Strategy comes down from the top. Productise the practice. Capture reference projects. Make the offerings repeatable. Nobody in the room disagrees, because the sentence is correct. The company has done good work for years. Somewhere in that work is a reusable shape. Marketing is told to go and find it.

So marketing does what marketing is allowed to do inside a serial organisation. They book meetings. They interview a project manager who ran something “like that.” Maybe they get twenty minutes with a technical lead. They ask the questions that make sense in marketing language. They leave with five sentences.

Those five sentences are not merely short. They are oversimplified along the wrong dimensions. They do not contain the reusable shape. They do not contain what the client actually bought. They do not contain the method that made the engagement work. They contain whatever survived a human translation hop performed by someone who does not speak marketing as a native dialect.

Then those sentences become a case-study page. Dull. Segment-average. Safe enough to publish, useless enough that sales cannot sell from it. A salesperson with a real relationship still has nothing specific to put in front of a client this month. Someone writes a statement of work anyway, with a budget attached. Delivery opens the pack and asks the only honest question left: what are we trying to do, and who set this number?

Nobody asked them.

The filing cabinet is wrong

Organisations know this pattern. They feel it in their calendars and their win rates. Then they file it in the wrong cabinet.

• Marketing is shallow. They never understand the work.
• Sales is hand-wavy. They oversell and under-specify.
• Technicals cannot communicate. They hide behind jargon.
• Project managers gatekeep. They translate everything into status beige.

Soft-skills training is scheduled. A lunch-and-learn about storytelling. A request that engineers “write for a business audience.” The hop structure stays identical. The five sentences stay wrong. The website stays dull. The SOW stays a surprise to the people who have to build it.

We think the people diagnosis is not just unkind. It is architecturally false. Competent people, operating honestly inside a serial telephone game, will still destroy the dimensions the next reader needs. That is not a character failure. That is what serial re-encoding does.

What this field guide owns

This book is about communication topology — how project truth moves through an organisation, and why the default path guarantees loss.

You will get four load-bearing claims:

1. Serial telephone. Technical → PM → marketing → sales → delivery is a chain of re-encodings, not a chain of copies.
2. Wrong-axis compression. Each hop compresses along the translator’s dimensions. Residual fidelity multiplies down the chain; it does not merely “add a little noise.”
3. Radial inversion. Hold ground truth joined at full fidelity. Give every audience a first-generation translation in their own dialect. Nobody is more than one hop from the truth.
4. Register as the fourth leg. Storage, structure, and conversational access are not enough. The missing leg is dialect — marketing-speak, sales-speak, technical-speak, finance-speak.

The cognitive exoskeleton pattern — AI saturates preparation, humans keep judgment — is adjacent and covered elsewhere. This piece is not that catalog. This piece is the topology underneath: why the organisation keeps misfiling a pipeline failure as a personality failure.

How we will prove it

Part I names the misdiagnosis, maps the serial chain, and makes the information-theory argument: loss compounds multiplicatively when each translator re-projects onto their own axes.

Part II runs one project — Aperture, a telco billing engagement — through five serial hops, then through radial first-generation dialects from the same ground truth. Then it names the fourth leg of the stack: register.

Part III applies the same doctrine to marketing, sales, and delivery, then to the CEO — who is already the organisation’s human translation layer, which is why pure dialect translation helps them least.

Your company already speaks five languages. The missing capability is not louder meetings. It is a translator that starts from the truth — once — and speaks every dialect as a first generation.

Start at the end, because the end is where the physics shows.

Delivery opens a statement of work. There is a fee. There is a timeline. There is a paragraph about “billing modernisation” written by someone who has never sat inside the actual method. The technical lead reads it twice and feels the familiar heat behind the eyes. What are we trying to do? What’s the budget? Nobody asked us.

That sentence is usually filed under “sales oversold” or “scoping was weak.” It is the same serial telephone that produced the dull website — running in reverse.

Name the chain

In a professional services firm, a project description rarely travels once. It travels as a chain of human translations:

Side branches make it worse: legal, finance, partner teams, offshore delivery managers. Each branch is another hop. Each hop is another chance to re-encode.

This diagram is the definitive placement of the serial chain for this book. Later chapters will not re-draw it. They will point here and show what the chain does to a single project.

