Which hadith count as canon, and how was that decided?

Summary

A hadith is a report of something Muhammad said, did, or approved.  Hundreds of thousands are in circulation, they contradict each other, and Islamic law is built on the ones judged sound.  Which reports count, and who decided, is therefore close to the centre of Islam.  This article describes the classical Muslim answer as accurately as it can be described, then tests it.

• "Canon" means two different things here, and blurring them causes most of the confusion. One meaning is a list of books — the six standard Sunni collections. The other is a grade on an individual report — sound (sahih), fair (hasan), or weak (da'if). A report can sit in a canonical book and still be criticised; a report outside all six books can still be graded sound. The list and the grade are separate systems.
• Nobody voted. The six-book list was not fixed by al-Bukhari (d. 256/870) and Muslim b. al-Hajjaj (d. 261/875), and their own contemporaries attacked the project. A four-book list is attested in the tenth century, a rival six in the eleventh that includes Ibn Khuzayma instead of Ibn Majah, and the list that finally won was written down by Muhammad b. Tahir al-Maqdisi (d. 507/1113), about 250 years after Bukhari died. Competing lists of five and six were still being published a century after that.¹
• The decisive step was subtler than the list. Jonathan Brown argues that al-Hakim al-Naysaburi (d. 405/1014) turned the two Sahihs from famous books into a shared yardstick for authenticity that both traditionists and their rationalist opponents would accept.¹ Once they were a measure, the law schools adopted them in turn, the Hanafis last, by the 1300s.²
• The famous sifting ratio is a conflation, and both Muslim apologists and Christian polemicists repeat it. "Bukhari examined 600,000 hadith and kept 7,275" pairs two numbers from different sources that count different things. The 600,000 comes from a praise-narrative first attested in al-Khatib al-Baghdadi (d. 463/1071), two centuries after Bukhari, and Brown's translation is careful to say narrations, not distinct sayings.³ The 7,275 is Ibn al-Salah's own count of the finished book including repeats, and he immediately warns that a single hadith with two chains is often counted twice.⁴ Nothing in the tradition claims that 592,725 specific reports were examined and rejected.
• The classical method is a source test, not a content test — and the tradition had a theological reason for that. Ibn al-Salah's definition of a sound hadith is entirely about the chain and its transmitters.⁴ Ignaz Goldziher's verdict has to be met head-on: "Muslim critics have no feeling for even the crudest anachronisms provided that the isnad is correct."⁵ Brown's defence is that the critics did use content but could not say so, because admitting it would have handed the argument to the rationalist Mu'tazila.² Both are partly right, and the receipts for each are below.
• Most hadith gradings a reader meets online are twentieth-century, not medieval. The grades printed on sunnah.com for Abu Dawud, al-Tirmidhi, al-Nasa'i and Ibn Majah come from Muhammad Nasir al-Din al-Albani (d. 1999) and from the Darussalam editions, and sunnah.com says so on its own about page.⁶ Al-Albani re-graded thousands of reports, authenticating many the classical critics had rejected and rejecting many al-Bukhari and Muslim had accepted.³ Writing "graded sahih" without naming the grader is a defect.
• The strongest Western criticism is not "the hadith are forged." It is three narrower claims: that the method is formal and prophetic foreknowledge makes anachronism unfalsifiable; that the tradition applied its rigour selectively, which is measurable in al-Tirmidhi's own self-grading; and that on the tradition's own account the requirement to name a chain only begins around 690 CE, roughly sixty years after Muhammad's death.³
• Harald Motzki's correction is real and has to be given full weight. Working the chains and the wordings together, he showed that G.H.A. Juynboll's flagship result — that the transmitter Nafi' was Malik's invention — was wrong, and that the report in question genuinely goes back to Nafi'.⁷ He stated the limit in the same sentence: the original wording "can be only partially reconstructed." The method dates chains, not texts.
• There are hadith in Bukhari and Muslim that academics date to long after Muhammad, and they can be named: the man kadhaba warning against lying about the Prophet (Juynboll), the wala' and Barira traditions (Schacht), the prohibition of lamenting the dead (Goldziher), and the "three mosques" pilgrimage hadith (Goldziher). The last of these is also the clearest case of the Western method over-reaching, and this article says so.²
• The comparison with New Testament transmission is usually made badly on both sides. These are two different machines. The New Testament is a written-copy tradition whose failure mode is copying error and whose fix is comparing branches of surviving manuscripts. Hadith is an oral tradition with claimed chains whose failure mode is false attribution and whose fix is cross-examining transmitters. Neither mechanism does the other's job.

Verdict: the Sunni hadith canon was not decreed;  it accumulated, and it was standardised by scholars three to five centuries after the reports it contains were supposed to have been spoken.  The grading science behind it is a genuine discipline with reproducible outputs — three independent third/ninth-century critics agree about transmitters roughly six times in seven⁸ — and it was audited from inside, most famously by al-Daraqutni (d. 385/995), who criticised 217 narrations in the two Sahihs themselves.¹  But it is a test of who said it, not of whether it happened, and it cannot reach behind the moment chains were first demanded.  The best modern work pushes reliable dating back to roughly 40 to 60 years after the events for a small number of traditions,⁹ which is a real achievement and is also far short of what the traditional claim requires.  A hadith graded sound is a report whose route survived a serious examination.  It is not a transcript.

• Weaknesses of this article's own case, stated plainly. Three of the sources it leans on could not be obtained: Michael Cook's Early Muslim Dogma, Behnam Sadeghi's statistical "Travelling Tradition Test," and a complete copy of the Dickinson translation of Ibn al-Salah — the available scan drops about 35 of its 327 pages, so Ibn al-Salah is cited by Category number first and page second. The Western critique rests heavily on arguments from silence, and Juynboll admitted his own crucial arguments were "practically all argumenta e silentio."¹⁰ The critical results are demonstrations on a handful of traditions, not a survey of the corpus, and Brown's warning applies to this article too: attitudes towards hadith authenticity "are necessarily based more on our critical worldview than on empirical fact."² A Christian arguing from that critique should notice that several of the same moves, applied with the same confidence, would cut into the Gospels.

The claim being tested

Muslims do not usually state a single claim about hadith.  Several distinct propositions travel together, and they need separating before any of them can be tested.

1. The transmission claim. Reports of Muhammad's words and deeds were passed down through named chains of transmitters (isnad, "supporting"), and those chains are real records of who heard what from whom.
2. The screening claim. Muslim scholars of the second and third Islamic centuries developed a rigorous science for testing those chains, and used it to separate sound reports from weak and forged ones.
3. The canon claim. The output of that screening is preserved in a recognised set of books, above all Sahih al-Bukhari and Sahih Muslim, whose contents can be relied on.
4. The authority claim. What survives that screening is binding — a second source of revelation alongside the Quran, since the Quran itself commands obedience to the Messenger.

These are separable.  A person can accept (1) and (2) while rejecting (3), which is roughly where Harald Motzki and Gregor Schoeler stand as historians.  A person can accept (1), (2) and (3) while disputing (4) for reports carried by only one chain, which is where Muhammad Abduh (d. 1905), Rashid Rida (d. 1935) and Mahmud Shaltut (d. 1963) stand as Muslims.  A person can reject the lot, which is what the Quranists do.²

This article tests (1), (2) and (3).  It does not test (4), which is a question of Islamic legal theory rather than of history.

Two terms have to be nailed down before anything else, because English writing on this subject constantly slides between them.

• Canon as a list of books. Al-kutub al-sittah, "the Six Books," is the standard Sunni set: the Sahih of al-Bukhari, the Sahih of Muslim, and the four Sunan of Abu Dawud (d. 275/889), al-Tirmidhi (d. 279/892), al-Nasa'i (d. 303/915) and Ibn Majah (d. 273/887).³
• Canon as a grade on a report. A hadith is sahih (sound), hasan (fair), or da'if (weak), and in the worst case mawdu' (forged). This is a judgement about one report, made by a named critic, and different critics reach different judgements about the same report.

The two systems are not aligned.  Being in a canonical book does not make a report immune from criticism — al-Daraqutni criticised hundreds of narrations inside the two Sahihs, and Ibn al-Jawzi (d. 597/1201) put at least three of them in a book of forgeries.¹  Being outside the six books does not make a report weak;  Ibn Majah's Sunan contains 1,339 reports found in none of the other five, and of those 1,339 Muhammad Fu'ad Abd al-Baqi's reckoning found 428 sound, 199 fair and 712 weak or worse.⁶  The most-quoted "obviously fabricated" hadith in polemics are usually in neither category:  they are reports the Muslim critics themselves rejected.

The strongest case for the Muslim claim

Stated at full strength, from Muslim sources, before any evaluation.

Islam is the only major religious tradition that built a discipline for auditing its own oral reports, and built it early.  No comparable apparatus exists in Judaism, Christianity, Buddhism or Zoroastrianism.  Christians did not compile biographical dictionaries rating the memory and honesty of everyone who passed on a saying of Jesus.  Muslims did — tens of thousands of entries, generation by generation, city by city.

