Guilty to the Value of Ten Pence

For a century and a half, English juries kept thieves off the gallows by writing down a price everyone in the room knew was false. In 1833 a peer stood up in the House of Lords, said so, and produced a count. This page sets his count beside an extraction from 197,751 trials in the Proceedings of the Old Bailey, and finds that 91.4% of the values extracted from verdicts are one of four numbers, each sitting one step below a line that carried a death sentence.

Artificial Wasteland · 28 July 2026 · source: the Proceedings of the Old Bailey, 1674–1913

On 28 June 1780 Richard Dorrell stood at the Old Bailey accused of stealing five iron bayonets from the Tower. The armourer who identified them was asked what they were worth. "Pretty near two shillings," he said, "rather under that." Pressed again, at a low valuation: "not more than one shilling, as they are damaged." The jury then returned its verdict, and the clerk wrote it down:

GUILTY of stealing to the value of 10 d. Old Bailey Proceedings, t17800628-3

Ten pence. Not eleven, which would have served just as well arithmetically. Ten. The line that mattered ran at a shilling: above twelve pence the theft was grand larceny, a felony that carried death at common law, and below it the theft was petty larceny, which did not. The jury had not valued the bayonets. It had chosen a side of a line and then worked backwards to a number, and the number it worked back to was the one juries had been working back to since before anyone in that room was born.

Blackstone had already given the practice its name, eleven years earlier, in the Commentaries:

It is true, that the mercy of juries will often make them strain a point, and bring in larciny to be under the value of twelvepence, when it is really of much greater value: but this is a kind of pious perjury, and does not at all excuse our common law in this respect from the imputation of severity, but rather strongly confesses, the charge. Blackstone, Commentaries on the Laws of England, Book IV, ch. 17 (1769)

So the phenomenon is not a discovery. It was visible to everyone at the time, and Old Bailey Online, whose transcription this page is built on, says so plainly on its own background pages: "In many cases the reduced value of the goods was blatantly implausible, but the jury engaged in such 'pious perjury' with broad judicial and public support in order to save lives."

What follows is the measurement. Where exactly the false numbers fall. How far they sit from the true ones. Which number a given defendant received, and why that one. What happened on the day Parliament moved the line. And, at the end, a check against the only other person known to have counted this, who did it by hand in 1833.

Four numbers

Take the values this parser extracts from verdict spans across the corpus: 5,388 of them across 240 years. Then count how often each number appears. This is not every printed value: the parser misses bare wording such as "Guilty 39" and reads "value 4 s. 11¾ d." as 4s. The figures below describe the extraction, including its limits.

These are not prices. They are the places you could stand and not be hanged, and they correspond one to one with Acts of Parliament. A fifth number, 9s, appears 48 times and belongs to a threshold that vanished in 1691.

Notice also how far under each line they sit. The 1699 shoplifting Act bites at "the Value of Five Shillings or more", so 4s 11d was safe; juries wrote 4s 10d. Blackstone puts petty larceny at goods "to that value, or under" twelvepence, so a shilling itself was safe; juries wrote ten pence. A number chosen to be just barely legal would hug the line. These numbers are not hugging a line. They are a formula, handed down.

Two distributions from the same trials

The obvious objection is that stolen property in eighteenth-century London might simply have been worth about ten pence rather often. A comparison is possible in trials where the parser finds both indictment values and a value in a partial verdict. These are values from the same trial, but the code does not match the goods or charges to one another.

In those 1,992 trials, the 3,377 extracted indictment values spread across 165 distinct values and carry 5.34 bits of information. Their commonest figures are 2s, 1s, 5s and 10s. Taking the first parsed value from the first partial verdict with a value in each of those trials gives 1,992 verdict values. They use 29 distinct values and carry 1.45 bits. The four commonest values in this subset, 10d, 4s 10d, 39s and 9s, cover 96.3% of them.

The verdict distribution carries less than a third as much information as the indictment distribution. That is a comparison of two differently assembled pools, not two prices for the same coat. The code also sums each trial's parsed indictment values and divides by its selected verdict value. Among the 1,867 trials where the sum is larger, the median ratio is 6.5× and the largest is 888×. Neither is a measured markdown on matched goods.

The largest ratio comes from a 1689 trial: the code totals £37, including a £30 necklace of which the defendant was discharged. The verdict reads "Guilty of the former so the value of 10 d." The arithmetic is reproducible; treating the whole £37 as the goods covered by that verdict is not justified.

The instrument

Set the dials yourself. Above the axis, in blue, is what prosecutors alleged the goods were worth. Below it, in gold, is what juries said. The red lines are the statutory thresholds. Nothing here is modelled or smoothed: each bar counts occurrences of a value in the parser's output. Extracted values above 120 shillings are off the end of the axis; they are 3.2% of all extracted values.

Different offences, different distributions

The offence categories let us compare where the extracted values fall. The thresholds were not general. Each belonged to a specific statute covering a specific offence, and Blackstone lists them in a single passage: benefit of clergy is denied "1. In all larcinies above the value of twelvepence, from a church, or from a dwelling-house, or booth, any person being therein … 3. In all larcinies to the value of 40 s. from a dwelling-house, or it's outhouses, without breaking in … 4. In all larcinies to the value of 5 s. from any shop, warehouse, coachhouse, or stable."

Blackstone's omitted second item gives housebreaking a 5s threshold, the same value as shop theft. The 40s rule concerns dwelling-house theft without breaking in. Those distinctions matter: a statute does not assign every defendant in an offence category a single verdict value. The table compares distributions across years, grouped by the first recorded offence.