Each hop is a re-encoding

People talk as if information is passed along. It is not. Information is projected.

• The technical team holds mechanism, constraints, and what was hard.
• The PM re-encodes into status, risk, timeline, and stakeholder hygiene.
• Marketing re-encodes into narrative, benefit language, and publishable safety.
• Sales re-encodes into objection hooks and relationship-safe claims.
• The SOW re-encodes into commitments, fees, and dates that procurement can sign.
• Delivery tries to reverse-engineer method from commitments that no longer match method.

The website becomes the source for sales. The sales deck becomes the source for the SOW. The SOW becomes the source for delivery planning. By hop four you are not reading the project. You are reading a translation of a translation of a translation — and treating it as ground truth because it is the document that exists.

Bidirectional damage

The chain fails in both directions, and organisations pretend the two failures are unrelated.

If you only fix the website, you still get bad SOWs. If you only tighten scoping templates, you still get unusable marketing. Both are symptoms of serial re-encoding.

Hop count as a diagnostic

You do not need a laboratory to start. You need a pencil.

1. Pick one project that became a public story or a signed SOW.
2. List every human and artefact that re-stated it before the final form.
3. Count the hops.

In mid-size professional services firms the count is often four to six before the description leaves the building. That is a shape-statement from lived consulting structure, not a fabricated industry average. The number that matters is yours.

Conway’s Law for offerings

Melvin Conway observed that organisations design systems that copy their communication structures.¹ The same law applies to offerings, case studies, and productised services. A serial communication structure does not merely slow marketing down. It manufactures serial, degraded product descriptions. The architecture of the talk becomes the architecture of the thing you claim to sell.

Myth versus reality

Hope is not a translation layer. Putting a better writer at hop three does not restore dimensions already discarded at hop two.

By the fourth hop you are reading a translation of a translation of a translation — and treating it as ground truth because it is the document that exists.

The project manager is not a villain.

They sat in the technical debrief and did their job. They kept the risks that could blow the plan. They kept the dates that status meetings require. They kept the stakeholder map that keeps politics calm. Then marketing asked what the project was about, and the PM told the truth — their truth. Schedule hygiene. Risk mitigated. Stakeholders aligned.

Marketing needed the reusable shape: what pattern this was, why a buyer would care, what differentiates the method. Those dimensions were already gone. Both people told the truth. Both destroyed value for the next reader.

That is the compression-dimensions problem. It is the definitive argument of this chapter. The flagship project in Part II will show it; it will not re-derive it.

Compression is not optional

Human working memory is limited in chunks, not infinite bits. A chunk is the largest meaningful unit the listener can hold. Every cross-role conversation forces reduction. Soft-skills training that pretends otherwise is theatre. The only open question is which axes get kept when the reduction happens.

Lossy compression is never neutral. It preserves some dimensions and throws others away. Once thrown away, they are not sitting politely in a footnote waiting to be recovered by the next hop. They are gone from the packet the next human receives.

What each role keeps

Read the table as a crime scene, not a personality chart. No row is “bad at communication.” Each row is a legitimate compression basis for a legitimate job. Serial topology forces one job’s basis to become the next job’s only input.

Multiplicative loss

People intuit serial loss as additive: a little muddying at each step, so four hops means four mild insults. That intuition is wrong for wrong-axis compression.

Think in residual fidelity. If hop 1 preserves only a subset of the dimensions hop 2 needs, hop 2 cannot invent them honestly from the packet. It compresses what remains. Hop 3 compresses the remainder of the remainder. Roughly:

F_final ≈ F₁ × F₂ × F₃ × F₄ × …

That is a shape-model, not a laboratory constant. We will not invent percentages of “value lost at hop two.” The structural claim is enough: most of the loss happens at the first wrong-axis hop, and by the fourth hop the reusable shape is usually gone. Later hops often polish the wreckage into prettier language. Prettier is not more faithful.

Cascades lock in the intermediate

Serial multi-agent systems show the same structural failure in software form: when agents are chained, errors often compound rather than cancel. A subtly wrong intermediate is treated as fact downstream and locks in.³ Your org chart has been running that architecture with humans for decades. The PM summary becomes the marketing brief. The marketing brief becomes the sales narrative. The sales narrative becomes the SOW. Nobody is lying. The intermediate is simply no longer optional.

