The Clause the Pencil Can Reach

I.The First Known Thing — the covenant saved by deletion

contract law · restraint of trade and the severance of a bad clause

An employer who wants a departing executive kept out of the market for a time must write a covenant, and the common law starts from the position that such a covenant is void as a restraint of trade unless it goes no further than the employer's legitimate interest requires. A covenant that reaches too far is unenforceable. The question that then arises is not whether the covenant is bad but what happens to the rest of the contract, and to the parts of the covenant that were perfectly reasonable.

what the court will and will not do about it

An English court will not rewrite the bargain. It will not narrow a five-year restriction to a reasonable twelve months, or read a worldwide clause down to the county where the employer trades, because to do so would be to make an agreement the parties did not make and to invite employers to draft as widely as they like in the knowledge that a judge will trim. What it may do is delete. If the offending words can be struck out — run a blue pencil through them — and what remains still reads as English, still stands as a promise, then what remains is enforced. The test is mechanical and it is about the text: removal without addition, without modification, without rearrangement.

the case that fixed the modern test

In Tillman v Egon Zehnder [2019] UKSC 32 the Supreme Court took up a non-competition covenant that forbade the employee to be “concerned or interested in” any competing business. The words “or interested” caught a minor shareholding, and so made the whole covenant unreasonably wide. The court severed those two words, left the remainder standing, and enforced it. It set out what severance requires: that the unenforceable provision be capable of removal without the need to add to or modify the wording that remains; that the remaining terms continue to be supported by adequate consideration; and that the removal not generate any major change in the overall effect of the restraints, with the burden of showing that on the employer. In doing so it endorsed the more permissive approach of Beckett Investment Management Group v Hall [2007] and departed from the older and stricter reading of Attwood v Lamont [1920].

the consequence for those who draft

What follows from a deletion-only rule is a drafting practice rather than a litigation tactic. The same restriction, in substance identical, survives or dies according to its grammar: written as one undivided sentence it goes entirely, because nothing can be taken out without rewriting; written as a series of separately numbered, self-standing restrictions, each with its own period and its own scope, the bad ones are struck and the good ones remain. Careful drafters therefore break covenants into independent units and cascade the durations, and they do this before any dispute exists, without knowing which unit will later be held too wide. Jurisdictions differ on the underlying rule — some courts will reform an unreasonable term to a reasonable one, some will only delete, some strike the covenant whole — but where deletion is the only remedy, the structure of the text decides the outcome, and the seriousness of the overreach does not.

II.The Second Known Thing — the series with a liability in it

drug discovery · lead optimisation and multi-parameter trade-offs

Walk to a hall where a chemical series is being made into a drug. A series is a scaffold with vectors: a common core that carries the interaction with the target, and positions around it at which substituents may be varied. Optimisation is the work of moving several properties at once — potency, solubility, metabolic stability, permeability, and a set of liabilities that must be brought under a limit rather than maximised: inhibition of the hERG potassium channel, a structural alert for genotoxicity, an inhibited or induced cytochrome, an unacceptable clearance. The chemistry proceeds by matched changes: make one substitution, measure what moved. Matched molecular pair analysis, set out for pharmaceutical properties by Leach and colleagues in 2006 and made routine by later algorithms and open implementations, is the field's instrument for reading those single changes across large datasets, and matched molecular series extend it to the ordered sets a chemist actually makes.

what happens when a liability appears

A liability discovered in the first analogues is not by itself a verdict. Some liabilities have well-mapped chemistry: hERG inhibition, as Jamieson and colleagues catalogued in 2006, is associated with a basic nitrogen and with lipophilicity, and is often reduced by lowering the base's pKa, by trimming logP, or by introducing an acidic or polar group in a tolerated direction. The practical question a project asks is whether that reduction can be had without losing the potency the same region provides. Sometimes a single substitution at a peripheral vector drops the liability by a log unit and leaves the target activity untouched. Sometimes every change that touches the liability touches the binding, and the two can be traded but not separated, so that gains have to be recovered elsewhere in the molecule — or the core has to be replaced, which is not optimisation of a series but the start of another one.

and how the decision is actually taken

Series are ranked, and most are stopped. The stopping decision is taken early, on a small number of compounds, and the quantities that enter it are magnitudes: how potent, how soluble, how far the hERG IC₅₀ sits from the projected free exposure, how large the ligand efficiency and the lipophilic efficiency. Those are properties of the compounds in hand. The field also knows that single changes do not always behave: nonadditivity analysis, formalised by Kramer in 2019 over double-transformation cycles, measures how often the effect of one substitution depends on what is present at another position, and finds that a material fraction of well-measured cycles are not additive. What is not computed, and not recorded anywhere in the decision, is whether the liability in this series sits on a vector that can be changed on its own.

