The local grab must leave one future

You are about to take the star hire, the exclusive distributor, or the supermarket that wants your SKU off every other shelf. From here the object looks best. That is not the test.

The test is what remains after you take it. If you cannot name one still-acceptable future that survives the grab, the grab fails. You do not get to argue the rule in general.

Why these three models

The decision is whether the next hire, channel, or customer is a locally shiny grab that burns every remaining good path. The features that fire are a many-peaked search space, a commitment that is costly to undo, and a sequence in which each grab makes the next one cheaper to take.

Three lenses, three outcome types. The terrain gives a complex answer: local search walks to a nearby peak, and the honesty test asks whether this step still belongs to at least one good remaining configuration. Options give a random answer: taking the object spends the right to wait. Path dependence gives a cycle answer: once complements pile up, the next grab arrives pre-justified. Complex, random, cycle.

LOOP is folded. The distributor has a next period, but the lever is whether a second channel remains legal, not a patience inequality. The shiny-object habit sits inside the first card as the reason the locally best object looks obvious. It adds no lever of its own.

1. The grab fails on one missing future

Jeff Edmonds opens the chapter by saying every two-year-old knows the greedy algorithm: to get what you want, start grabbing what looks best.¹ The rest of the chapter is one question. After you commit to the object that looks best, have you gone wrong.

The counterexample is a coin set. Denominations are 4, 3, and 1. The amount is 6. Greedy commits to a 4 because it makes the most progress at a cost of one coin. What remains is 2, which you fill with two 1s. The optimum was two 3s. By taking the 4 you deleted the only optimal remaining solution.¹

Edmonds does not ask you to keep every good future. He asks you to keep one. Instead of requiring all optimal solutions to contain the “best” object, prove that at least one does. Committing to that object may eliminate some of the optimal solutions. It must not eliminate all of them.² The loop invariant is that we have not gone wrong: if a solution exists, at least one optimal solution is still consistent with the choices made so far.² You may burn a few bridges. You may not burn all of them.²

A weaker invariant fails. “No conflicts yet” among the commitments so far is not enough, because those commitments may already have backed you into a corner where future conflicts are inevitable.⁸ A hire who is polite in the interview and who will not share the function is a clean first step with a blocked leftover. The test is the leftover, not the absence of a fight this week.

The conflicts in the book are ordinary. If you take the pool, the yard is too full for the other prizes. A lion might impress the neighbours and eat the dog.³ A head of product who will not share the function with a complementary ops lead is the 4-cent coin. An exclusive Nairobi distributor who forbids dukas and agents is the same coin. Each is locally best. Each can delete the only remaining mix that still works.

The honesty test is five lines. Run it on the live object.

• Name the object you are about to grab.
• Name the conflicts it creates: who it blocks, which channel it forbids, which customer it makes illegal.
• List the remaining solutions that are still legal after the grab.
• Pass only if at least one of those is still optimal, or still acceptable by a bar you wrote down before you saw this object.
• Fail on one counterexample. Edmonds kills a greedy rule with a single coin set. So do you.

Acceptable is the operator version of optimal. You are deciding whether a coherent team, channel mix, or customer set still exists after this signature. If the only remaining futures are “we hope they become flexible” or “we will renegotiate later,” you have already taken the 4.

Cormen, Leiserson, Rivest and Stein put the same split in different words. A greedy algorithm makes the choice that seems best at the moment. That heuristic does not always produce an optimal solution.⁴ It is legal only when the greedy-choice property holds: you can assemble a globally good solution by making locally good choices. In the other regime the next choice depends on solved subproblems, so you do not make the first choice before you know what the leftovers are worth.⁴ How you split a scarce seat, shelf, or exclusive now changes the value of every later split. If you cannot exhibit greedy-choice, stop calling the heuristic optimal.

Levinthal’s organisational version is why the locally best object keeps winning the meeting. On a rugged terrain, neighbourhood search walks uphill until it hits a local peak, then stops. The number of those peaks rises as attributes interact.⁵ The honesty test is how you check, before the step, whether this neighbour still belongs to at least one configuration you would accept.