Among the extracted threshold-adjacent values, pickpocketing is concentrated at 10d and shoplifting at 10d and 4s 10d; each has 39s exactly once. The category for theft from a specified place is concentrated at 39s, while 99s predominates in its 1828–1834 selection. Housebreaking has 43 at 10d, 34 at 4s 10d and 34 at 39s. The association fits the historical account of juries working below legal thresholds; this grouping does not establish that statutes were the only influence.

1827: Parliament moved the line, and the number moved with it

The cleanest experiment in the corpus was run by Parliament. The Larceny Act 1827 (7 & 8 Geo. 4 c. 29) provides at section 1 "That this Act shall commence on the First Day of July in the present Year", abolishes the distinction between grand and petty larceny at section 2, and at section 12 makes it capital to "steal in any Dwelling House any Chattel, Money, or valuable Security to the Value in the whole of Five Pounds or more; every such Offender, being convicted thereof, shall suffer Death as a Felon."

Five pounds is a hundred shillings. A jury genuinely appraising goods would not notice. A jury reverse-engineering a number from a threshold would have to move, and would have to move when the Act commenced on Sunday, 1 July 1827.

In this extraction, the last all-39s sitting before the first recorded 99s was 31 May 1827. The first sitting with an extracted 99s value was 12 July 1827. The statute took effect between those two dates. By the following year the older number was gone entirely. It recurs in the extracted series on 29 October 1829 and 17 October 1833; even 25 October 1827 was another all-39s sitting.

The prosecutors, meanwhile, were pulling the other way. Under the old Act an indictment had to allege 40 shillings or more to reach the capital charge at all, and in dwelling-house cases between 1713 and 1826 the value 40s appears in indictments 217 times against 5 for 39s. One shilling, contested from both ends of the courtroom: the prosecutor puts the goods at forty, the jury puts them at thirty-nine, and the shilling between is a man's life.

Somebody else counted this, in 1833

On 2 August 1833 the House of Lords debated robberies in dwelling houses. Lord Suffield rose to argue that the severity of the law was corrupting the juries administering it, and he had done the arithmetic:

Such a state of the law induced Juries to violate their oaths. He held in his hand a list of 555 perjured verdicts delivered at the Old Bailey in fifteen years, beginning with the year 1814, for the single offence of stealing from dwellings, the value stolen being in these cases sworn above 40 s., but the verdicts returned being to the value of "39 s. only." If required, he could produce the name of every one of these 555 prisoners … It deserved remark, that when the Legislature raised the sum to 5 l. in June, 1827, the Juries at the same time raised their verdicts to 4 l. 19 s., still keeping it low enough to save the offender's life. Lord Suffield, HL Deb 2 August 1833, vol 20 cc277–83

Fifteen years beginning with 1814 is 1814 to 1828. Over that window, the pipeline counts 524 verdict spans tagged theftunder40s, which is 94.4% of Suffield's 555. The definitions differ: the code does not require a dwelling-house offence, an alleged value above 40s, or an explicit 39s verdict. Hansard records Suffield holding a list; it does not establish how that list was compiled.

Of those 524 tagged spans, 349 yield 39s, 23 yield another value and 152 yield no parsed value. The 152 are already inside the 524; they do not locate or explain the difference of 31 from Suffield's figure. He also says he could name "every one of these 555 prisoners"; the code's associated defendant tally is 617, without matching those people to his list. These figures are a comparison, not a replication of his count.

His arithmetic is not flawless. He offers 11,064 jurymen alongside the 555 verdicts, and those two figures cannot both describe twelve men on each of 555 juries, which would be 6,660. We have not been able to reconstruct what he was counting.

Four months earlier, in the Commons, Mr Lennard had made the same observation without the list:

Under the old law, Juries were in the habit, in a great number of cases, of finding the property stolen under the value of 40 s.—namely, 39 s. And what did they do now? Why, they found it to the value of 99 s. Mr Lennard, HC Deb 16 April 1833, vol 17 cc156–78 em-dash-ok

Which is exactly the shift plotted above, described from the floor of the House of Commons six years after it happened, and located here in the extracted session series. Lennard also gives the housebreaking evasion figure as 4s 11¾d. This parser cannot establish that figure's frequency: it misreads fractional pence. In the extracted housebreaking values, the mode is 10d, with 43 occurrences; 4s 10d and 39s each occur 34 times. That distribution does not establish that Lennard's examples were wrong.

What the shilling bought

Restrict to trials that leave no room for ambiguity: theft from a dwelling house, while the 40-shilling threshold stood, with exactly one offence, one verdict and one sentence recorded, so the sentence is unmistakably the consequence of that verdict on that charge.

The comparison is not a controlled experiment and should not be read as one: a partial verdict removed the mandatory death sentence as a matter of law, so the near-zero figure on the left is close to definitional, and juries chose whom to spare, which is selection of the strongest kind. The number that carries the weight is the one on the right. 40.5% is what waited for a defendant the jury did not extend the formula to. Four in ten went to the gallows list.

When the reason went, the practice went

If the four numbers were a custom of valuation, dismantling the Bloody Code would leave them untouched. If they were a device for defeating the death penalty, they should die when the death penalty for theft died.

At its height in the 1720s, 63.1% of all guilty theft verdicts at the Old Bailey were partial verdicts of this kind. Nearly two convictions in three came with a fictional price attached. The practice then declines for a century as capital statutes are repealed one by one, and after 1834 it stops. Not tapers: stops. In the 20,338 guilty theft verdicts of the 1840s there is not one.

Which is the whole argument, made by the data rather than about it. A custom does not end on a date. A workaround ends when the thing it was working around is gone.