III.The Bridge

no severance case cites a screening cascade

The halls share nothing. Severance belongs to contract and to the law of restraint of trade; lead optimisation to medicinal chemistry. No review spans them, and where chemistry has borrowed from outside it has borrowed from optimisation theory and from statistics, not from doctrines of drafting. Nor is the bridge the familiar counsel to design the liability out, which every chemist would recognise and which says nothing about when that is possible. What the two share is a narrower structure: a composite object, one part of which is found defective after the object was made, where the object survives only if that part can be deleted without rewriting what remains — so that survival is decided by the separability of the parts and not by the gravity of the defect, and separability was fixed when the object was drafted, before anyone knew which part would fail.

X: what saves a composite is the seam, not the size of the fault

Call it excisability. Where a composite must survive the condemnation of one of its parts, and where the only permitted repair is deletion — because rewriting the remainder makes a different object, a contract the parties did not agree to or a scaffold the project has not validated — the outcome is governed by whether a seam runs between the condemned part and the part that does the work. So the quantity that predicts survival is the separability of the defect-bearing part from the load-bearing part, and not the severity of the defect; and because separability is a property of how the object was composed rather than of the defect, it is fixed at drafting and can be estimated before any defect is known. Three things follow that the chemical hall does not currently say. First, the magnitude of an early liability, which is what triage uses, is close to the wrong quantity: a series with a hundred-fold hERG problem on a peripheral vector may be in better condition than one with a ten-fold problem written through its pharmacophore. Second, excisability is measurable in the language the field already speaks — a single-point transform that moves the liability while leaving potency within noise is a clean severance; a change that requires compensating substitution elsewhere is reformation, which the law of contract will not perform and which in chemistry costs a second optimisation; a liability inseparable from the core is an unseverable covenant, and the remedy is a new scaffold. Third, because nonadditivity is exactly the failure of a change at one position to act independently of another, the field's own measure of nonadditivity around the liability-bearing vector is an estimate of how far the blue pencil will reach in that series.

IV.The Question That Becomes Possible

neither discipline would ask this alone

The chemical hall asks how bad the liability is and whether the team can fix it, and answers with experience. The legal hall has no molecules, but it learned, from covenants that died of their punctuation, that when the only repair is deletion the fate of a composite is settled by its seams and can be arranged in advance by whoever composes it. Neither asks whether the series that fail on a liability are the ones whose liability was large, or the ones whose liability had no seam — nor whether that distinction can be drawn from the first few dozen compounds, before the money is spent.

the excisability hypothesis

Can a lead series be scored, from its first few dozen measured analogues, for the excisability of a named liability — the extent to which a single-position change moves that liability by a useful margin while leaving target potency within assay noise and without requiring compensating change at another position — and does that score predict whether the series later reaches a compound meeting both the potency and the liability criterion, better than the magnitude of the liability at the same point in the series, and beyond what lipophilic efficiency and the compounds' potency already explain?

why it would matter

Because series triage spends most of its evidence on magnitudes, and magnitudes are what a small set of early compounds reports least reliably about a scaffold. If the hypothesis holds, a project has a quantity computable from data it has already generated — no new assay, no new chemistry — that says whether the liability in front of it is a clause that can be struck or a term written through the whole agreement, and the two deserve opposite decisions at the same measured potency. It would also give the chemistry of scaffold selection something to aim at. A drafter of covenants writes in separable units because a defect will appear somewhere and only separable units survive it; the corresponding chemical practice would be to prefer, among comparable hits, the scaffolds whose liability-prone features sit on independently variable vectors — a criterion applied at the moment of choosing the series, when it is nearly free, rather than discovered two years later when it is not.

V.How It Could Be Proven — or Broken

a candidate must be breakable, or it is merely poetry

An intercognix candidate earns nothing until it names its own test. This one needs no new experiment. It needs series with their chronology intact — compounds, dates, target potency and one liability measured on a common assay — which is what a company's electronic notebook holds and what the public literature approximates in published series and in the deposited sets behind them.

the honest status

Note the narrowness. The claim concerns one liability at a time, in series large enough to contain matched single-position changes, and one outcome — whether a compound satisfying both criteria was ever made — chosen because it is recorded rather than because it is the only thing that matters. It does not claim that magnitudes are irrelevant: a liability far inside the projected exposure is fatal whatever its seams, and a liability that never mattered needs no severing. It claims only that between those extremes, where most series sit, the field ranks by the quantity that is easier to measure rather than by the one that decides, and that the deciding quantity is available in the same data. It is also worth saying what does not transfer. A court's pencil is exact, its object is text, and the remainder after deletion is the same remainder it was before; a molecule has no grammar, and every deletion is a synthesis with consequences of its own. What transfers is the observation that where repair is restricted to removal, composites should be graded on their seams and composed with their seams in mind — not the certainty that makes the deletion clean on paper. The whole thing rests on a comparison an existing dataset would allow: whether the series that failed on a liability were the ones whose liability was large, or the ones whose liability could not be reached on its own. If they were the first, this is a well-dressed restatement of triage as it is already practised, and should be said so plainly. Until that comparison is made, this remains exactly what its stamp says: a candidate. A structural echo between two halls of the library, waiting for someone with the tools of both to walk the corridor.