyperscalers to operators who need it but don't operate at that scale. All value-prop and pillar language was audited for this distinction. "Hyperscaler-level" was also changed to "hyperscaler-grade" throughout for precision.

• Additional market context integrated → Resolved 2026-04-18. Three concrete differentiators added to the positioning language, each slotted into the pillar where it most naturally belongs: 1. Deep optical-systems expertise — including complex high-end coherent transmission (400ZR, 800ZR) and long-haul fiber-optic networks — added to Pillar 1 (Trust / Expertise) as a domain-depth clarifier, and to the terminology guide as example gear vocabulary. 2. Supply chain capability at scale — ability to procure and deliver cutting-edge components in massive volumes, including SKUs that are on-allocation elsewhere — added to Pillar 3 (Value / Economics) as a supply-chain clarifier. 3. Nimble problem-solving — small team, short decision paths, adaptive when plans change — added to Pillar 4 (Execution) as the operational-style clarifier, and reflected in the Brand Personality section. The Executive Summary was extended with a second paragraph naming all three. No new pillars were added; the four-pillar structure remains intact.

• Overclaim language softened → Resolved 2026-04-18. Three specific phrases implied a closer structural relationship with hyperscalers than is accurate. Each was rewritten: 1. "at the pricing the hyperscalers actually pay" → "at the grade of pricing normally reserved for the largest buyers" 2. "from the same factories hyperscalers use" → "directly from the manufacturers" 3. "hyperscaler-tier pricing because we have the direct manufacturer relationships" → "the grade of pricing normally reserved for the largest buyers, because we source directly from manufacturers in volumes most channels can't match" The value claim is preserved (buyers get access to a grade of pricing usually requiring massive purchasing power) but we no longer imply we have identical commercial agreements to hyperscalers themselves. Consistent with the reframe from "we're the people hyperscalers hire" to "we bring hyperscaler-grade capability to operators."

• Live website (petabitscale.com) integrated as a source → Resolved 2026-04-18. A dedicated "Live Website Reference" section now maps the site's four-service taxonomy (Architecture and Design · Network Procurement · Strategy and Insight · Turn-Key Infrastructure) to the four v1.0 pillars, so new content production stays consistent with what's publicly promised. The site's public client roster (GTT Communications, Deft, CoreSite, Ziply Fiber, Vultr, Kentik) was added as already-citable without additional permission work. Specific site copy that violates v1.0 voice ("solutions," "world-class," "Supercharge," etc.) was flagged for the homepage-rewrite project.

• Agent-facing orientation docs added → Resolved 2026-04-18. Two new files at the repo root make the brand system discoverable and usable by automated tools without exploring the tree: CLAUDE.md (prose orientation — what's canonical, read-first file map, hard rules, common agent task recipes) and brand-summary.json (machine-readable snapshot of palette, voice attributes, vocabulary, pillars, logos, typography, and guardrails). The brand voice guide remains the human-readable source of truth; these two are navigation + lookup layers for machines.

• Voice de-personified: "Richard" → team-level throughout → Resolved 2026-04-18. Earlier drafts anchored many voice claims in a specific individual ("Richard's emails," "Richard's CV," "Richard ran networks"). That framed the voice as one person's rather than the team's. Every reference was audited and rewritten to team-level language ("our team," "our outbound," "our operators"). The voice is no longer a Richard-ism being scaled to others; it's the team's voice, which the team codifies and carries. The brand identity metadata was also genericized to "Petabit Scale (team-level source of truth)" rather than naming an individual owner.

• "Field evidence" sections removed from voice attributes → Resolved 2026-04-18. Each voice attribute previously ended with a specific "Evidence" line citing an individual email or customer interaction. Those anchors served the generation process but became clutter in the finished guide. Each attribute now carries only What it means / How it shows up / What to avoid / Confidence — crisper and more transferable.

• New core value added: Real customer outcomes → Resolved 2026-04-18. Five core values now expressed in voice: Candor, Precision & craft, Independence, Peer-respect (implicit), and Real customer outcomes — "solve real problems, deliver when no one else can, save customers meaningful money along the way. The measure of our work is what the customer gets, not what we sell." This complements Pillar 3 (Value / Economics) and Pillar 4 (Execution) as a values-level ethos rather than a positioning claim.