The rule the discipline runs on was stated by Abdallah b. al-Mubarak (d. 181/797) and is quoted in every survey of the field:  "The isnad is part of religion, if not for the isnad, whoever wanted could say whatever they wanted.  But if you ask them, 'Who told you this?' they cannot reply."²  That is a demand for sourcing, and it is the same demand a modern historian or journalist makes.

The method has three steps, and Brown's reconstruction of what the early critics actually did — as against what the later manuals say they did — is the clearest statement of it.³

1. Demand a source. A report with no chain is not evidence.
2. Rate the transmitter. Two criteria: moral uprightness ('adala) and, much more importantly, accuracy (dabt). Accuracy was tested by collecting everything a man narrated and checking it against what his fellow students of the same teacher narrated. A transmitter with no two known students citing him is majhul — unknown — and his reports are rejected automatically.
3. Seek corroboration. A parallel narration through the same source (mutaba'a), or an independent report from another Companion (shahid).

Three features of this system deserve more credit than critics usually give them.

It was not a piety test.  Heresy was largely tolerated if the man was accurate.  Al-Bukhari includes two hadith in the Sahih through Imran b. Hittan, a Kharijite who wrote a poem praising the murderer of the caliph Ali;  Muslim narrates through the Shiite Adi b. Thabit.  Yahya b. Ma'in (d. 233/848) said of one Shiite transmitter that he was "trustworthy, sincere, and Shiite, but who would rather fall from the sky than misrepresent half a word."²  The criterion was competence, not orthodoxy.

It worked backwards, which is the critical posture.  Brown's observation is worth quoting because it answers the charge that the critics simply assumed what they wanted:  "Muslim critics worked backwards in time when authenticating hadiths… For a Muslim hadith critic, a hadith was at first just an unverified claim;  its isnad began with the person who told him the hadith."²

It produced reproducible results.  This is the strongest single piece of evidence in the whole defence, and it is quantitative.  Scott Lucas compared the reliability grades that Ibn Sa'd (d. 230/845), Ibn Ma'in and Ahmad b. Hanbal (d. 241/855) assigned independently to the same transmitters.  He found 87% agreement between Ibn Sa'd and Ibn Ma'in, and 86% between Ibn Sa'd and Ibn Hanbal.⁸  Three critics of one generation, working from overlapping but not identical materials, agreed about six times in seven.  That is what a discipline looks like, as opposed to a set of private preferences dressed up as scholarship.

And the tradition audited its own canon in public.  Al-Daraqutni's Kitab al-tatabbu' criticises 217 narrations from the two Sahihs — 100 from Muslim, 78 from al-Bukhari, 32 shared.¹  Ibn Ammar al-Shahid found 36 defective narrations in Muslim.¹  Al-Tirmidhi graded the hadith in his own book.  Ibn al-Jawzi (d. 597/1201) included narrations from both Sahihs in his book of forgeries.¹  A tradition that wanted to hide its problems would not have written them down and copied them for a thousand years.

The last argument on this side — that a fraud on the scale the critics require would have had to fool a hostile, fractured, multi-continental scholarly community for four centuries — is stated in full in the section on the strongest Muslim responses below, after the criticism it answers.

The vocabulary, defined precisely

Every hadith has two parts.  Goldziher's anatomy is the standard one and both terms are Arabic words a reader will meet constantly:  the isnad is "the chain (silsila) of attestors, from its originator to its last transmitter," and the matn is "the actual wording of the saying."⁵  A printed hadith gives the chain first and the saying second:  A told me, from B, from C, from D, that the Messenger of God said…

The standard reference work for the technical vocabulary is the Muqaddima of Ibn al-Salah al-Shahrazuri (d. 643/1245), translated into English by Eerik Dickinson as An Introduction to the Science of the Hadith.  It is organised into sixty-five numbered "Categories," and Category numbers are the stable way to cite it.  Every definition below is from that book.

Four of these carry most of the weight.

Sound (sahih).  Ibn al-Salah's definition is one sentence and it is entirely about the chain:  "The sound hadith is a 'supported' hadith (al-hadith al-musnad), the isnad of which coheres continuously through the transmission of one upright and accurate person from another up to its point of termination.  The sound hadith can be neither 'anomalous' (shadhdh) nor 'defective' (mu'allal)."⁴

Fair (hasan).  This is the grade the tradition itself never settled.  Ibn al-Salah quotes the definitions of al-Khattabi and al-Tirmidhi and then says of both:  "All of this is ambiguous and inadequate.  Nothing in the definitions of Tirmidhi and Khattabi distinguishes the fair hadith from the sound."⁴  He then builds his own two-part account.  It is worth pausing on:  the tradition's most authoritative systematiser found its second-highest grade ill-defined, and said so in print.

Weak (da'if).  Weak is a residual category, not a verdict of falsehood:  "Every hadith in which the traits of the sound hadith and the fair hadith mentioned above do not come together is a 'weak' hadith."⁴  This is why classical scholars, including Ibn Hanbal, could still use weak hadith for moral exhortation.  A weak report is one whose route failed the test, not one shown to be false.

Forged (mawdu').  Here the tradition states its own limit with unusual clarity.  Ibn al-Salah:  "A hadith can be known to be forged only by the acknowledgement of its forger or something equivalent to his acknowledgement.  Sometimes scholars come to recognize a forgery by inference from the state of the transmitter or that of the report he transmits."⁴  A confession, or an inference from the man.  Not a textual test.

One more distinction runs through every modern Muslim argument about hadith, so it has to be introduced here.  A report transmitted by so many independent chains in every generation that collusion is impossible is mutawatir, "massively transmitted";  everything else is ahad, a "single" report, even if it has several chains.  The translator's note in Ibn al-Salah gives the classical definition:  its transmitters "were so numerous in each generation that under normal circumstances it would have been impossible for all of them to enter into a conspiracy to put a lie in circulation."⁴  Mutawatir reports were held to yield certainty ('ilm);  ahad reports only strong probability (zann).

Two facts about that distinction matter.  First, it is not native to hadith criticism — Brown traces it to Mu'tazilite theology, entering Sunni hadith science through al-Khatib al-Baghdadi (d. 463/1071), more than four centuries after Muhammad.²  Second, the number of chains required was never agreed;  Mu'tazilite proposals ran from four to seventy.  Sayyid Ahmad Khan (d. 1898) applied a strict standard and concluded that only five hadith met it.²

The five conditions of a sound hadith

Ibn al-Salah's one-sentence definition unpacks into the five conditions every later manual repeats.

1. Continuous chain (ittisal al-sanad) — each transmitter actually received the report from the one above him, all the way back.
2. Uprightness ('adala) of every transmitter.
3. Accuracy (dabt) of every transmitter.
4. No anomaly (shudhudh) — it does not contradict what better-attested transmitters relate.
5. No hidden defect ('illa).

Three qualifications belong with those conditions, and all three come from Ibn al-Salah himself on the same page as the definition.  They are more modest than the popular presentation of the science.

• Meeting the standard does not make a report certainly true: "The standard of the sound hadith does not require that it be definite that the hadith is sound in reality, since this standard admits the hadith that a single upright transmitter is alone in relating."⁴
• Failing it does not make a report false: "when they say regarding a hadith that it is 'unsound,' this is not a definite statement that it is actually a falsehood, since it may in reality be a truth. All that is meant is that its isnad is not sound according to the aforementioned standard."⁴
• And on ranking chains against each other: "we think it is better to refrain from judging any isnad or hadith to be the absolutely most sound, although a number of the authorities in hadith have ventured into that morass and their opinions were therefore contradictory."⁴

There is a further point that a fair account cannot leave out.  Al-Bukhari and Muslim never wrote down their conditions.  The five-condition formula is a reconstruction by Ibn al-Salah and Ibn Hajar al-Asqalani (d. 852/1449), four to six centuries later, working backwards from what the two compilers actually did.  Brown states it plainly:  "It is difficult to know exactly how early hadith critics defined sahih, since they were very laconic in their works."²  The earliest compiler to state a standard, Ibn Khuzayma (d. 311/923), gave four conditions and not five — no shudhudh, no 'illa.

That does not mean the standard was invented.  It means the standard was inferred, and that anyone reading the two Sahihs is reading the output of criteria their authors left unstated.

The six books, and who decided there were six

The Six Books are these.  The counts are Brown's, and they are counts of narrations, which is not the same as counts of distinct sayings.³

Brown's own two books give different totals for Ibn Majah — 4,485 in Hadith and about 4,341 in The Canonization of al-Bukhārī and Muslim, which is also sunnah.com's figure.  Both are printed above rather than reconciled, for the same reason the ten-years and sixteen-years compilation figures are both printed below.

The two Sahihs overlap on 2,326 traditions and share about 2,400 narrators.  Together, all six books contain roughly 19,600 hadith, or about 35,000 counting repetitions.²  That is a much smaller corpus than the six-figure numbers in circulation suggest, and the gap between the two is the subject of the next section.

