The CEH-001 situation

Why start here

CEH-001 is this book's running exam case: a credit-equity hybrid on an equity desk. Not a vanilla CDS, not a listed share. Two honest marks land near 1.50 and 2.10. The midpoint 1.80 is denied. Later chapters assume you can see that fight.

Think of this page as a short physical law: the statement is small (two honest marks, refuse the average), the machinery underneath is the rest of the series. The companion book, Financial AI, already introduced North Desk -- a trading account, a "Coin" representing a bet with some win probability, and a "Forecaster" (a model) that proposes trade ideas through a named alias. This book uses that same account to ask a different question, not a second, unrelated example. A dashboard or a language model that simply prints one confident-looking number instead of showing the real disagreement between two honest methods is exactly the failure mode this book exists to prevent.

Bid, ask, mid (quick)

Bid: highest buy price now. Ask: lowest sell price. Mid is usually (bid + ask) / 2. On a liquid stock the crowd prints that for you. On CEH-001 there is no trustworthy mid for the claim itself, so the firm must construct a mark (the value it books for inventory, P&L, limits).

The rare building in distress

Imagine you must appraise a rare, historic building in the middle of a city. No one has sold this exact building in years, so you cannot look up its market price. Worse: the company that owns the building is heading toward financial distress (roof leaking, tenants leaving, bankruptcy risk rising).

That is CEH-001: a one-off claim with real legal cashflows, no clean public tape for the claim itself, and an issuer under stress. The finance team still has to answer: what is this worth on the books today?

Two math teams use two honest methods and get two different answers. That disagreement is the whole point of this chapter.

What is a credit spread?

When a company borrows money, it is like a landlord asking investors to fund a building.

• The baseline. A perfectly safe, government-backed claim down the street might guarantee about 5% yearly return.
• The extra risk. Our building's owner is in trouble. Investors say: if I put money here, I need a premium for the chance of loss or delay.
• The spread. That extra premium is the credit spread. If the distressed company must promise 8% while the safe baseline is 5%, the credit spread is 3% (8% minus 5%).

Wider spread means markets want more pay for the risk. Tighter spread means less fear is priced in.

What is a basis point (bp)?

Distress levels and yields move in tiny fractions. Professionals use basis points (bps, said "bips") so they do not drown in decimals. Think of them as pennies on a percentage point:

• 1 bp = 0.01% (one hundredth of a percent)
• 100 bps = 1.00%
• 200 bps = 2.00%
• 300 bps = 3.00%

In the table below, the market demands about 3.00% extra return (300 bps) to take a chance on our distressed issuer. Similar names nearby only require 2.00% to 2.40% extra (200-240 bps). That gap says buyers see this company's distress as a major threat relative to peers.

Recipe A: neighborhood fear (spread frame)

Team A says: do not trust only the internal books of a distressed owner. Look at the neighborhood.

• Real-time truth. Internal accounting can be late or optimistic. Market prices of peer names update as soon as fear moves.
• Sensor network. If a panic headline hits at 09:00, peer prices can move by 09:01. Peers act as immediate risk sensors for how expensive distress looks.
• Relative value. Issuer at 300 bp versus peers at 200-240 bp is a clear "this name is worse" signal.

Recipe A maps that neighborhood fear (through a contracted transform) into the desk's internal value units for CEH-001. On this morning it lands near 1.50, with a plausible band [1.42, 1.58].

What is 1.50? An internal ticket unit (points, percent of notional, or another model unit), not necessarily dollars on a retail screen. The series keeps the unit abstract so you watch the fight between frames. When inventory is 40,000 units, a 0.60 gap becomes money.

Recipe A stays alive because risk files, stress packs, and some hedges already speak spread language. Forcing them onto pure cashflow present value makes them translate silently and lose the audit trail.

Recipe B: rent check (curve / schedule frame)

Team B rejects neighborhood gossip as the whole story. They open the legal lease (the contract calendar) and ask: exactly what cash will this pay over time, and what is that worth today?

• The promised cash (schedule). Toy example only: Year 1 $100, Year 2 $100, Year 3 $100, Final $1000.
• The time discount. You cannot add those to $1,300 and call it today's value. Money today is worth more than money years from now. Team B uses a discount curve (built from stable bank / rates instruments under named conventions) to shrink future rent checks into today's dollars.
• Safety adjustments. On top of time discounting, they layer assumptions about default odds and recovery (what you might salvage if the issuer fails).

Because Recipe B focuses on the rigid legal schedule and time infrastructure, its math lands at a different internal score: 2.10, band [2.01, 2.22].