• Terminology update: "EVPN-VXLAN" replaces "BGP-EVPN" → Resolved 2026-04-18. Canonical gear-terminology examples now use EVPN-VXLAN — the more commonly referenced full protocol stack — rather than BGP-EVPN. All preferred-terminology examples and domain-depth lists updated accordingly.

• Layout refinement: Stacked + Mark share a row in the Brand Guide PDF → Resolved 2026-04-18. Section 09 (Lockups) and the on-dark page previously stacked all three lockups vertically. Stacked and Mark are now placed side-by-side in a two-column row, saving a full page of vertical space on each section while preserving the Horizontal lockup's full-width primary position above.

• Dark-background page correction (full-bleed charcoal) → Resolved 2026-04-18. The "On dark backgrounds" page initially showed on-dark logo files on a cream content page. Corrected to full-bleed charcoal canvas so the demonstration is authentic to real-world deployment. The color variants (transparent background) are placed directly on the charcoal page — the facets sit on the surface just as they would in production. The on-dark files (plate-included) remain the right choice for photography or variable-color surfaces we don't control.

• Color palette hex codes → Resolved 2026-04-18. Values extracted directly from the vector master drawing instructions (PyMuPDF page.get_drawings() on Petabit-Scale-logo.ai and cross-checked against Petabit-Scale-logo2.pdf — both agree exactly). Canonical seven-color palette: Primary Teal #4A7876, Forest Green #375E4E, Bronze #C59E79, Deep Bronze #A7815B, Warm Sage #A19680, Cream #FDDCA4, Charcoal #1F272B. A previously-missed Deep Bronze shadow tone was discovered during extraction. An earlier PNG-histogram pass produced values within 1-bit rounding of the canonical set; the vector-extracted values supersede. See Color Palette in Visual Identity.

• Font files installation → Resolved 2026-04-18. Three-tier typographic system installed:

• Raleway (display) — variable font (Raleway.ttf + Raleway-Italic.ttf), covers weights 100–900 — at /assets/fonts/raleway/
• Lato (headers) — six static files (Light 300, Regular 400, Bold 700 + italics) — at /assets/fonts/lato/

• Open Sans (body) — variable font (OpenSans.ttf + OpenSans-Italic.ttf), covers weights 300–800 — at /assets/fonts/opensans/ All files sourced from the google/fonts GitHub repository. See the Typography section for the tier-selection rule and usage mapping.

• Vector + raster asset rebuild → Resolved 2026-04-18. Full brand-asset system generated programmatically from vector masters at /assets/brand/: 9 canonical SVGs (horizontal / stacked / mark × color / white / dark), 27 PNG renders at @1x / @2x / @3x, complete favicon set (SVG + 7 PNG sizes + Apple / Android aliases), and palette files (JSON / CSS / SCSS). Legacy PNG files from /assets/logos/ and /assets/vectors/rasterized/ moved to /assets/brand/archive-legacy/. See /assets/brand/README.md for structure, naming convention, and usage guidance. Original designer .ai / .eps / .pdf files remain untouched in /assets/vectors/masters/ as authoritative sources.

• Empty-superlative policy → Resolved as use-with-caution (not a full ban). Terms like "next-generation," "cutting-edge," "industry-leading" remain allowed when they carry specific technical meaning (e.g., "next-generation 1.6T optics"); cut when they're empty modifiers. See the Use-with-Caution table in the Terminology Guide.

• 2018-era proposals (GIX, Pilot Fiber) → Resolved as archived. No longer tonal reference or capability document. The proposal template refresh starts from first principles using this voice guide. The two files are retained as historical artifacts only.

• Cold outreach → Resolved as opportunistic — no standing templates. Petabit Scale does not run cold outbound as a motion; discovery path remains reputation + referral. When occasional targeted outreach happens, it's written fresh. The tone matrix (First-touch / intro row) covers the register if needed.