Rugged landscape, the search lens

• Assumes: the next object is a local step on a many-peaked terrain, and one good remaining configuration must survive it.
• Fits because: the hire, channel, or customer looks best from here and conflicts with other objects.
• Breaks when: the problem has greedy-choice, so the local grab is global.
• Evidence: grade B for the terrain; Edmonds’s predicate is a method, not a field result.
• Counteracts: treating “best available this week” as proof of safety.
• May reinforce: refusing a grab that still leaves a coherent team standing.

2. Taking it spends the right to wait

The second lens prices the same grab as an exercise.

When a firm makes an irreversible investment, it exercises, or kills, its option to invest. It gives up the possibility of waiting for new information that might change the desirability or the timing. That lost option value is an opportunity cost and belongs in the cost of the move.⁶ An irreversible grab is a different transaction from a reversible one with the same sticker.

An exclusive distributor agreement kills the option to put the same SKU through agents next quarter. A star hire with a veto on complementary senior seats kills the option to complete the team another way. Each can still be the right move. Each has a second price: the futures you can no longer enter.

Ask whether a reversible version of this same grab still gets you most of the benefit. A contractor for ninety days, with no veto on the next senior seat, is the hire without the 4-cent lock. A non-exclusive city agreement at a worse unit rate is the channel without the lock. A supermarket listing that leaves kiosks legal is the customer without the lock. You pay a visible premium, or accept a smaller first order, to keep one remaining future intact. That premium is the option. Teams cut it because it shows up on the invoice and the thing it buys does not.

Pindyck notes that even spending on new workers can be partly irreversible.⁶ The irreversibility sits in the conflicts the person is allowed to create, not in the salary. Write those conflicts into the offer, or refuse the offer. A star who will work alongside a complementary lead is a 3. A star who will not is a 4. Same CV. Different remaining futures.

Real options, the wait lens

• Assumes: the grab is costly to reverse, and new information may still arrive about the leftover mix.
• Fits because: exclusivity, veto seats, and flagship demands are hard to unwind once signed.
• Breaks when: reversal is cheap (month-to-month, no exclusivity, a contractor you can end).
• Evidence: grade A on the killed option; transfer to a hire or channel is analogical.
• Counteracts: comparing two grabs on first-year revenue only.
• May reinforce: delay dressed up as option value when nothing new is coming.

3. The next grab will be cheaper, and worse

Run the test now, not after the object has been in the building for a year.

Arthur’s dynamics of choice under increasing returns have a small set of properties that transfer. The aggregate outcome can be inefficient even when each choice is rational. Once shares lock in, ordinary later incentives do not always move them. Small historical events are not averaged away. They can determine the path, and the process remembers its small-event history.⁷

The first exclusive or veto hire is a small historical event. It feels like the obvious next step. Complements then accumulate: staff who know that distributor, a pick-pack process built for that supermarket’s carton, a roadmap that assumes that head of product will approve every adjacent hire. Each complement raises the cost of leaving, so the next grab arrives pre-justified. After you have taken the 4, the remaining menu is 1-cent coins. Taking another 1 looks cheap. It is how you finish going wrong.

Last year’s lock-in is already sunk. The live question is the next object. Arthur’s warning is that the next object will look safer than it is, because the path has already begun to remember. The honesty test gets harder to pass the longer you wait to run it, which is why it belongs on this week’s offer letter.

A second salesperson in a seat that already works is activity-selection. You can take the locally best person and leave the rest of the calendar intact. A first exclusive in a city where you still need dukas, agents, and one supermarket is the other regime. The split you make now changes the value of every later split. Name which problem you are in before you congratulate yourself for being decisive.

Path dependence, the sequence lens

• Assumes: early grabs attract complements, and those complements select among later equilibria.
• Fits because: staff, process and roadmap pile up around the first exclusive or veto seat.
• Breaks when: the grab attracts no complements, or you can replay from kept inputs.
• Evidence: grade B. Arthur’s lock-in is an allocation model, not your city.
• Counteracts: treating the next grab as independent of the last one.
• May reinforce: fatalism about a path you can still refuse to extend.

GEER: write the remaining futures before you sign

Position the boundary first. Inside: the object, the conflicts it creates, and the futures that remain legal after it. Outside: the person’s charm, the distributor’s brand, the supermarket’s footfall. You have agreed to be surprised by those. Remaining futures are a stock. The grab is an outflow. You cannot set the stock. You can only refuse an outflow that drains it to zero.