The canon was not fixed by Bukhari and Muslim, and their contemporaries objected to the idea.  Abu Zur'a al-Razi (d. 264/878), a leading critic of the same generation, said of the two of them:  "These are people who wanted prominence before their time, so they did something of which they could boast."³  His specific worry about Muslim was that a book of sound hadith would arm heretics:  they would "see that they can respond to a hadith that we use as proof against them by saying 'That is not in the Sahih!'"  Under that pressure, "al-Bukhari and Muslim defended themselves by saying that their books did not include all the sahih hadiths in circulation."²  The compilers themselves did not claim to have produced a canon.

The list took about 250 years to settle, and it settled differently in different places.  Brown's trace is the fullest available in English.¹  The earliest attested selection is four books, not six:  al-Bukhari, Muslim, Abu Dawud and al-Nasa'i.  Sa'id b. al-Sakan of Egypt (d. 353/964), asked which books to study, "entered his house and reemerged with four books in his hands.  'These are the foundations of Islam,' he said."³  Ibn Manda of Isfahan (d. 395/1004–5) names the same four.  Abu Bakr al-Bayhaqi (d. 458/1066) lists six, but his sixth is Ibn Khuzayma.  The list that won was set out by Abu al-Fadl Muhammad b. Tahir al-Maqdisi (d. 507/1113), who "described the Six Books as the Sahihayn, the Jami' of al-Tirmidhi, and the Sunans of al-Nasa'i, Abu Dawud and Ibn Majah" — and who also wrote the first comprehensive book on the criteria of the six.¹  Rivals persisted for another century:  al-Saghani (d. 650/1252) put al-Daraqutni sixth, the Andalusian al-Saraqusti (d. 524/1129) put Malik's Muwatta' sixth, and al-Silafi (d. 576/1180), al-Hazimi (d. 584/1188–9) and al-Nawawi (d. 676/1277) counted only five.¹  The unchanging core across every version is the same four books Ibn al-Sakan pulled off his shelf.³

The decisive event was not the list but the yardstick.  This is Brown's central thesis, and it is the answer to the question "how was that decided."  The two Sahihs became canonical not when someone put them on a list but when scholars started using them as a shared measure of what counted as authentic:

"One scholar in particular inherited the body of scholarship on the Sahihayn and harnessed the two works as a new measure of authenticity for evaluating reports attributed to the Prophet.  Al-Hakim al-Naysaburi (d. 405/1014) recognized that the Sahihayn possessed tremendous polemical value as common measures of hadith authenticity that met the requirements of both the transmission-based scholars whom he championed and the Mu'tazilites whom he bitterly opposed."¹

The mechanism was a genre of book called ilzamat — collections of hadith that al-Bukhari and Muslim "should have included."  Al-Hakim's Mustadrak, with about 8,800 hadith, was the largest.  To argue that a report met their standard you first had to say what their standard was;  saying it turned the two books into a ruler.  From there the law schools fell in line one at a time — Hanbali and Shafi'i in the early 1000s, Maliki soon after, and "by the 1300s even the hadith-wary Hanafi school had found acknowledging this convention unavoidable."²

Goldziher, whose scepticism about hadith was total, admitted he could not date the moment:  "we cannot establish with chronological accuracy the date which brought the consensus publicus for the two Sahihs to maturity."¹

What canonisation bought, and what it cost.  Ibn al-Salah records the mature doctrine and — unusually for a medieval author — records that he changed his own mind to reach it.  The material both Sahihs agree on is "definitely settled.  Theoretical and certain knowledge occurs through it, contrary to the doctrine of those who deny this, arguing that their agreement does not in principle produce more than the presumption (zann)… Formerly, I had inclined toward this view and regarded it highly.  Then it became clear to me that the doctrine we had chosen in the first place is the correct one."⁴

The certainty there does not come from the chains.  It comes from the community's agreement to receive the books — ijma'.  That is a theological warrant, not a historical one, and it is the door through which consensus enters a system that presents itself as evidential.  On the same page Ibn al-Salah closes the era of independent authentication altogether:  "the way to identify sound and fair hadith is now confined to the consultation of the two Sahihs and other authoritative books."⁴  He is honest about why:  by the thirteenth century live transmission could no longer meet the old standards, so the books had to stand in for the process.

One last fact that belongs here because it is rarely mentioned on either side.  The text of Sahih al-Bukhari itself is not fixed to the letter.  Different students transmitted different recensions of it.  Hammad b. Shakir's recension had 200 fewer narrations than al-Firabri's;  Ibrahim b. Ma'qil al-Nasafi's had 300 fewer — roughly 3% of the book.¹

The 600,000 figure, and why the famous ratio is a conflation

The most repeated fact about Sahih al-Bukhari is that its author sifted 600,000 hadith down to about 7,000.  Muslim apologists use it to show the rigour of the screening.  Christian polemicists use it to show that 98.8% of the corpus was junk.  Both are using the numbers wrongly, and the error is the same error.

Where 600,000 comes from.  It is not a statement by al-Bukhari in any surviving work of his.  It is a report transmitted by al-Khatib al-Baghdadi (d. 463/1071) in his Ta'rikh Baghdad, and repeated by later authors down to Ibn Hajar.¹  Al-Khatib died about two centuries after al-Bukhari.  Brown's own footnote for the claim in his shorter book cites a source later still, Taj al-Din al-Subki (d. 771/1370).  Brown does not call the figure false.  He does something more useful:  he places it in the chapter of his study devoted to what he calls "canonical culture" — the body of praise-narrative that grew up around the books to establish their authority.  The same passage in al-Khatib reports that al-Bukhari performed ablutions and prayed two cycles of prayer before inserting any hadith into the book.

The tradition does not even agree with itself on how long it took.  Al-Khatib's report says ten years.  Brown's own textbook, and sunnah.com's about page, say sixteen.³

Parallel claims exist for the other compilers, in the same genre.  Muslim is said to have selected his Sahih from 300,000, on a report al-Khatib carries through al-Hakim.¹  Abu Dawud is quoted saying "I have written 500,000 ahadith of Allah's Messenger, from which I selected 4800 of the most authentic ahadith for this book."⁶  Ahmad b. Hanbal "claimed he had sifted the contents of his Musnad from over 750,000 hadiths."³

The word that decides the argument is "narrations."  Brown's rendering is careful:  al-Bukhari sifted his book "from a pool of six hundred thousand narrations."³  A narration is a text plus a claimed route.  The same saying, heard from two teachers, is two narrations.  That is how the compilers themselves counted, and a man who watched Muslim do it said so:  Ahmad b. Salama al-Bazzar, with Muslim for fifteen years while he wrote the Sahih, reported that "if he heard the same tradition from two shaykhs, he considered it to be two hadiths."¹

Once that is understood, a 600,000-narration pool is not 600,000 different claims about Muhammad.  It is a pool of chain-and-text pairs in which one saying can appear hundreds of times.  Nabia Abbott's explanation of how the pool got so large is the strongest one on offer.  Early collections were written on expensive papyrus and parchment, and recorded a saying with perhaps one chain.  Cheap paper arrives at the end of the eighth century.  A "hadith" then becomes identified with its chain rather than its text, and collectors began recording every route they met.  Ibn Hanbal aimed at seven narrations per tradition.  "As isnads developed and became interlaced, this number increased even more, while the actual number of Prophetic traditions remained relatively small."²  The scale this reached is hard to overstate:  al-Hasan al-Masarjisi of Nishapur (d. 365/976) compiled a personal collection that "if published today would occupy an astounding 182 volumes."²

Where 7,275 comes from, and why it does not pair with 600,000.  The 7,275 figure is Ibn al-Salah's, and it comes with his own warning attached:

"Bukhari said, 'I have one hundred thousand sound hadith and two hundred thousand unsound.'  The total in his book Sahih is 7,275 hadith, including some repeated hadith.  It has been said that with the omission of the duplicates the total is four thousand."⁴

And immediately after:  this figure "may include the accounts (athar) of the Companions and Followers, and often a single hadith related with two isnads is counted as two hadith."

So the two numbers are not two ends of one audit.  Here is what each actually counts.

Nothing in the Muslim tradition claims that 592,725 specific reports were examined and thrown out one by one.  The honest statement of the screening is much less dramatic and much more defensible:  a large pool of chain-and-text pairs, containing a far smaller number of distinct sayings, was worked down to a book of about 2,600 distinct traditions carried by the chains al-Bukhari trusted.  The "98.8% was fabricated" framing should be retired, and so should the apologetic version of the same arithmetic.

One number that is worth keeping, because the tradition itself produced it and it is a like-for-like comparison:  al-Daraqutni, a rigorous critic, disapproved of 217 narrations in the two Sahihs combined, and of only two of their narrators.  Al-Dhahabi (d. 748/1348) counted no fewer than a thousand weak narrations among Ibn Majah's roughly 4,341.¹  That is a real quality gradient inside the canon, measured by Muslim critics, and it tells a reader far more than the 600,000 ever will.

Who grades hadith today, and why the answer matters

Open sunnah.com — the site almost every English-language argument about hadith is fought on — and look up a report in Sunan Abi Dawud.  Under the text you will see a line like "Grade: Sahih (Al-Albani)."  Look up one in Jami' at-Tirmidhi and you will see "Grade: Sahih (Darussalam)."  Look up one in Sahih al-Bukhari and you will see no grade line at all.