• Voice attribute #1 renamed: "Operator-grade" → "Carrier-grade" (v1.2) → Resolved 2026-04-18. The audience IS operators (the top term in our vocabulary; also the audience noun in the value prop), so "operator-grade to operators" named us in our audience's own identity rather than in a quality they value in a counterparty. Carrier-grade is a native industry term of art (telco-tier reliability / engineering standard) that makes the same pedigree claim without the echo, and it pairs cleanly against "Theoretical / consultant-flavored." Pillar #1 renamed to "Carrier-Grade Expertise" to match. The use-with-caution entry for "carrier-grade" was revised: it's now canonical brand language in voice-attribute / pillar / pitch contexts, with the caution narrowed to empty marketing filler ("carrier-grade solutions"). Attribute detail and pillar core-idea prose were revamped (not just word-swapped) to connect carrier-grade semantics — the five-9s engineering standard — to the existing pedigree claim.

• Value-prop tail sharpened (v1.2) → Resolved 2026-04-18. The softer "for operators with unique large-scale needs" replaced with the direct "for the unique challenges of operating networks at massive scale" — surfacing the deep-specialization differentiator. The real positioning point is that Petabit Scale is deeply specialized in the problems that only emerge at massive scale, and the new tail names that directly instead of softening into "needs" language. Applied to primary value prop, 10- and 30-second pitches, OG/Twitter meta, JSON-LD description, and all mirrored copies across the brand system.

• Business card recreated digitally + added to the brand guide (v1.2) → Resolved 2026-04-18. New builder at scripts/build-business-card.py generates a print-ready 2-page PDF at assets/brand/applications/business-card.pdf (3.5″ × 2″ + 0.125″ bleed + crop marks) matching the produced reference card. Fresh Bronze vector-icon family replaces the original envelope/handset/pin/cursor set: paper plane (email), smartphone (phone), teardrop map pin (address), globe with meridian (URL). Icons drawn as ReportLab paths inline — no external icon-library dependency, no vendored SVG. Section 13 added to the brand guide PDF (brings it from 24 → 25 pages) showing front + back previews with a spec table and regeneration instructions. Template workflow is now: edit the draw_front(...) call in the script with per-member info, re-run, ship the PDF to print.

• Audit discipline: grep for split-string phrases (v1.2 lesson) → Resolved 2026-04-18. During the initial carrier-grade rename, a single-line grep missed the executive-summary value prop because it was split across two Python string literals with implicit concatenation — the phrase "operators with unique large-scale needs" wrapped across lines 327–328 of the PDF builder. Lesson: when verifying a rename, also grep the rendered artifact (PDF text-extract via PyMuPDF) to catch cases where source-code line breaks hide the phrase. The post-rename audit protocol is now: grep source files → grep rendered outputs → open the PDF and spot-check the high-traffic pages (cover, exec summary, pillars, pitch ladder).

Watching for

The voice matures as the team grows and the market shifts. Near-term items the guide expects to revise — not open questions (those have formal recommendations above) but an evolving edge:

• Procurement email canonical. Currently intentionally omitted; will anchor to real reference threads once 3–5 accumulate.
• Customer case studies. Will land when permission + outcome data align for IMC (new enterprise onboarding), Vultr (Chennai MI355x buildout), or SambaNova (ongoing advisor relationship). First to land sets the case-study template for the others.
• AI-segment-specific vocabulary. As AI compute becomes our majority revenue segment, the specific register (training fabric · inference · GPU cluster interconnect) may deserve its own subsection under Tone by Context.
• Tagline decision. Deferred from v1.0; revisit after the homepage rewrite. If a natural tagline emerges from the hero copy, promote it. Otherwise, leave the wordmark to carry.
• Community voice in practice. The "substantive teacher" register is documented; it'll sharpen once we have a few published blog posts or NANOG-adjacent pieces to cross-reference.
• Team voice consistency as the team grows. Today the voice lives primarily in one or two writers' outbound. As more people produce content, the guide may need tighter enforcement tooling — a pre-publish checklist, a copy-review cadence, or both.

If you find yourself writing in a context the guide doesn't address, log it here in a commit — that's how this section stays useful.