The cheapest lever is a page. Before the offer letter or the exclusivity clause, write the object, the conflicts, and the remaining legal futures. If the remaining list is empty, the grab fails. Time cost is an hour. Reversal cost is zero, because you have not signed.

The next lever is a rewrite. Convert the same object into a version that leaves one future intact: drop the exclusivity, drop the veto, shorten the term to a quarter, make the listing additive rather than exclusive. You keep most of the first-year revenue. You keep one path.

The expensive lever is a long jump: refuse this object and search outside the neighbourhood. Size that valley in cash and months before you walk into it. Most weeks you will not need it. The page and the rewrite handle the common case.

No-lever flag: if the object is already signed and the complements have piled up, this ensemble can still inventory what remains. It cannot unmake the last grab. It can stop the next one.

RADAR: what to do before the offer letter goes

Do now, T+0 to T+3, effect visible the same week. Take the live object. Write the five lines of the honesty test. Name one remaining future in a sentence a colleague could run. If you cannot, the letter does not go. Reversible, cheap, and dominant across every story about how good this person or this channel is.

Hedge, T+7 to T+14, premium is the whole loss of a slightly worse first deal. Convert the grab before it is signed. Non-exclusive language. No veto on complementary seats. A contractor or a ninety-day city agreement. Cover must be live before the signature, because the option dies at the signature, not at the first bad quarter.

Defer and trigger, size fixed now. Do not sign an exclusive or a veto seat this week. Pre-commit the observable: you sign the irreversible form only when you can still name, in writing, one remaining team or channel mix that meets the bar you set before this object appeared. The trigger is that sentence. Decide the bar now, while nobody is waiting in reception.

A portfolio whose page arrives after the signature is misordered. The page is the DO NOW. The signature is the thing it is allowed to stop.

CHAIN: what a shiny grab usually becomes

Name the mode first. This is lock-in. Early choice plus increasing returns plus complements. The shape is a short random walk, then a hardening that later incentives do not easily reverse.

Has a rule changed, has an actor entered or left, has a measurement become a target? If the person in the seat is now compensated on being the only senior in the function, or the distributor is measured on exclusive volume, the historical relationship between “best available object” and later flexibility describes a different process. The base rate does not transfer. If nothing broke, continue.

Match the class on structure: first exclusives, first veto seats, first flagship demands that rewrite the offer. Arthur’s allocation model says that once the process locks, later policy does not always unlock it, and small early events select the path.⁷ The directional base rate is unflattering. The locally best first exclusive tends to become the shape of the city. The locally best veto hire tends to become the shape of the team. Founders credit the object’s quality. The sequence did the work.

Subtract what would have happened anyway. Some teams would have been narrow without this person. Some cities would have had one channel without this clause. Credit the grab only with the futures it uniquely closed. If the object uniquely closes the last remaining mix you would accept, it fails, whatever it would have done for this quarter’s revenue.

Matrix-break flag. If the clause is already non-exclusive, the term is already a quarter, and the hire has no veto, the irreversibility assumption is not in force. Run a lighter check and move. Applying this article’s full brake to a reversible contractor is a misuse.

The coin this ensemble cannot weigh

These three models can tell you whether a remaining future still exists on paper. None of them can tell you whether the future you named is actually executable. A sentence that says “we can still hire an ops lead” is not the same as an ops lead who will take the job after you have already installed a veto. The named leftover can be a courtesy. The models cannot see the courtesy.

They also cannot see the room. Refusing a locally impressive person or a branded distributor reads, in the meeting, as a lack of ambition. That social cost does not appear in Edmonds, Pindyck, or Arthur, and it is often the reason the 4-cent coin wins.

One property none of them contains: the complementary person you did not hire, watching. They update on whether this company still has a seat for them. That update is an information event the grab produces and no card tracks.

The one action that survives the ignorance: before the next offer letter or exclusivity clause leaves the building, write one remaining future that is still acceptable after the grab, in a sentence a colleague could execute without you. If the sentence will not stand, the grab fails. One counterexample. The next object, tested once.