Recipe B stays alive because client fair-value language, some accounting paths, and cashflow-bucket hedges speak schedule language.

Discount curve in one line. It is a priced calendar for money: each maturity has a discount factor between 0 and 1. Change day-count, bootstrap set, or interpolation without a hash, and the same coupon dates present-value to a different mark. That is why later chapters care about curve honesty and "as-of" clocks.

Same building, two answers

Same claim, same legal document, same clock:

• Recipe A (neighborhood spreads): 1.50
• Recipe B (contract cash flows): 2.10
• Gap: 2.10 - 1.50 = 0.60

They disagree because they measure different things. The English word "mark" pretends they measured the same thing.

The $12,000 midpoint lie (why 1.80 is banned)

The desk owns 40,000 units of this product. How the booked number moves:

In a meeting, 1.80 feels fair. It is a lie relative to both contracted geometries. It sits in neither band (the corridor between 1.58 and 2.01 is empty). Averaging does not shrink risk; it hides a $12,000 invisible error behind one smooth float. Downstream systems will treat 1.80 as a measurement.

You are not fighting elegance. You are fighting a $12,000 invisible error wearing the costume of a single number.

Keeping the pipeline honest

We keep both recipes because departments speak different languages. Risk may need neighborhood panic (Recipe A) for limits and hedges. Client reporting may need the schedule (Recipe B) for fair-value language. An honest system does not say "models agree around 1.80." It says:

Under the neighborhood frame, 1.50 [1.42, 1.58]. Under the contract frame, 2.10 [2.01, 2.22]. We are not blending them. State your business purpose so we can log which frame you choose today.

That is what later chapters mean by rulers, as-of honesty, powered eval, and Clearance rights: software must refuse midpoint peace, not manufacture it.

Job identity is part of the mark. A morning risk file may still speak spread language, because that's the sensor risk actually trusts when panic is moving fast. A client fair-value call may speak schedule language instead, because that's the contract math the client's own auditors expect to see. A hedge residual has to name which mark it is measured against, spread-frame or curve-frame, because the P&L of a hedge only means something relative to one specific ruler. The 1.80 midpoint tries to serve all three jobs, risk, client reporting, and hedging, by serving none of them honestly.

One morning (short)

09:05-09:40. Peer spreads move. Curve points update. Freeze what was knowable at the decision clock. Tomorrow's ten-year already in today's table cheats the exam.

09:47. Dual marks on the ticket: 1.50 and 2.10, bands, frame ids. No blend field.

10:06. Bad agent writes mark = 1.80 from a "consensus" tool. Good agent cannot.

11:30. Client call chooses a frame on the record, or keeps both clouds visible.

Soft leakage that looks like calm

There is a second, subtler failure, more dangerous than the midpoint because nobody is proposing a wrong number on purpose. Thursday afternoon a feature looks "better behaved" -- model error shrank. The cause was a peer scaler fitted on a window that included Friday's co-break day, information the Thursday 14:00 decision was not allowed to use in aggregate form. Every row timestamp still looked legal on its own. The world looked calmer because the yardstick itself was secretly stretched, not because the desk actually knew more.

That is why later chapters refuse to treat "all timestamps ≤ t" as the whole information contract. The CEH-001 fight is not only spread versus curve. It is also honest rulers versus quiet cheats that manufacture midpoint-shaped peace.

Where AI helps, and where it peacemakes

• Scorer: proposes marks inside a frame. Fine. Training against a blended 1.80 label is not.
• Agent: drafts tickets. Useful if tools obey the clock. Dangerous if it can spend or pull post-clock data.
• Peacemaker: consensus dashboards that love 1.80 with pretty intervals covering neither honest cloud.

What "stock-trading desk" means here

This series is not a rates-trading detour that wandered into equities. The showcase is an equity research and trading desk that must mark and hedge single-name risk. Shares, options, portfolio loss cards, and Clearance rights to spend are the centre.

Discount curves appear because any stock-linked schedule still needs a coin for time. Spread language appears because distress-linked single-name structures are marked with peer information when they do not print.

If your mental picture is only "predict the stock," you will misread half the gates later on. If your mental picture is "mark a never-printed single-name exposure with two contracted rulers while an AI tries to average them," you will recognise every later chapter as the same morning wearing thicker tools.

Numbers to keep

How to read what comes next

Whenever a later chapter says "refuse 1.80," return here: two panes, two jobs, one empty corridor, and a position whose honest statement is a pair.

Next: time-infrastructure tricks in Recipe B, and as-of honesty when the market moves hard after 09:05.

Book: Chapter 1 in Rulers Before Models.