Data Gaps & Recommendations

• [ ] PII in Google Drive: Discovery flagged an unredacted credit-card spreadsheet and a corporate-docs file containing EIN, DUNS, VAT, CAGE at default Drive permissions. Not a branding issue — but fix it separately. Move sensitive files to a restricted share or lock them with specific-user ACLs.
• [ ] No customer case studies anywhere: Despite deep customer relationships, there are no written case studies documenting completed engagements. Recommendation: Pick 2-3 named customer projects with permission and draft short case studies in the voice defined here. Anchors the "proof" pillar with durable artifacts.
• [ ] Website folder (Drive) is dead since 2020: No current website copy drafts or landing-page iterations exist. Recommendation: The homepage copy rewrite (the user's chosen first artifact) is high-leverage here; its outputs should populate a fresh Website working folder with version history.
• [ ] No AI-specific positioning content despite AI as leading segment: Guide treats AI as a leading segment of customers, not a brand identity. Once the first content piece lands, write one or two follow-ons specifically about AI-fabric infrastructure to establish the segment-specific voice.
• [ ] Social media / LinkedIn activity not observed: No LinkedIn drafts or social content found in discovery. If social becomes a channel, Phase 6 tone matrix covers it; publish consistently or skip the channel — inconsistent presence is worse than absence.
• [ ] Procurement email voice not yet canonical: The tone-matrix row is defined (tight, transactional, SKU-fluent) but no worked example could be confidently produced during v1.0. Recommendation: Capture 3-5 actual procurement reply threads (redacted) and use them to establish the canonical example. Revisit the Customer Email — Procurement section when ready.

Live Website Reference

The current live site at petabitscale.com is a source — and a rewrite target. The service taxonomy is useful; the voice isn't yet aligned to the current guide.

Current site → pillar mapping

Four services on the live site map cleanly to the four pillars. Use this alignment when writing new content so nothing gets orphaned.

Public client roster (already named on the homepage)

The site already publicly names these clients, so they're citable in new content without additional permission work: GTT Communications, Deft, CoreSite, Ziply Fiber, Vultr, Kentik.

Additional customers identified in discovery and customer-email evidence (SambaNova, CoreWeave, Lambda, Voltage Park, IMC, Equinix) are not on the public site — those still require explicit reference permission before public citation. See Open Question #2.

Site copy flagged for homepage rewrite

The live site voice predates this guide and drifts into generic-vendor register. Before the homepage rewrite, these specific phrases should be replaced:

Target homepage hero (from the voice guide): "We architect, build, and supply large-scale network infrastructure. Hyperscaler-grade expertise, strategy, and pricing — for the unique challenges of operating networks at massive scale."

Appendix: Sources

Document owner: Petabit Scale (team-level source of truth). Living document; edits version-tracked via git. Next scheduled review: when the first content artifact (homepage copy rewrite or proposal template refresh) reveals friction or gaps.

Provenance

Generation details for the original v1.0 guide, kept for traceability. v1.1 and v1.2 changes are logged under Resolved Decisions above. Moved to the bottom so the canonical rules stay at the top where writers reach first.

• Created: 2026-04-18
• Version: 1.2 (v1.0 first-time formalization 2026-04-18; v1.1 ten-improvement review pass 2026-04-18; v1.2 voice-attribute rename "Operator-grade" → "Carrier-grade" + value-prop tail reframe 2026-04-18 — same day, live session iteration)
• Replaces: None — greenfield guide
• Owner: Petabit Scale — team-level source of truth
• Cadence: Living doc; edits welcome, version-tracked in git
• Sources: Discovery across Google Drive, Slack, Gmail, Zoho Desk (~24 brand-relevant artifacts) + 10-phase founder interview + local visual asset library + live website + customer email field research
• Documents processed: 10+ (overview deck, 2 proposals, bio, corporate docs, 30+ customer email threads, logo masters, live site)
• Conversations analyzed: 0 dedicated transcripts; customer email corpus used as conversational proxy
• Discovery report used: Yes — full triage report
• Overall confidence: High (weighted aggregate ~0.90)