In March 1930 a satyagraha — a campaign of disciplined non-violent pressure — began at the Kalaram temple in Nashik for the right of Untouchables to enter it. It was organised by his colleagues, it was conducted with exactly the restraint the method requires, and it ran, on and off, for five years. Volunteers were beaten. The doors did not open. Set beside the Mahad litigation, which took ten years to establish a right to drink water that two resolutions had already granted, it is the second of two long experiments with the same result.

Ramabai died in May 1935, after twenty-nine years of a marriage conducted almost entirely in poverty and mostly in his absence. She had buried four of their five children. He was forty-four, and by every account it broke something in him that did not mend.

On 13 October 1935, at a conference at Yeola in Nashik district, he told an audience of some ten thousand people that he had been born a Hindu because it had not been in his power to prevent it, but that dying one was.

Read carefully, that is not the announcement of a destination but of a departure, and twenty-one years passed before he named where he was going. The distinction matters for the last chapter of this account: at Yeola the decision to leave was taken and made public, and everything afterwards is the question of what to leave for. Representatives of several faiths approached him within weeks, and he let the question stand open for two decades while he studied it.

The next year the Jat-Pat Todak Mandal of Lahore — the Society for the Breaking of Caste, a reformist Hindu body whose entire stated purpose was the abolition of caste — invited him to preside over its annual conference. He wrote the address and sent the text ahead, as they had asked. The committee read it and wrote back asking for passages to be removed: the section arguing that caste rested on the authority of the Vedas and the shastras, and the closing declaration that he would not remain a Hindu. He replied that he would not change a comma. The conference was cancelled.

He printed the undelivered address himself, at his own expense, in May 1936, under the title Annihilation of Caste. The first printing sold out. It is the most widely read thing he wrote, and it is still in print in more languages than he could read.

The argument of the book is not that caste is unkind, and not an appeal for sympathy; it is that caste is a system of graded inequality, in which almost every group has some other group beneath it, so that the people injured by the system are also invested in it and will defend the step they stand on. From there it makes the claim that got the conference cancelled: that reformers who attack caste while treating as sacred the scriptures that sanction it are asking people to disobey what they believe to be divine command, and that this is not a strategy but a wish.

He also disposed of the defence most often offered then and now — that caste is simply a division of labour, an economic arrangement that acquired a religious coating. His answer is that a division of labour would let a man change his work when the work ran out, and caste does not; that it is not a division of labour but a division of labourers, ranked; and that no other system of work in the world assigns a man's occupation before he is born and forbids him to leave it when it is starving him.

The remedies he proposed in the book are the part almost nobody quotes, and they are startling. Intermarriage, on the ground that endogamy is the mechanism and everything else is enforcement. The abolition of the priesthood as a hereditary office, with priests instead qualified by state examination and licence, subject to the same law as any other citizen, and limited in number. One scripture in place of the many, so that there is a single text to argue with. He was proposing to reorganise a religion from the outside, and he said plainly that he did not expect to be thanked for it.

Gandhi replied in his journal Harijan in July 1936, in an article that took the address seriously as the challenge to Hinduism it was. He defended varna — the fourfold ideal order — as a thing distinct from caste as practised, argued that a religion must be judged by its best exponents rather than its worst, and said that Ambedkar's quarrel was with the failures of Hindus rather than with Hinduism. Ambedkar answered in turn, at length, and did not concede the distinction: an ideal that has never once been practised anywhere, he wrote, is not being judged unfairly by its results.

Both replies are printed at the back of later editions, which is the right way to read the book: as one half of an argument that was never resolved, between two men who each had a serious answer to the other and neither of whom moved.