Inferential distance stacks

Communication cost is also inferential distance: explanation fails when the speaker assumes a shorter distance than exists between their mental model and the listener’s.⁴ Each role boundary is a distance jump. Serial hops stack those jumps. Asking a PM to spontaneously preserve marketing dimensions is asking them to compress along axes they do not use to do their job. Sometimes a gifted bilingual does it. Topology that requires genius at every hop is not a system. It is a wish.

Designed versus accidental compression

There is nothing wrong with compression. Designed compression — first-generation, along the destination’s axes, from full-fidelity ground truth — is how useful artefacts get made. Accidental compression — each human reducing for themselves, then handing the remainder downstream — is how telephone games get made.

High compression can preserve intelligence when the compression is intentional and the source is intact. Accidental multi-hop compression preserves the politics of the last meeting.

The PM kept budget risk. Marketing needed the reusable shape. Both told the truth. Both destroyed value for the next reader.

Theory is cheap. Side-by-side text is not.

We take one consulting engagement — call it Aperture — and run it through the serial telephone. The names are composite; the topology is not. If you have lived inside professional services, you have watched a version of this movie.

Ground truth (full fidelity)

Before any hop, here is what actually happened.

That packet is rich. It has method, proof, differentiation, and constraints. Serial topology will now destroy it one legitimate job at a time.

Hop 1 — Technical

Mediation layer over legacy rating. Three adapters (CRM, usage export,
settlement). Parallel-run for six weeks. Avoided full rewrite. Main risk
was dual-write consistency during the window; mitigated with replay
harness and nightly reconciliation.

Preserved: method, constraints, hardness.
Missing for marketing: offer packaging, buyer language, productised name for the pattern.
Fidelity: high. Wrong audience packaging, not yet wrong content.

Hop 2 — Project manager

Integration phase delivered on time. Key risks mitigated via parallel-run
approach. Stakeholders aligned across billing and CRM. Budget held with
minor change control. Recommended as a solid reference engagement.

Preserved: schedule hygiene, risk theatre, political calm.
Dies: disintermediation pivot; why the method mattered; reusable pattern; dual-write physics as a product fact.
Fidelity: this is the expensive hop. The wrong-axis compression from Chapter 3 lands here. Marketing will never recover what the PM did not keep.

Hop 3 — Marketing

Helped a leading telecommunications provider modernise its billing
capability and improve customer outcomes through a structured
transformation programme.

Preserved: publishable safety, industry noun, positive vibe.
Dies: everything specific. This is now a segment-average sentence any competitor could claim.
Becomes: the seed of the case-study page. Dull on arrival.

Hop 4 — Sales

We do billing transformation for telcos. Strong delivery record.
Happy to introduce the team and talk about how we partner for change.

Preserved: relationship-safe generality.
Dies: proof shape (three adapters, six-week parallel); objection path against rewrite-shaped RFPs; differentiation.
Behavioural consequence: the salesperson fills silence with generic prowess language. Not because they are dishonest — because the packet no longer contains sellable specifics.

Hop 5 — SOW / delivery brief

Fixed-fee billing modernisation. Deliver upgraded rating capability
within nine months. Assumes availability of modern APIs and a single
cutover window. Client provides data dictionaries in week two.

Preserved: fee, date, procurement-friendly certainty.
Dies: mediation-first method; dual-run requirement; no-greenfield constraint; the actual reason the original engagement worked.
Delivery inherits: a budget that fights physics. The blow-up is not a surprise. It was scheduled when hop 2 and hop 4 erased the method.

Annotation table

Nothing in this chain required a villain. Topology was enough. The multiplicative loss model from Chapter 3 is visible without a single invented percentage: the reusable shape is mostly dead by hop 2; hops 3–5 just style the corpse and attach a number.

The internal email that named the pivot never reached marketing. The SOW that denied the method reached delivery with a signature on it.

Now open Aperture again. No PM interview. No five sentences. No telephone.