Those three outcomes reflect three different things, and an argument that does not distinguish them is broken before it starts.

Bukhari and Muslim carry no grade because inclusion is itself the claim.  The compiler's decision to put the report in a book titled Sahih is the verdict.  There is no external grader.  That is not a stronger endorsement than a grade line;  it is a different kind of statement, and it is the reason al-Daraqutni's 217 criticisms and Ibn al-Jawzi's inclusions matter — they are what external scrutiny of these books looks like.

The four Sunan carry modern gradings.  Sunnah.com states this on its own about page:

"We are working hard to add grade information for each hadith not in the Sahihain (Sahih al-Bukhari and Sahih Muslim).  At this point we are displaying grade decisions by Shaykh al-Albani and Darussalam (Hafiz Zubair 'Ali Za'i).  Eventually we hope to have grade assignments from several other distinguished muhaddiths such as Shaykhs al-Arna'ut, Ahmad Shakir, and Abu Ghuddah wherever applicable."⁶

A survey of the site's collections found exactly four grader labels in use — Al-Albani, its Arabic equivalent, Darussalam, and Zubair 'Aliza'i — and they are not spread evenly.  Sunan Abi Dawud carries al-Albani's name;  al-Tirmidhi, al-Nasa'i, Ibn Majah and Musnad Ahmad carry Darussalam, which sunnah.com glosses as the work of Hafiz Zubair Ali Za'i.  Al-Arna'ut, Ahmad Shakir and Abu Ghuddah appear nowhere as graders;  they are aspirations listed on the about page.¹¹

So when a debate quotes "Sunan Abi Dawud 585, graded sahih," the grade is the judgement of Muhammad Nasir al-Din al-Albani (d. 1999), a Syrian-Albanian watchmaker turned hadith scholar who died within living memory.  That is a fact about a modern Salafi critic's opinion.  It may be a good opinion.  It is not a medieval consensus, and it is certainly not al-Bukhari's.

How far al-Albani's verdicts depart from the classical tradition is documented.  Brown:

"Al-Albani completed two voluminous series, 'The Series of Weak Hadiths and their Negative Effect on the Muslim Community' and 'The Series of Authentic Hadiths,' in which he revaluates thousands of hadiths.  Many that he authenticates had previously been declared unreliable, and many hadiths that he criticizes had earlier won the approval of great classical critics like al-Bukhari and Muslim."³

His theological position is worth stating because it explains the method.  Against Muslim modernists who hold that single-chain reports cannot establish doctrine, al-Albani held that they can, on the ground that "we cannot distinguish between God and His Prophet."³  And where a hadith's content is troubling, the fault must lie in the chain — the Saudi hadith scholar Abdallah al-Sa'd states the principle as sharply as it can be stated:  "It is impossible for a hadith to have an untrue meaning without there being a flaw in the isnad."³

The same method produces opposite results in different hands, which is a problem for it.  The Moroccan Ghumari brothers — Ahmad (d. 1960), Abdallah (d. 1993) and Abd al-Hayy (d. 1995) — used al-Albani's method and reached the opposite conclusions, embracing Ali as the best of the Companions and defending Sufi grave-visitation.  Abdallah al-Ghumari repeatedly accused al-Albani of heresy;  a Wahhabi hadith scholar returned the compliment.  Brown's verdict on the modern revival of independent grading is that it "has undeniably led to a proliferation of erratic rulings."²

Muslims do not agree about how much weight a grading should carry.  Brown's four-way map of the modern field is the standard one.²  Islamic Modernists and Quranists reject hadith authority in whole or in large part;  Mahmud Abu Rayya's Adwa' 'ala al-sunna al-muhammadiyya (1958) drew at least eight book-length rebuttals, and Fazlur Rahman (d. 1988) granted that Goldziher and Schacht were "essentially correct" about most reports while arguing that the concept of sunnah was authentic.  Modernist Salafis — Muhammad Abduh, Rashid Rida, Mahmud Shaltut, Muhammad al-Ghazali — hold that the Quran is the criterion, that single-chain reports cannot found dogma, and that content criticism must be revived;  Rida rejected the Sahih al-Bukhari report that the sun passes under the earth and prostrates before God's throne because it contradicts astronomy.  Traditionalist Salafis, al-Albani's school, reject weak hadith for any purpose, breaking with classical practice, and re-grade the corpus themselves.³  Late Sunni Traditionalists — al-Kawthari (d. 1952), al-Buti (d. 2013), Ali Jum'a — affirm the classical method wholesale but give jurists the last word on what a hadith means.  Al-Ghazali called the classical science unequalled "in the history of human culture in terms of establishing principles for verification" — and still rejected a hadith al-Albani had authenticated, "In the meat of a cow is disease," as false whatever its chain.²

A skeptic should notice that this is a live, unresolved argument inside Islam about its own second scripture, conducted by people with far more at stake than any outsider.  A Christian should notice that it is conducted at a high technical level and that the participants are not embarrassed by it.

The Shia canon is a different system

Almost everything above is Sunni.  Twelver Shia Islam has its own canon, built on different assumptions, and confusing the two produces bad arguments in both directions.³

Three structural differences come first.  A Shia hadith is a report from any ma'sum — any infallible figure — which means the Prophet or an imam;  whether the chain between an imam and the Prophet is complete "was of no importance."  The Companions are not collectively upright, since anyone who did not support Ali's claim is at best complicit with injustice.  And content criticism has a larger, more open role, because Shia scholarship absorbed Mu'tazilite epistemology in the eleventh century — al-Sharif al-Murtada (d. 436/1044) held that every report attributed to the Prophet or the imams had to be authenticated by reason.

The Four Books are al-Kafi of al-Kulayni (d. 329/939), Man la yahduruhu al-faqih of Ibn Babawayh (d. 381/991), and al-Tusi's (d. 460/1067) Tahdhib al-ahkam and al-Istibsar.  Unlike the Sunni six, these are four different kinds of book.  Brown notes the limit on their standing:  "Their canonicity derives from their widespread acceptance and use, not their infallibility."³  The scale has five grades where the Sunni scheme has three, and the published numbers are remarkable:  of the 15,181 traditions in al-Kafi, 5,072 are graded sound and 9,484, about 62%, are graded weak by Shia scholars themselves.¹²

Two honest notes.  The Shia critical apparatus is later than the Sunni one:  al-Tusi appears to have been the first Imami scholar to grade transmitters systematically, and the first Shia manual of hadith terminology, written in the sixteenth century, "is basically a digest of the Sunni Ibn al-Salah's famous Muqaddima."³  And the counts for the other three books could not be obtained from an academic source, so they are not printed here.

The academic critique, from Goldziher to Motzki

Western study of hadith has run in four phases, and Brown's map of them is the clearest available:  Orientalist (Goldziher, Schacht, Juynboll — doubting individual reports while keeping the outline of early Islamic history);  philo-Islamic apology (Nabia Abbott, Mustafa al-Azami);  revisionist (Patricia Crone, Michael Cook, John Wansbrough, from 1977 — doubting the outline itself);  and Western revaluation (Motzki, David Powers, Andreas Görke, Gregor Schoeler, Fred Donner, from the 1980s — rejecting revisionism, and finding the Muslim tradition more sophisticated than the Orientalists allowed).²

Brown does something else that this article will borrow:  he states the assumptions of the historical-critical method out loud so they can be examined instead of smuggled.  His three are a presumption of doubt about a text's authenticity, a general suspicion of orthodox narratives, and the conviction that a scholar can sort reliable from unreliable material "by identifying which parts of the text served which historical agendas."  Two named tools go with them:  the Principle of Analogy, that human societies always work in essentially the same way, and the Principle of Dissimilarity, that a report embarrassing to the orthodoxy is probably true.³  These are assumptions, not findings.  The Muslim side has its own, stated just as bluntly by Mulla Ali al-Qari:  "it is manifestly obvious that if something has been established by transmission [from the Prophet], then one should not heed any contradiction with sense perception or reason."²

Ignaz Goldziher (d. 1921)

Goldziher's Muhammedanische Studien, volume two, published in Halle in 1890, is where the modern study of hadith begins.  His thesis is two sentences long:

"Closer acquaintance with the vast stock of hadiths induces sceptical caution rather than optimistic trust regarding the material brought together in the carefully compiled collections… [We] will probably consider by far the greater part of it as the result of the religious, historical and social development of Islam during the first two centuries.

The hadith will not serve as a document for the history of the infancy of Islam, but rather as a reflection of the tendencies which appeared in the community during the maturer stages of its development."⁵

His attack on the method is more damaging than his attack on the corpus, and it is the passage every defender has to answer:

"Muslim criticism had chiefly formal points of departure… Traditions are only investigated in respect of their outward form and judgement of the value of the contents depends on the judgement of the correctness of the isnad."⁵

"Nobody is allowed to say:  'because the matn contains a logical or historical absurdity I doubt the correctness of the isnad.'"⁵

"Muslim critics have no feeling for even the crudest anachronisms provided that the isnad is correct."⁵

His worked example on the same page is the pilgrimage stations:  hadith assigning starting points for pilgrims include provision for Iraqi caravans that did not exist in Muhammad's lifetime, and the critics who rejected the Iraqi version rejected it "not because of the anachronism implied but because of the difficulties of the isnad."

