Guide · Market microstructure · Updated 2026-09-19

The book ranks you by the instant you arrived, and adding quantity throws that away

A limit order is usually described as an instruction about price. That is the half written on the ticket. The half that decides whether the order ever trades is a rank the trader never sees and cannot ask for: its position inside the line of everything else resting at that price. Most traders spend their attention on the price and then destroy the rank by accident, with a keystroke that feels like housekeeping.

The rule, stated exactly

The matching engine keeps two keys. Price first: the highest bid and the lowest offer are served before anything worse. Time second: among orders at the same price, the one the system accepted earliest is matched first, then the next, to the back. The protocol states it from the entry side. An incoming order is tested against the book, and if it does not match it is placed in the appropriate book with its price and its time stamp. That stamp is the rank, and nothing else about the order changes it.

Two consequences follow, and both are routinely got wrong. The first is that your order does not trade because the price reached it. It trades because everything ahead of it was executed or cancelled while the level was still standing. A price can touch your limit, print there, and leave without reaching you. The market did not skip you. There were 6,000 shares in front of you and only 2,000 traded.

The second is that what you hold is a per level object, not a per order one. Your rank exists only in relation to one price, so moving the price does not carry it with you.

The rank is also invisible. The depth window publishes total quantity at each price, not your place inside it. The only honest estimate of your position is the quantity resting at your price when your order was accepted, which is why a trader who never records that number cannot tell whether a fill was a reward for patience or a warning.

What a modification does, field by field

A modification is not an edit. It is a request, and the engine answers it against a fixed table of what it is permitted to keep. Below is that rule as published in the trading system protocol for the capital market segment. The derivatives protocol carries the same table in the same words.

Read the disclosed quantity line twice, because it shows what the rule is really testing. An iceberg style order displays part of its size and hides the rest. Changing the displayed portion costs priority only when the change increases what the book can see, and toggling the attribute costs priority only when that increases the displayed quantity. The engine is not asking whether you touched the order. It is asking whether you increased your claim on the front of the line.

The stop loss line has a practical edge. A stop managed by repeatedly modifying the trigger, which is how most discretionary trailing stops are implemented, is re-stamped on every adjustment. That is free while the order sits in the stop loss book, because an untriggered stop is not in the price queue at all. It stops being free the moment the stop triggers and becomes a limit order in a queue, which is the moment the trader cared about.

The asymmetry is principled, not arbitrary

Reducing quantity cannot disadvantage anyone. Every order behind yours moves closer to the front and nothing ahead of you is affected, so there is nothing to re-rank.

Increasing quantity is a different transaction. The shares being added have no claim on the time you arrived, because at that time you had not decided to trade them. An exchange could split the order, leave the original quantity in place and stamp only the addition at the back. The published rule does not do that. It re-stamps the whole order, so the priority of the shares that were already there goes as well.

A price change loses everything for a cleaner reason: the queue you were ranked in is not the one you are joining. That includes improving your price. An order moved to a better bid does not arrive at the front of the new level. It arrives at the back of it, behind whoever was already there.

The habit this makes expensive

Two patterns account for most of the priority thrown away in practice. The first is topping up: a working order for 500 shares near the front, and a decision to make it 800. Increasing the quantity sends all 800 to the back. A separate order for the extra 300 leaves the original 500 exactly where it was.

Take the level in the computed model below, with 10,000 shares resting and a tick of 0.10. The difference in expected capture between the front and the back is 0.4785 of a tick per share. On the 500 shares already at the front, the careless version of that modification costs ₹23.92. That is small enough to be invisible on a contract note and large enough to matter as a habit: forty of them in a session is ₹957, and none of it appears as a loss anywhere, only as fills that did not happen and fills that happened at a worse moment.

The second pattern is chasing the touch. The bid moves, the trader modifies a resting order to follow it, and the order is re-queued at the new level whatever else was true about it. Traders who work with this rule treat that as two decisions: cancel, because this price is no longer where I want to be, and then place, knowing that placing means starting at the back. The move is then made once, on purpose, rather than twenty times by reflex.