The ground truth from Chapter 4 still exists: mediation layer, three adapters, six-week parallel-run, internal pivot framing, no-greenfield constraints. In a serial organisation that packet only travels by human re-encoding. In a radial architecture it sits joined once — proposals, emails, tickets, deliverables, post-mortems addressable together — and every audience asks in their own dialect.

Topology inversion

The centre is not a summary. Summaries are already compressions. The centre is full-fidelity residue, structured enough to navigate. Construction mechanics are a sibling problem. This chapter owns the topology: spokes instead of chains.

Each spoke compresses along the destination’s dimensions. Marketing does not receive a PM packet. Sales does not receive a marketing packet. Delivery does not receive a sales packet. Each receives a first-generation artefact cut from the same source.

Aperture after: five first-generation dialects

Marketing dialect

Offer pattern: disintermediation without rip-and-replace.
We productise mediation-first modernisation for billing estates that
cannot fund a multi-year rewrite. Proof: three production adapters,
six-week parallel-run, rewrite deferred then cancelled. Best fit:
operators whose board fears greenfield risk more than integration work.

Sales dialect

If the client believes they need a rewrite, the objection path is:
mediation layer first; legacy becomes optional; dual-run proves safety;
rewrite becomes a choice, not a hostage situation. Do not promise
greenfield APIs. Do promise a fixed parallel window with reconciliation.
Reference proof points: three adapters live; six weeks dual-run.

PM dialect

Method-true plan: mediation build → adapter sequence → dual-run window
with nightly reconciliation → decision gate on legacy retirement.
Real risk was dual-write consistency, not "stakeholder alignment."
Done meant parallel proof, not a single cutover slide.

Technical dialect

Reuse: mediation boundary, adapter contracts, replay harness,
reconciliation checks. Anti-pattern: greenfield rating rewrite under
fixed fee when APIs are not clean. Constraints that mattered: limited
change windows; dual-run mandatory; settlement path cannot pause.

Finance dialect

Fee shape: fixed-fee is safer under mediation-first with a bounded
parallel window than under rewrite-shaped scope. Margin risk spikes
when SOWs assume modern APIs the estate does not have. Historical
tell: rewrite deferred/cancelled after dual-run proof — cost of
optionality was lower than cost of certainty theatre.

Before / after comparison

Fidelity is not “more words.” Fidelity is the right axes for the reader, compressed once from source. Marketing’s page can still be short. It is no longer empty.

What AI is for here

AI’s job is not to invent marketing from model priors. That path produces the segment-average page you already hate. AI’s job is fluent register translation over your joined corpus: marketing-speak, sales-speak, technical-speak, finance-speak, cut from the same ground truth.

Humans keep taste, brand judgment, and the publish decision. The serial hop as a lossy intermediate is what dies.

Technology amplifies whatever process you encode.⁵ Auto-writing a case study from the PM’s five sentences automates the telephone. Auto-producing first-generation dialects from joined ground truth automates the radial architecture. Same model class. Opposite topologies. Opposite results.

The website is no longer a translation of a translation. The SOW is no longer folklore with a number attached.

You can already imagine the frustrated executive.

We digitised the archive. We joined the soft data. We put a conversational interface on top. The demo looks good. Then marketing asks about a project and receives a technically correct page they cannot put on a website. Sales asks and receives a risk log. A new hire asks and receives a dump that assumes ten years of tribal context. The stack is “working.” The organisation still cannot talk to itself.

Three legs were never enough.

The three legs you may already have

Earlier work — especially the ebook Your Life Compiles to One Language — framed a compiler stack for personal and corporate residue. We reference it here; we do not re-argue it.

1. Storage. Everything useful eventually wants to be text, because text is the model’s home turf. Dead formats and sealed silos are storage failures.
2. Structure. Text joins into a navigable graph — claims, edges, pointers — not a SharePoint landfill. Soft joins across stores make cross-corpus walks possible.
3. Access. Conversation is the runtime. Turn-sized claims beat document archaeology. The graph is addressable in a single conversational breath.

Those three legs answer: can the machine represent it, join it, and retrieve it in conversational time? They do not answer: can this human use the answer in their job without another human translator?

The fourth leg: register

Register is the organisational dialect of the person asking. Marketing-speak. PM-speak. Technical-speak. Sales-speak. Finance-speak. HR-speak. Legal-speak. Not natural-language pairs. Sociolects.