He also has a political case.  The caliph Abd al-Malik, wanting to stop pilgrimage to Mecca while his rival Ibn al-Zubayr held it, promoted a substitute pilgrimage to the Dome of the Rock, and "the pious theologian al-Zuhri was given the task of justifying this politically motivated reform of religious life by making up and spreading a saying traced back to the Prophet, according to which there are three mosques to which people may make pilgrimages."⁵

That specific case does not hold, and saying so costs nothing.  Brown's rebuttal is good:  there are other early chains for the hadith that do not run through al-Zuhri at all, and "should we reconsider our conclusion or assume, quite without reason, that these other isnads were forged as well?  The Al-Aqsa Mosque is mentioned in the Quran, so is it so inconceivable that the Prophet would order his followers to pay special attention to it?"²  Brown also catches Goldziher inverting a second piece of evidence.  Ghiyath b. Ibrahim forged a hadith about pigeon-racing to please the caliph al-Mahdi, and was immediately caught — and Muslim sources tell the story precisely as a textbook illustration of forgery detection.  "Goldziher… uses a story designed to illustrate an exception to represent the rule."³

The pattern behind Goldziher's inferences is worth naming, because it is what makes them fragile:  as Brown summarises one case, "the simple fact that Abd al-Rahman served as an Umayyad functionary meant that he must have forged hadiths to support Umayyad causes."²  A rule like that will always find what it is looking for.

Joseph Schacht (d. 1969)

Schacht's The Origins of Muhammadan Jurisprudence (Oxford, 1950) is the most consequential book in the field.  Three formulations carry the argument.

Backward growth of chains, from the summary of results at the front of the book:

"traditions from Companions and other authorities underwent the same process of growth, and are to be considered in the same light, as traditions from the Prophet;  the study of isnads often enables us to date traditions;  the isnads show a tendency to grow backwards and to claim higher and higher authority until they arrive at the Prophet;  the evidence of legal traditions carries us back to about the year 100 A.H. only."¹³

The rule of presumption, which reverses the burden of proof:

"every legal tradition from the Prophet, until the contrary is proved, must be taken not as an authentic or essentially authentic, even if slightly obscured, statement valid for his time or the time of the Companions, but as the fictitious expression of a legal doctrine formulated at a later date."¹³

And the famous verdict, continuing on the same page:

"We shall find that the bulk of legal traditions from the Prophet known to Malik originated in the generation preceding him, that is in the second quarter of the second century A.H., and we shall not meet any legal tradition from the Prophet which can be considered authentic."¹³

A correction is needed here, and it cuts against the polemical use of Schacht.  The version of that sentence that circulates in apologetics, in polemics, and in a good deal of secondary scholarship inserts the word positively:  "which can positively be considered authentic."  That word is not in the text.  Two independent scans of the Oxford printing agree on the shorter wording, and a search of the whole book for "positively" returns six hits, none of them this sentence.¹³  Note which way the error runs:  the real sentence is stronger than the misquotation, because "positively" would soften it into a claim about proof rather than about authenticity.  Anyone quoting Schacht should quote what he wrote.

Schacht's method for dating a tradition is an argument from silence, and he knew it:  "The best way of proving that a tradition did not exist at a certain time is to show that it was not used as a legal argument in a discussion which would have made reference to it imperative, if it had existed."¹³  He defends it with a line from al-Shaybani about the Medinese — "unless the Medinese can produce a tradition in support of their doctrine, but they have none, or they would have produced it" — and he concedes a counter-case a page or two later.

The common link is his most productive idea, and his own statement of it is far more cautious than his reputation:

"The original promoter N.N. would have provided his tradition with an isnad reaching back to an authority such as a Companion or the Prophet, and this higher, fictitious part of the isnad would often acquire additional branches… But N.N. would remain the (lowest) common link in the several strands of isnad… the existence of a significant common link (N.N.) in all or most isnads of a given tradition would be a strong indication in favour of its having originated in the time of N.N."¹³

The reasoning is simple.  Draw the chains of a hadith as a tree.  If a dozen chains all pass through one man in the second generation and then fan out below him, the natural reading is that he is where the report entered circulation.  Schacht notes that Muslim scholars had spotted the phenomenon without drawing the inference — al-Tirmidhi calls such reports "the traditions of N.N." and often classes them as gharib, rare.

Schacht's worked examples land inside the canon.  The wala' hadith forbidding the sale of a freedman's clientage runs Malik – Abdallah b. Dinar – Ibn Umar – the Prophet, and "as Zurqani points out, Abdallah b. Dinar is the common link in the isnads of its several versions, and it can therefore be dated in the generation preceding Malik."¹³  He dates the Barira tradition the same way, to Hisham b. Urwa (d. 146/763).  Both hadith are in both Sahihs.

G.H.A. Juynboll (d. 2010)

Juynboll took the common link and made it a programme.  He is explicit about the debt — "I have in particular benefited from his theories that 'isnads have a tendency to grow backwards' and his 'common link theory'"¹⁰ — and equally explicit that Schacht under-used his own idea:  "it must be conceded first of all that, in my opinion, the common-link theory is a brilliant one."¹⁰

His showcase is a hadith predicting doom for a city "built between Dijla and Dujayl."  Eighteen chains converge on Sufyan al-Thawri (d. 161/776), and Juynboll dates the report to 765–770 CE:  "Ruling out the coincidence that a dozen traditionists invented individually a saying which they then, quite by chance, all attributed to the same master, we have to come to the inescapable conclusion that it is Sufyan himself who has to be held accountable for it."¹⁰  It is a clean demonstration, and it is a safe one — the report is in Ibn al-Jawzi's book of forgeries, so Muslim critics rejected it too.

His hard case is the one that matters.  Man kadhaba — "Whoever lies about me intentionally, let him take his seat in Hell" — is the most massively transmitted hadith in Sunni Islam, in both Sahihs, ascribed to somewhere between 40 and 200 Companions depending on who is counting.  Juynboll's conclusion:

"My own attempt at unraveling the multitude of different isnads of a 'genuine' mutawatir tradition resulted in the unforeseen, but in the final analysis inevitable, conclusion that tawatur as such is no guarantee for the historicity of a hadith's ascription to the prophet."¹⁰

He places its origin in "Iraqi sunnite traditionist circles flourishing in the second half of the second century."¹⁰  If that is right, the tradition's own highest certainty grade does not establish that Muhammad said the thing.

Juynboll's own statement of his method's limits is more modest than his reputation, and honesty requires quoting it:

"apart from the relatively clear-cut case dealt with above… it is mostly impossible to prove with uncontrovertible certainty that isnads are not invented in their entirety.  Thus the common link, if there is one, is often only a useful tool from which to distil an approximate chronology and possible provenance of the hadith.  But that is certainly better than nothing, and may form a suitable starting point for further investigation."¹⁰

And on the same page as his man kadhaba conclusion he concedes the structural weakness:  "the present study's crucial arguments are practically all argumenta e silentio.  This may, at first sight, seem unsatisfactory… To this I can only reply that if there had been other, 'stronger' arguments, these would certainly have been adduced first, but there were none as far as I was able to ascertain."¹⁰

His most aggressive claim was that Nafi', the client of Ibn Umar — the transmitter in the chain Muslim critics rated soundest of all — was not a historical narrator at all, but a device Malik invented.  That claim is what brought the field's most important correction.

Harald Motzki (d. 2019), and the correction that has to be given full weight

Motzki's objection to Juynboll is not that scepticism is impious.  It is that pure chain analysis throws away half the evidence.  His method, isnad-cum-matn analysis, works the chains and the wordings together, and it rests on three premises:  that the variants of a tradition are at least partly the product of a real transmission process;  that the chains at least partly reflect the real path;  and that if the wordings coming out of one common link resemble each other closely, that is a genuine moment of transmission.²

The test is whether variation in wording tracks the shape of the chain tree.  If two common links have versions too different to have been copied from one another, they cannot both descend from either;  they must both descend from something earlier — which pushes the report back behind both of them.  Brown's worked example is the clearest short demonstration:  two versions of the "God descends at night" hadith preserved by Ma'mar b. Rashid are too different for one to be a copy of the other, one running through al-Zuhri in the Hijaz and one through Abu Ishaq in Kufa, both men dead in the 740s.  Their common source therefore lived in the late seventh century.³

Applied to Juynboll's flagship case, the method broke it.  Motzki:

"One of his conclusions claimed that all the Prophetic ahadith with the isnad Nafi' – Ibn Umar found in the 'canonical' collections… do not go back to Nafi' but rather to Malik ibn Anas.  Using the same example as Juynboll, namely the hadith on the alms of the breaking of the fast, we were able to show that his conclusion is wrong.  There is no doubt that this hadith goes back to Nafi' and was not invented by Malik or brought by him into circulation first.  However, the original wording that Nafi' gave to this tradition can be only partially reconstructed."⁷

That last sentence is the honest measure of what the method delivers.  It recovers a chain, not a text.