Deliberate cancel and replace is not free either, for a regulatory rather than a mechanical reason. The exchanges run an order to trade ratio framework and levy penalties where ratios are excessive, so an algorithm that re-places on every quote update is spending that allowance. A professional execution stack is built to modify where modification is free, cancel where it must, and do neither on reflex.

What a place in the line is worth

Priority is worth something, and the something can be computed once the assumptions are written down. Here they are in full. They are assumptions, not measurements.

Take one price level at the touch and work in units of 100 shares. One event happens at a time and the four kinds are drawn independently in fixed proportions. Twenty eight parts in a hundred are ordinary executions, each taking one unit off the front. Seventy one parts are cancellations of one unit by somebody already at the level, and total depth is held constant, so a cancellation falls ahead of you with probability equal to your share of the queue. Eight parts in a thousand are the level being abandoned, meaning the market moves away and trading no longer happens there, which ends the episode with no fill. Two parts in a thousand are a sweep, in which one order takes whatever is left and the quote then sits one tick worse, which ends the episode with a fill.

One more assumption sets the units of value. The spread is one tick, so resting at the bid captures half a tick against the prevailing mid when the quote holds and gives up half a tick when the fill was the sweep that moved it. Everything below follows from those numbers alone.

The first column is what traders think they are buying. The other two decide whether it was worth buying. At the front, the order fills almost always and almost never because somebody ran it over. At the back of the same level it fills about a third of the time, and about half of those fills are the sweep.

This is why a queue position behaves like an option rather than an asset. You hold the right, not the obligation, to trade at that price ahead of everybody behind you, and you can walk away for nothing by cancelling. What makes it valuable is not your presence at the price. It is that the people behind you cannot get there first.

The last column is the point. The deeper the level, the more a front position is worth, because what you hold is the right to go ahead of everything in it. It does not rise without limit: past a few hundred units the back of the queue is already close to hopeless, so further depth adds nothing and the value flattens near 0.58 of a tick. The rule of thumb that priority is worth more in thicker books describes the left half of that column and stops before the flattening.

The same arithmetic gives a threshold. On a level of 30,000 shares with this event mix, expected capture from joining the queue turns negative once more than 6,300 shares rest ahead. Past that point the order is paying for the privilege of being adversely selected. Change the assumptions and the threshold moves, which is the honest way to use a number like that: not as a level, but as a reason to estimate your own.

The fill you waited for is the fill to be suspicious of

There are exactly two routes to the front of a queue and they mean opposite things. The first is attrition: orders ahead of you executed or were cancelled, the quote held throughout, and your turn arrived at a price that was fair when you posted it and is still fair now. The second is displacement: one participant took the entire level, your order was inside what they took, and the quote now sits one tick worse. You were not chosen. You were in the way.

The second kind is concentrated at the back of the queue, because an ordinary trade never reaches that far. Only something large does, and something large arriving at a price level is the most reliable available signal that the price is about to leave it. Deep in a queue you are not trading with the market. You are trading with the part of it that is in a hurry, and in a hurry is expensive.

This inverts how most execution reviews are read. A high fill rate on passive orders is taken as evidence that the prices were good. It is equally evidence that the orders sat where informed flow wanted to trade. The diagnostic is not the fill rate but what the mid did in the seconds after each fill, which any trader keeping an order log can measure and almost nobody does. It feeds straight into cost modelling in a backtest, where a passive strategy that assumes a fill whenever price touched its limit has assumed both a queue position it never held and an absence of adverse selection its real fills would have contradicted.

The tick decides how long the queue is

Queue length is not a property of a stock. It is a property of the grid the stock trades on. The tick sets how many distinct prices exist in a given range, and the same resting interest divided among fewer prices makes each queue deeper.