A correct answer in the wrong register is still operationally wrong. Marketing does not want the technical page. Sales does not want the PM risk log. The CEO does not want a forty-page dump five minutes before the board pack is due. Retrieval succeeded. Register failed.

Japanese ↔ English is table stakes. The enterprise prize is fluent register translation over your joined intellectual property. Staff stay in their comfort zone. The system brings the organisation’s work into the room in the language of the person asking. That is radial topology (Chapter 5) made first-class in the stack.

Failure modes per missing leg

This is the definitive placement of the four-leg stack. Part III will apply register; it will not re-justify it.

Organisational amnesia makes register urgent

Institutional memory is a sibling doctrine: the organisation never held the memory — its people did. We will not rebuild that ebook here. The wedge matters for register.

A large share of workplace knowledge is unique to individuals, and knowledge workers already burn hours every week waiting for colleagues or recreating knowledge that exists somewhere else.⁶ A new marketing hire or salesperson arrives with complete amnesia of the last decade. CRM crumbs are not memory. Even if you solve amnesia with structure and access, a day-one hire still needs answers in a dialect they can use. Radial register translation is how inheritance becomes operational on Monday, not after six months of corridor osmosis.

Adjacent, not this book

The cognitive exoskeleton pattern — AI saturates pre-work, humans keep judgment, relationship, and accountability — is adjacent territory. The per-role prep morphology is covered elsewhere. One sentence is enough here: the exoskeleton was the mechanism; the joined wiki is the payload; this book is the communication topology and the fourth leg that makes dialects first-class.

We also will not dig into wiki construction — ingestion, janitors, graph design. Those mechanics live in prior work. Register assumes structure exists. It does not re-litigate how to build it.

You already built three legs and still cannot talk to marketing. The missing leg is register.

Picture the salesperson at the end of the month. Best client relationship in the firm. Trust is high. Budget season is open. And they have nothing specific to sell — not because the firm has done nothing, but because everything the firm has done lives in other people’s heads, other people’s dialects, and five-sentence pages that could have been written by a competitor.

That is not a CRM hygiene issue. That is serial topology plus missing register.

This chapter applies the same doctrine from Parts I and II. It is not a new framework. It is not a per-role exoskeleton catalog — that morphology is covered elsewhere. It is three variants of radial first-generation translation.

Variant A — Marketing: articulate before you publish

Marketing’s valuable job is often mis-specified as “write the webpage.” The load-bearing job is upstream: articulate what the company has actually done as an offer. Name the reusable shape. Productise the pattern. Only then does a page have something true to say.

Serial path: calendar meetings, wrong questions, five lines along the wrong dimensions, dull page, generic model-assisted rewrite that makes it smoother and emptier.

Radial path: marketing asks in marketing dialect over joined projects. The system returns first-generation offer language — pattern name, proof points, fit, anti-fit — cut from ground truth. Aperture becomes “disintermediation without rip-and-replace,” not “billing modernisation outcomes.”

Variant B — Sales: join what no one person holds

A salesperson structurally cannot hold the other hundred and ninety-nine people’s work. The CRM page holds what the previous owner bothered to type. The capability map of the firm is scattered across proposals, delivery notes, and corridors.

Serial path: sales inherits marketing’s flattened packet, invents talk tracks, over-promises or under-sells, and blames marketing when the deal needs specifics.

Radial path: the human brings deep client knowledge; the system brings the firm’s full capability map in sales register. Cross-sell becomes a join, not a brainstorm. “Nothing to sell this month” often means “nothing I can see in my dialect from the last ten years of work.”

Proposal compilation is a sibling productisation of the same idea — kernel times context, bespoke at speed. The light claim here is enough: both the firm’s IP and the client history want to be read from one joined substrate, then spoken in sales register.

Variant C — Delivery: SOWs in physics, not folklore

Delivery is where serial reverse-path damage becomes calendar time and margin.

Serial path: sales folklore becomes SOW language; fee attaches to a shape that never existed; delivery discovers mediation-first physics after signature; the post-mortem blames estimation.