Motzki's three methodological objections to Juynboll are worth listing because they are the substance of the correction:⁷

1. Treating single-strand chains as presumptively forged "is methodologically unjustifiable… This point of view would condemn the majority of sources for early Islamic history as inauthentic and put them beyond the pale for historians studying the period." He names it in a footnote: it is a conclusion from silence.
2. Juynboll's evidence base was largely confined to the canonical collections, because he worked from al-Mizzi's index of the Six Books. Motzki works from earlier and later material — the Musannaf of Abd al-Razzaq al-San'ani (d. 211/827) at one end, al-Bayhaqi at the other.
3. Pure chain analysis wastes evidence: "only the combination of an analysis of the asanid and the relevant texts is capable of closing the many gaps."

He also ends with a courtesy that says something about how this argument was conducted:  "I cannot stress enough the huge amount of benefit that I have gained from Juynboll's preliminary work."⁷

Motzki's defence of the chain system is the strongest paragraph anyone has written on that side, and it belongs here in full:

"Neither Schacht nor Cook have convincingly shown that 'spread of isnads' was really practised on a significant scale.  They have only shown that there were several possible ways how isnads could be forged and that Muslim scholars could have had different motives to do so.  Apart from possibilities, Schacht and Cook produced only scarce evidence that isnad forgery really happened.

On the basis of mere possibilities and a few instances of the real forgery, it makes no sense to abstain completely from using the isnads for dating purposes.  The historians of the European Middle Ages would not abstain from using diplomas as historical sources because there were cases of forgery which are not easy to detect… If the scholarly isnad system was only or mainly used to feign reliability, then the whole system of validating traditions by isnads would have become absurd."¹⁴

From which he draws his working rule:  "we must, therefore, presume that isnads are, in principle, reliable, except, perhaps, around the time when the system came into being."¹⁴

And his closing candour is the fairest last word available on the whole field:

"Dating traditions is not possible without having recourse to assumptions… these assumptions will always be generalisations based on a limited number of particular facts.  Depending on which facts we generalise, the views on the cultural history of early Islam can be very different.  Therefore, whether the dating of a tradition is considered reliable or not, depends not only on the dating methods applied, but also on our preconceptions of early Islam which we have formed."¹⁴

How far back the best modern work actually reaches

Andreas Görke and Gregor Schoeler took the same tools to the biography of Muhammad rather than to law, reconstructing the corpus of Urwa b. al-Zubayr (d. 94/712) from the independent versions of his students.  Görke concedes the source situation without flinching:  "There are no non-literary sources that pertain directly to the life of Muhammad… Coins and inscriptions with a specific Islamic content or directly referring to Muhammad only appear at the end of the 1st/7th century, about 50 or 60 years after his purported death."⁹

His result is the number this article uses:  "the gap between the oldest traditions or texts that can be reconstructed and the events referred to in these traditions may be reduced to some 40 to 60 years."⁹  What that buys for the Meccan period is about six sentences — Muhammad is frightened by his first revelations, then preaches;  Meccan reactions run from indifference to hostility once he attacks their gods;  Muslims are harassed and emigrate.  "This is approximately the extent of what can be deduced as probable historical facts from the traditions of Urwa for the time before the hijra."⁹  The meagreness is what makes it credible.

Schoeler's summing-up is the most carefully hedged pro-authenticity statement in the field.  Against the revisionists:  "we have tested a method which often allows us to shorten the time span of around two centuries between an event and our available historical sources by half or reduce it even further… It is therefore wrong to claim that we cannot step over the 'magical threshold' of the first century AH."¹⁵  Against the conservatives, in the same breath:  "even in the case of authentic traditions, we should not expect to have objective reports on actual events.  What we have are 'memories' at best, if not actually 'memories of memories'."¹⁵  He also grants that Crone and Cook "rightly stressed the inconsistencies affecting early Islamic historical transmission," and that conservative historians including Watt "often treated the material in an overly credulous manner."¹⁵  His working rule:  "Normally, however, we have no reason to doubt the historicity of the main outlines of reports which emerged one to two generations after the death of the oldest living members of a community at the time an event took place."¹⁵

Does the method check content, or only the chain?

Overwhelmingly the chain — but not exclusively, and the exceptions matter.

The case for "only the chain" is Goldziher's, quoted above, and Muslim reformers agree with him.  Sayyid Ahmad Khan conceded that classical scholars had not performed proper content criticism.  Abu Rayya said outrageous reports were declared sound.

The case for the defence is Brown's, and it comes with receipts.  The early critics did use content, but had a doctrinal reason not to advertise it.  Three documented instances:³

• Al-Bukhari rejects a report that "the signs of the Day of Judgement are after the year 200/815," because "these two hundred years have passed, and none of these signs have appeared."
• Al-Bukhari rejects a hadith about breaking apart Muslim coins, because Muslims did not mint coins until Umayyad times — "they did not exist at the time of the Prophet." That is an anachronism argument, from the man Goldziher says had no feeling for anachronism.
• Muslim rejects a report that five chapters of the Quran each equal a quarter of the book — five quarters — as "reprehensible, and it is not conceivable that its meaning is correct."

Why did they not say so openly?  Brown's explanation is the best available.  The hadith scholars were locked in a fight with the Mu'tazila, for whom reason and the Quran were the criteria and chains were worthless.  "If hadith critics admitted that a hadith could have an authentic isnad but still be a forgery because its meaning was unacceptable, then they would be admitting that their rationalist opponents were correct!"²  So when the content was impossible, they went looking for the flaw in the chain, and filed the report under the transmitter's biography as evidence against him.

What the classical critics did with a sound chain carrying an implausible text is documented by Goldziher, and it is four moves.⁵  Harmonise, through "the work of a subtle harmonistic, which often extends to the smallest details."  Invoke abrogation for legal conflicts.  Apply formal preference rules — where an affirmative and a negative report both have correct chains, the affirmative wins.  And, fourth, appeal to prophetic foreknowledge:  "Muhammad's prophetic gift is used as a factor to smooth over such difficulties."

That fourth move is the strongest single point in the whole Orientalist case, and it should be stated in the plainest possible terms.  If a report describes something that could not have been known in 630, and the answer is that the Prophet foresaw it, then no anachronism can ever count as evidence against any hadith.  A test that cannot fail is not a test.

After about 1100 the picture changes.  As Mu'tazilism faded, open content criticism became publishable.  Al-Jawzaqani (d. 543/1148–9):  "Every hadith that contradicts the Sunna is cast away."  Ibn al-Jawzi:  "any hadith that you see contradicting reason or fundamental principles [of Islam], know that it is forged."  Ibn Qayyim al-Jawziyya devoted a whole section of al-Manar al-munif to the marks of a forged text.  But the mainstream fenced it immediately:  al-Suyuti held that a hadith could be rejected on content only after every attempt at reconciliation had failed, and Mulla Ali al-Qari added that "the door of possible interpretation is definitely wide."²

The verdict.  Chain criticism is a source-reliability test, not a content-plausibility test.  It was built to answer "who told you this, and is he corroborated?", and it answers that question with real rigour.  It was not built to answer "could this have happened?", and for most of the classical period the tradition's own theology forbade asking.  Brown's own summing-up is fair to both sides:  "Muslim critics like al-Bukhari did in fact use the contents of hadiths to prove that they were unreliable, although their degree of skepticism never approached that of the HCM."²

Can a chain be forged? What did the classical scholars do about it?

Yes, chains were forged, and the Muslim critics said so themselves, in detail.  This is simultaneously the best evidence of the tradition's self-awareness and the best evidence against treating a sound chain as proof.

The vocabulary is a confession.  Sariqat al-hadith means "stealing hadith."  Tarkib al-asanid means "rigging chains."  Both are technical terms in the Muslim critical literature.  Three documented mechanisms:²

• To get a report admitted. Amr b. Ubayd (d. 144/761), a Mu'tazilite, had heard the report "He who carries weapons against us is not one of us" but had no chain for it, so "he manufactured one from his teacher al-Hasan [al-Basri] so that he could use it in debates."
• To manufacture corroboration. Al-Daraqutni describes a notebook of hadith praising human reason, forged by Maysara b. Abd Rabbihi, then re-equipped with a fresh set of chains by Dawud al-Muhabbir, then again by Abd al-Aziz b. Abi Raja', then again by Sulayman b. Isa al-Sinjari. "A person who came across the hadiths in this book therefore could find four different sets of isnads leading to four different scholars for hadiths that were in fact total forgeries."
• To manufacture a short, prestigious chain. Al-Tabarani (d. 360/971) narrates through a three-link chain to the Prophet; he is the only scholar ever to narrate from the man in the middle of it.