The person who has to run this is the one who will sign. Everyone else in the room is holding the shiny object. Bring the remaining-future sentence to that meeting. If they cannot name one either, you have your answer, and it cost a page.

Sources and notes

1. Jeff Edmonds, How to Think About Algorithms, Cambridge University Press, 2008, Chapter 16, printed p. 227 (PDF p. 243; printed = PDF minus 16). Making Change, continued. With 4-, 3-, and 1-cent coins and amount 6, “the optimal solution contains two 3-cent coins. We go wrong by greedily committing to a 4-cent coin.” The chapter opens at printed p. 225: “Every two-year-old knows the greedy algorithm. In order to get what you want, just start grabbing what looks best.” Landing page (200, title and 19 May 2008): cambridge.org.
2. Edmonds, How to Think About Algorithms, CUP 2008, Chapter 16, printed p. 229. “At Least One Optimal Solution Remaining: Instead of requiring all optimal solutions to contain the ‘best’ object, what we need to prove is that at least one does.” The chosen loop invariant: “we have not gone wrong. If there is a solution, then there is at least one optimal solution consistent with the choices made so far.” He adds that it is acceptable to burn a few bridges so long as you do not burn all of them. Same landing page as note 1.
3. Edmonds, How to Think About Algorithms, CUP 2008, Chapter 16, printed p. 226, Example 16.1.1 The Game Show. Prizes conflict: “if you take the pool, then your yard is too full to be able to take many of the other prizes. A lion might impress your neighbors, but it might eat your dog.” Used for the conflict step of the honesty test, not for a hiring result.
4. Thomas H. Cormen, Charles E. Leiserson, Ronald L. Rivest and Clifford Stein, Introduction to Algorithms, 3rd ed., MIT Press, 2009, Chapter 16, pp. 423 to 425. “This heuristic strategy does not always produce an optimal solution.” The greedy-choice property and optimal substructure are the two key ingredients. A greedy algorithm makes its first choice before solving any subproblems; dynamic programming usually does the reverse. Catalogue record (200, third edition, September 2009): dl.acm.org.
5. Daniel A. Levinthal, Adaptation on Rugged Landscapes, Management Science 43(7), 1997, pages 934 to 950. Neighbourhood search continues until organisations reach a local optimum; the number of local optima rises as the terrain becomes more rugged; long jumps are frequent early and rare once fitness is high. Open-access course mirror (200, PDF, claim present): NTNU course PDF. Publisher abstract: pubsonline.informs.org. The transfer is structural. No claim is made that a specific company has a measured ruggedness.
6. Robert S. Pindyck, Irreversibility, Uncertainty, and Investment, Journal of Economic Literature, Vol. XXIX, September 1991, pages 1110 to 1148. “When a firm makes an irreversible investment expenditure, it exercises, or ‘kills,’ its option to invest.” The lost option value is an opportunity cost that belongs in the cost of the move. Investment in new workers may be partly irreversible. Author PDF (200, claim present): web.mit.edu. Working-paper mirror: NBER w3307.
7. W. Brian Arthur, On Competing Technologies and Historical Small Events: The Dynamics of Choice under Increasing Returns, IIASA Working Paper WP-83-090, September 1983. Published as “Competing Technologies, Increasing Returns, and Lock-In by Historical Events,” Economic Journal 99(394), 1989, pages 116 to 131. Four properties include inflexibility (market shares become locked in) and non-ergodicity (small historical events are not averaged away and can determine the path). The process “remembers its small-event history.” Open-access working paper (200, claims present): pure.iiasa.ac.at. Author bibliography confirming the 1989 publication of WP-83-90: sites.santafe.edu/~wbarthur.
8. Edmonds, How to Think About Algorithms, CUP 2008, Chapter 16, printed p. 228. The “No Conflict Yet” candidate invariant (commitments so far do not conflict) fails because those commitments “may have backed the algorithm into a corner so that future conflicts are inevitable.” Used in section 1 to distinguish a clean first step from a leftover that still exists.

A note on what is deliberately absent. This piece does not tell you which remaining future is the best one, and it does not score the star hire’s actual skill. The honesty test is a fail rule. It removes objects that leave the leftover set empty. It does not rank the objects that pass.

Joshua Agonya Pi’Rwot, Founder.