Radial path: commitments are first-generation artefacts in delivery dialect over how the firm actually executes. Aperture’s dual-run constraint appears in the SOW because it appears in the ground truth, not because a hero PM fought for it in a late review.

This is the same radial rule as marketing and sales. Inbound commitments are spokes from truth, not the last translation in a telephone game.

One mechanism, three artefacts

What we refuse to expand here

• Per-role exoskeleton prep morphologies — sibling territory.
• Full agentic RFP orchestration — a different book.
• Read-path metrics and resurrection dashboards — a different book.

The braid stays tight: serial versus radial, compression dimensions, register. Variants prove the doctrine. They do not invent three new doctrines.

Marketing stays in marketing’s comfort zone. Sales stays in the relationship. Delivery stays in the method. The system brings the IP into the room in the right dialect.

The CEO floats an initiative before it leaves the building. Margin is thin. People are busy and excellent. Something has to change. The idea is half-formed and still plastic — which is exactly when translation damage is cheapest to prevent and most expensive to ignore.

Today the pre-flight is gut, politics, and the last person who spoke loudly in the hallway. Finance will hate it — maybe. HR will slow it — maybe. Sales will rephrase it into something unrecognisable — almost certainly. Delivery will discover the constraints after the all-hands. The CEO, who can already speak four dialects poorly and two well, tries to hold the whole telephone game in one skull.

The CEO as human translation layer

We think this is under-named. The CEO role is described as strategy, capital allocation, culture, and decision-making. Quietly, it has also included being the organisation’s human translation layer — expected to convert technical into board language, finance into product pressure, sales reality into strategy, and HR constraints into something the exec team can act on.

That observation has a sharp consequence: pure dialect translation helps the CEO least. Handing them “marketing-speak for this project” is not the unlock. They already do hops in their head. They are bilingual enough to survive. The serial telephone hurts them less as readers and more as people forced to be the network switch for every departmental dialect.

What helps instead

Still radial. Still register. Different spokes.

Take the initiative while it is still plastic and ask for first-generation stakeholder lenses from the company’s actual history:

• Finance register: where similar initiatives destroyed margin; where fixed-fee was safe; what the last two analogous bets did to cash.
• HR / workforce register: capacity patterns, change fatigue, skills that actually existed last time you tried this shape.
• Sales register: how buyers will hear it; which proof points exist; which promises the firm cannot keep.
• Delivery register: method constraints, dual-run realities, the physics that killed the pretty version last time.

Generic model priors say “projects like this sometimes fail.” Joined organisational history says “here is the post-mortem, here is the SOW that lied, here is the email that named the real constraint.” Shape-of-failure with receipts — not a full multi-agent think-tank redesign. The CEO keeps the call, the politics, and the accountability. The system stops forcing them to be the only working translator in the building.

What to do Monday

You do not need the perfect wiki to change the filing cabinet. You need one chain and a pencil.

1. Pick one chain. A project that became a case study, a sales talk track, or an SOW.
2. Write the hop list. Who re-encoded it, in what order, into what artefact. Use the Chapter 2 chain as your template.
3. Mark the dead dimensions. At each hop, what destination-critical fact disappeared. Use Chapter 3’s axes table.
4. Locate ground truth. Not the interview notes — proposals, emails, tickets, deliverables, the internal framing that never went external.
5. Produce two first-generation translations for two audiences from that ground truth. Compare them to the serial artefacts. The gap is your proof.
6. Optional: hop-count one path end-to-end on your org chart. The number that matters is yours.

Hope is not a translation layer. Soft-skills training is not a topology change. Automating the telephone is how you lose faster.

Closing doctrine

Organisational communication fails as a serial game of telephone. Each hop re-compresses along the translator’s dimensions. Residual fidelity multiplies down the chain. Marketing’s five dead sentences, sales’ empty month, and delivery’s shocked SOW are the same architecture at different hops.

The fix is radial: joined ground truth at full fidelity; first-generation dialects as spokes; nobody more than one hop from the truth.

The stack is incomplete until it includes register:

Your company already speaks five languages. The missing capability is not louder meetings. It is a translator that starts from the truth — once — and speaks every dialect as a first generation.

It was never a comprehension problem. It was a lossy serial pipeline. Stop training people to shout more clearly down the telephone. Change the topology.