How they fought it.  Corroboration analysis at every tier of the chain, plus an obsessive tracking of tadlis.  Shu'ba (d. 160/776) said "tadlis is the brother of lying" and audited his own teacher's transmissions.  Yahya b. Sa'id al-Qattan (d. 198/813) identified tadlis even in Sufyan al-Thawri.  Ali b. al-Madini and al-Husayn al-Karabisi "wrote multivolume books identifying the names of those who committed tadlis and the degree of their laxity."  Against that record, Juynboll's flat claim that tadlis "was hardly ever detected" is hard to sustain, and Brown says so.³

But Michael Cook's version of the objection survives that rebuttal, and it is the sharper one.  Cook's point is not that dishonest men went undetected.  It is that a structural effect was never modelled.  In a traditional society "the relevant issue is not originality, but authority:  sharp practice consists in falsely ascribing my view to a greater authority than myself."  If a scholar hears a report from a peer but does not want to look as though he learned religion from an equal, he attributes it to his own teacher instead and extends the chain backwards.  History then preserves two apparently independent chains where there was one.  The critics were hunting liars;  they were not counting how many chains a habit like that inflates.³  Cook's book could not be obtained for this article, so his argument is reported through Brown's summary of it and no words are quoted from Cook's own text.

And the tradition's own limit on detecting forgery is stated by Ibn al-Salah:  a hadith "can be known to be forged only by the acknowledgement of its forger or something equivalent to his acknowledgement."⁴  A confession, or an inference from the man's character.  Not a test of the text.

Two confessions survive from inside, and both belong in any honest account.  Yahya b. Sa'id al-Qattan:  "I have not witnessed lying [about the Prophet] in anyone more than I have seen it in those known for asceticism and piety."  And Ghulam Khalil (d. 275/888–9), a venerated saint of Baghdad whose funeral shut the city's markets, asked about dubious hadith he narrated:  "We forged these so that we could soften and improve the hearts of the populace."³  An early jurist said of legal reports:  "When we arrived at an opinion through reasoning we made it into a hadith."

Goldziher had already described the psychology, and it is not a charge of cynicism:  after Muhammad's death his followers "added many salutary sayings which were thought to be in accord with his sentiments and could therefore, in their view, legitimately be ascribed to him."⁵  Sincere people can create false attributions.

Are there hadith in Bukhari or Muslim that academics date late?

Yes.  Naming them, with the argument and the counter-argument, is more useful than a general claim.

Two qualifications this article will not omit, because leaving them out is the commonest polemical error in this field.

First, the most-quoted "obviously forged" hadith are not in Bukhari or Muslim, and Muslim critics rejected them too.  "If you see Mu'awiya on my pulpit, kill him."  "There will be in my community a man named Muhammad b. Idris, and the strife he brings will be worse than Satan," aimed at al-Shafi'i.  The Abbasid black-banners hadith.  The pigeon-racing hadith.  Ibn Adi, al-Jawzaqani and al-Dhahabi all held the Mu'awiya report unreliable or fabricated, and "the hadith condemning al-Shafi'i was used by Muslim scholars as a textbook example of forgery."²  Using these against the canon is a category error, and a reader who knows the field will notice.

Second, the Qadarite hadith show what drift looks like as distinct from fabrication.  Reports in which the Prophet condemns the Qadarites by that proper name are anachronistic, since the name is later.  But Brown's observation is sharp:  "for every hadith in which the Prophet condemns the Qadarites by this proper name there is a corresponding, non-anachronistic narration in which he refers to them as 'the people of qadar'… In fact, these non-anachronistic narrations are the most reliable ones according to Muslim scholars."²  An anachronism creeping into the later, weaker versions of a report is what transmission drift looks like.  It is not the same thing as inventing the report.

And one structural admission from the tradition's own leading modern defender.  Brown states that Sunni critics "dealt stringently with the isnads of hadiths dealing with law and dogma but were lax with material concerning history (maghazi), the virtues of people or acts (fada'il), pious preaching (wa'z), the end of days (malahim), good manners, and the meaning of Quranic terms (tafsir)."  He measures it in al-Tirmidhi, who graded his own book.  Reports flagged as lacking corroboration run 17% in the chapters on tithing and fasting and 7% on inheritance — but 35% on apocalyptic strife, 52% on the virtues of early Muslims, 50% on pious invocations, and 27% on manners.²

That table is the single most useful measurement in this whole subject.  It says the classical method was applied hardest where the community cared most, and relaxed where it did not — and the places it relaxed are precisely the places a historian most wants to use.  Brown's own reading of it is a defence, and a reasonable one:  "It is possible that it was prioritization of law over other areas that led to the inclusion of large numbers of unreliable hadiths in Sunni collections, not the failings of Sunni hadith-critical methods."

Hadith transmission compared with New Testament transmission

This comparison is made badly by nearly everyone, so start with the structure and only then the numbers.  The difference is not "more copies" against "fewer copies."  These are two different machines with different failure modes.

• The New Testament is a written-copy tradition. Its unit is the manuscript. Its evidence is physical: about 5,300 Greek manuscripts, roughly 54 of them predating AD 300. Its failure mode is copying error and interpolation. Its correction mechanism is collation — because copies branch, an error entering one branch does not enter the others, and a late copy can be checked against an early one that still physically exists. Its residual weakness is the century before the earliest surviving copy, which cannot be checked directly. The Berean Archive's audit of those numbers is at the New Testament manuscripts article, which also cuts several popular apologetic figures down to size.
• Hadith is an oral tradition with claimed chains, written down later. Its unit is the narration — a text plus a claimed route. Its evidence is a claim about the past made by the person making the claim. Its failure mode is not copying error but false attribution: a report may be word-perfect and still never have been said. Its correction mechanism is cross-examination of transmitters, which is genuinely powerful against a careless or lying individual. It is much weaker against a whole generation adopting a convention, and it cannot test a chain at all for the period before chains were demanded.

When were chains first demanded?  On the tradition's own account, about sixty years after Muhammad's death.  Muhammad b. Sirin (d. 110/729):  "In the early period no one would ask about the isnad.  But when the Strife began they would say 'Name for us your sources' so that the People of the Sunna could be looked at and their hadiths accepted, and the People of Heresy could be looked at and their hadiths ignored."³  The Strife is most probably the Second Civil War, 680–692 CE.  That date is not a Western reconstruction.  It is what Muslim sources say.

What each side gets wrong.

• Christians who say "the hadith were written 200 years later, so they are worthless" are describing the wrong failure. The 200-year gap is not a copying gap. The tradition never claimed the texts were written down contemporaneously, and the whole chain apparatus exists to bridge it. And they are usually comparing the New Testament's earliest evidence with the hadith corpus's latest. The honest comparison for the historical-Muhammad question is Görke's: the reconstructible Urwa material sits about 40 to 60 years after the events. That is not two centuries. It is the same order as the gap between Jesus and Mark.
• Muslims who say "we have chains and you only have anonymous manuscripts" are also describing the wrong thing. A chain is a claim about provenance made by an interested party. A manuscript is a physical object whose date can be measured against the claim. The classical critics could not do to a chain what a textual critic does to a papyrus, because there was no chain independent of the man reciting it. The tradition is honest about the consequences: Ibn al-Salah closed independent authentication in his own century precisely because live transmission had degraded, and even before that the licensing system had loosened to the point where scholars obtained transmission rights by letter, for books they had not read, and "for their infant children or even for children not yet born."²
• Both sides ignore that the two traditions have opposite audit trails that point the same way. The New Testament's disputed passages are known because the manuscripts disagree openly and the disagreements survive. The hadith corpus's disputes are known because the critics wrote them down — al-Daraqutni's 217, Ibn al-Jawzi putting canonical narrations in a book of forgeries, al-Tirmidhi grading his own book. Both traditions are self-documenting, and both look worse from outside than a tradition that had suppressed its problems would. That is a point in favour of both.

One asymmetry deserves stating plainly, because it is the most consequential one and it is not in either side's talking points.  For the New Testament, the textual question and the historical question are separable.  Textual criticism establishes what the authors wrote;  whether they told the truth is a different argument run on different evidence.  In hadith they are not separable, because the chain is simultaneously the transmission record and the historical claim.  That is why hadith criticism is harder.  It is a fact about the material, not a failure of the scholars who work on it.

The strongest Muslim responses

This is the defence of hadith reliability at full strength, answering the critique rather than avoiding it.  An argument refuted only in its weak form is not refuted.

1.  The conspiracy required is not credible.  For the corpus to be a fabrication on the scale Goldziher and Schacht require, a scholarly community spread from Spain to Central Asia and split by the Sunni-Shia schism, the Kharijite wars, the mihna inquisition and four rival law schools would have had to co-operate in a fraud, over four centuries, while publicly attacking each other's transmitters, and leave no trace.  Fred Donner's version is that it is inconceivable the decentralised early community could have orchestrated a comprehensive redaction of the tradition into unified form.³  Motzki's version is a question nobody has answered:  "Was the whole system of Muslim Hadith criticism only a manoeuvre of deception?  Who had to be deceived?  Other Muslim scholars?"¹⁴

2.  The burden of proof has been placed in the wrong place.  Motzki's rule — presume chains are reliable in principle, except perhaps around the time the system was forming — is the same rule medieval European historians apply to charters, which were also sometimes forged.¹⁴  David Powers goes further and puts the burden on the sceptic, arguing that it falls on those who would deny the authenticity of reports attributed to the Prophet.³

3.  Convergence is hard to fake, and this is the empirical core.  Powers on the Sa'd b. Abi Waqqas bequest hadith:  it is "either strange or a remarkable coincidence that half a dozen Successors, living in different cities of the Umayyad empire and presumably working independently of one another, adopted the same story to illustrate the origins of the one-third restriction, tracing it back to the Prophet by means of fabricated isnads, all of which converge on one and the same Companion."³  Motzki's study of the killing of Ibn Abi Huqayq finds several independent common links, which pushes that material back to "no later than the last third of the seventh century."³  The "God descends at night" case does the same in miniature.