India runs a price linked tick grid in the capital market segment, introduced in May 2024 with a single step at ₹250 and extended in March 2025 into the full slab structure below with effect from the trade date of 15 April 2025. The tick of a security is reviewed monthly and determined from the closing price on the last trading day of the previous month, and the same determination carries into the corresponding stock futures.

The interesting part of a slab grid is not the slab. It is the boundary, where a one paisa move in the underlying price changes the structure of the market that trades it.

Follow the ₹1,000 boundary through to priority. The tick doubles, so the same edge measured in ticks is worth twice as much in rupees. The same resting interest spreads over half as many prices, so each queue is about twice as deep, and the computed value of a front position rises with depth. Together, a front position worth ₹0.0239 per share on the finer grid is worth ₹0.0562 on the wider one, about 2.3 times as much on the same event mix.

The mechanism runs the other way too, and that is why a fine tick produces short queues. When the tick is a large fraction of the price, the spread is pinned at one tick and there is no room to improve a quote: the only ways to trade sooner are to wait or to cross and pay a full tick, so priority approaches the spread in value and people queue for it. When the tick is small, anyone who wants to be ahead of you prices one tick better for a fraction of a basis point. Nobody queues long behind a position that can be leapfrogged for almost nothing, and the book rearranges into many shallow levels. Both regimes run in the same market on the same day at different slabs, which is why the quoted spread and the depth behind it move together rather than independently.

The monthly review has a consequence worth stating. A security closing one month on one side of a slab boundary and the next month on the other has the same business, the same shareholders and the same flow, and a different queue structure from the first trading day. Stock options took a version of the same change from the trade date of 3 November 2025, when the tick for contracts whose underlying trades below ₹250 moved to one paisa.

Where the queue stops deciding anything

Price and time priority is the rule for continuous trading. It is not the rule for the whole day, and the exception is larger than it was a year ago.

The pre-open session has never used it for allocation. Orders are collected without matching and one equilibrium price is computed, chosen for maximum executable volume and then narrowed by minimum imbalance and nearness to the reference price. Everyone who matches trades at that price whatever they bid. The full cascade is worked through in the pre-open auction guide, including the detail that matters here: an unmatched limit order carries into the continuous session with its original time stamp, so a queue position earned at 9:01 is real from 9:15 onward.

The larger change is at the other end of the day. From 3 August 2026, for stocks on which derivative contracts are available, continuous trading ends at 3:15 pm and the closing price comes from a closing auction running to 3:35 pm. Any page describing the queue deciding fills until 3:30 pm now describes only half the market, because stocks without derivative contracts do continue in continuous trading until 3:30 pm. For twenty minutes the same exchange runs two allocation rules side by side.

Read the order of service carefully, because it reverses the intuition the rest of the day builds. A market order entered at 3:24 pm is matched before the limit order queue is reached at all. A limit order resting since 9:20 am, carried into the auction with its original time stamp and correctly ranked ahead of every limit order entered inside the session, is still served after that market order. Priority was not abolished. It was demoted to a tie break underneath a different rule.

Three further details change what a resting order can do there. Orders outside the band of plus or minus 3 per cent around the reference price are rejected or cancelled. Disclosed quantity orders are not accepted at all, so an iceberg strategy has no expression in the last twenty minutes. A trade struck in the auction cannot be cancelled by request. What survives intact is the reduction rule: inside the auction, as in the continuous session, lowering quantity without touching the price keeps the order's place.

What this is actually for

None of this tells anyone what price to bid. It decides something narrower and more durable: whether the execution you assumed when you designed a strategy is the execution you will get. A passive strategy that has never estimated its queue position has assumed the best case twice over, a fill whenever price touched and no adverse selection in the fills it got, and both assumptions flatter it in the same direction. That is why they survive a backtest together.

The skill is not memorising the modification table, although that takes ten minutes and pays for itself. It is knowing which of your instructions to the market are free, which are expensive, and which are expensive in a way that will never appear on a statement. That judgement is what the choice between order types rests on, and it is learned by working through mechanisms rather than collecting rules.