4.  A forger has no motive to record his own ignorance.  Motzki's study of Abd al-Razzaq and Ibn Jurayj finds both maintaining a perfect formal distinction across a large corpus between "I heard" and the vaguer "from," both repeatedly admitting that they do not know where a report came from or what it means, and Ibn Jurayj recording second- and third-hand transmissions when he could simply have claimed to have heard them directly.  Al-Zuhri transmits a ruling he personally disapproved of, which Motzki reads as evidence of his integrity as a transmitter.³  This is the same kind of argument the Berean Archive makes for the New Testament from useless preserved detail — the cloak left at Troas, the greetings to named individuals — and it should be weighed the same way in both places.

5.  The discipline is measurable and it audited itself.  Lucas's 86–87% agreement between three independent critics;  al-Daraqutni's 217 criticisms inside the canon;  al-Tirmidhi grading his own book and publishing the weak grades.  Brown's summary is not extravagant:  "In theory as well as practice, the Three-Tiered system of demanding a source, investigating its reliability and seeking out corroborating evidence is an effective way of determining the authenticity of a report.  Modern reporters, after all, employ a similar method."²

What the strongest defence does not establish, and this belongs in the same breath.

• It establishes dates, not texts. Motzki's own result on the Nafi' hadith: the original wording "can be only partially reconstructed."
• It establishes a late-first-century floor, not the Prophet. Görke's 40 to 60 years, with a Meccan-period yield of about six sentences.
• It has been demonstrated on a handful of traditions, not on the corpus. Brown concedes this himself: "the hadith tradition is so vast and our attempts to evaluate its authenticity so inevitably limited to small samples, that any attitudes towards its authenticity are necessarily based more on our critical worldview than on empirical fact."²
• It works best exactly where the tradition worked hardest — law and dogma — and worst where the critics admitted they relaxed: history, virtues, preaching, eschatology, exegesis. Which is where most of the reports anyone actually argues about live.
• And Schoeler's warning stands over all of it: what survives is "'memories' at best, if not actually 'memories of memories'."

Common objections and responses

"Bukhari sifted 600,000 hadith down to 7,275, so 98.8% of the corpus was fabricated."

No.  The two numbers count different things and come from different sources, as set out above.  600,000 is a narration count in a praise-report first attested two centuries after al-Bukhari;  7,275 is Ibn al-Salah's count of narrations in the finished book, repeats included.  A "narration" is a text plus a route, so one saying can occupy dozens of slots in the pool.  The claim that al-Bukhari examined and rejected 592,725 distinct reports about Muhammad appears nowhere in the tradition.  Christians should stop making this argument, and Muslims should stop making the mirror-image version of it.

"The hadith were written down 200 years later, so they are worthless."

This describes the wrong failure mode.  Nobody claims the hadith were written contemporaneously;  the chain apparatus exists because they were not.  The right question is whether the chains carry information, and the answer from the best current work is a qualified yes for a subset of material, reaching about 40 to 60 years back from the earliest reconstructible texts.  A 200-year gap in a copying tradition would be devastating.  In an oral tradition with claimed chains it is the ordinary situation, and it is why the chains have to be tested rather than dismissed.

"The isnad system proves the hadith are reliable."

It does not, and the tradition never claimed it did.  Ibn al-Salah says a sound grade does not make a report definitely true, and an unsound grade does not make it false.⁴  A chain establishes a route, and only if the chain itself is genuine.  The critics knew chains could be rigged and coined technical terms for it.  The system raises the cost of forgery;  it does not make forgery impossible, and the strongest available defence of it — Motzki's — is that no one has shown large-scale chain forgery actually happened, not that it could not have.

"This hadith is graded sahih, so Muslims must accept it."

Ask who graded it and when.  If the report is in Abu Dawud, al-Tirmidhi, al-Nasa'i or Ibn Majah and the source is sunnah.com, the grade is very likely al-Albani's or the Darussalam editors', both twentieth-century.  Al-Albani authenticated many reports the classical critics rejected and rejected many al-Bukhari and Muslim accepted.  A modern grading is a real scholarly opinion and worth citing;  it is not a fact about the medieval consensus, and no Muslim is obliged to follow it.

"Here is an absurd hadith — therefore the whole system fails."

Check first whether Muslim scholars also rejected it.  The favourites of internet polemic — "if you see Mu'awiya on my pulpit, kill him," the hadith condemning al-Shafi'i, the pigeon-racing hadith — are in none of the six books and were classed as forgeries by Muslim critics.  Quoting them against the canon shows only that the polemicist did not check.  The reports that actually test the system are the ones inside Bukhari and Muslim, and those are named earlier in this article.

"Aisha rejected a hadith on content, so content criticism was always part of the method."

The Aisha reports are real and they matter — she is shown rejecting the punishment-by-lamentation hadith as contradicting Quran 53:38.  But single episodes do not make a discipline.  The classical manuals define soundness by the chain, and Ibn al-Salah's own test for forgery is a confession or an inference from the transmitter, not an examination of the text.  Content criticism existed;  it was intermittent, and for centuries it was theologically awkward to say out loud.

"Western scholars are hostile outsiders with an agenda."

Some were, and the honest reply is to test the arguments rather than the men.  This article rejects Goldziher's three-mosques case on the evidence, prints Schacht's most famous sentence without the softening word that polemicists add, and gives Motzki's demolition of Juynboll's Nafi' thesis at length.  It is also worth noticing where the criticism came from inside:  Sayyid Ahmad Khan, Rashid Rida, Mahmud Shaltut, Muhammad al-Ghazali and Fazlur Rahman were Muslims, several of them senior religious authorities, and they raised the content-criticism objection independently.

"Christians have no right to make this argument, because the Gospels are worse."

Partly fair, and worth conceding.  A Christian who deploys the argument from silence, the common-link method and the principle of dissimilarity against hadith has to accept the same tools turned on the Gospels, where they have been used for two centuries.  The reply is not that the tools are illegitimate.  It is that the two traditions have different evidence:  the New Testament's transmission can be checked against thousands of physical objects, and the hadith corpus's cannot, because the chains left no artefacts.  That is an argument about evidence, not about who is allowed to speak.

What evidence would change the conclusion?

What would strengthen the Muslim claim.

• Isnad-cum-matn analysis applied at scale rather than to individual traditions. The method's results so far are demonstrations. A systematic study of, say, a thousand reports from the two Sahihs, published with its negative results, would move this article's verdict substantially.
• A statistical test of the corpus that does not depend on the analyst's judgement about each chain. Behnam Sadeghi's "Travelling Tradition Test," which correlates a tradition's content with the cities in its chains, is the only genuinely statistical method in the field. It could not be obtained for this article and nothing is attributed to it here. Its results, checked and replicated, would be the strongest evidence available on either side.
• Documentary evidence from the first Islamic century: a dated papyrus or inscription quoting a hadith that also appears in the canonical collections. One such find would do more than a shelf of arguments.
• Successful reconstruction of the wording of an early tradition, not merely its route. Motzki's honest admission that Nafi's original wording is only partly recoverable is the current ceiling. Raising it would matter.

What would weaken it.

• A demonstration that chain-spreading through tadlis was widespread rather than merely possible. This is precisely what Motzki says Schacht and Cook failed to show; if someone shows it, the common-link inference and the tradition's own corroboration analysis both lose their footing at once.
• Early manuscript evidence of a canonical hadith circulating with a materially different text, or with a chain that is later replaced by a better one.
• More cases like the Qadarite reports, but running the other way: anachronisms concentrated in the most reliable versions rather than the least.
• A systematic study finding that the inter-critic agreement Lucas measured collapses when the sample is widened, or that it reflects shared dependence rather than independent judgement.

What would not change anything.  Producing another absurd report that Muslim scholars already classified as forged.  Recounting the 600,000 story in either direction.  And any argument that requires treating hadith as decisive evidence against Muhammad in one paragraph and as worthless in the next.

Sources

Quotations are reproduced word for word from the sources actually opened.  In the body of the article, long marks and dots on transliterated Arabic have been dropped to match this site's style, inside quotations as well as outside them;  no other change has been made to any quotation.  In the footnotes below, quotations keep the publisher's own spelling, so that they can be matched against the source character for character.  Ibn al-Salah is cited by Category number first, because the only obtainable scan of the Dickinson translation is missing about 35 of its 327 pages and no constant page offset works;  page numbers were read off each page's own printed running head.