On 17 July 2026, Dominik Wujastyk posted Experiments with Claude on the Project’s blog, reporting the preliminary results of an experiment in using Claude to edit Suśrutasaṃhitā, Sūtrasthāna 6 (Ṛtucaryā). He asked Claude to study 23 chapters critically edited by the Suśruta team, alongside the witnesses on which they are based, to derive a set of rules or patterns capturing their editorial practices, and then to apply those rules to Saktumiva’s mechanical collation of chapter 1.6. The results were posted here.

Since then, Jason Birch has critically edited chapter 1.6 on ṛtucaryā without consulting Claude’s edition. He has recently compared his draft edition of the first 20 passages with Claude’s and offers the following general observations.

1. Claude improved on Saktumiva’s mechanical collation of the witnesses by adopting K in more instances where the witnesses differ. In at least one instance, it correctly chose a reading of K that was the lectio difficilior and wrote a footnote explaining why the reading is plausible.

2. Claude’s edition was limited by its inability to check the draft transcripts against the manuscripts. This might be possible in the future if Claude is given access to the manuscript images. Many of the defects in Claude’s edition resulted from mistakes in the draft transcripts; examples are given in the detailed comments below.

3. Claude often uncritically favoured K (perhaps because it was told that K is the most important witness). As a result, it adopted incorrect readings of K. In quite a few instances, these incorrect readings were errors in the draft transcript, which Claude failed to identify. For example, it adopted triṃśatmuhūrttam (instead of triṃśanmuhūrttam), evauṣadhyau grīṣmaniḥsārā … bhavaty (instead of evauṣadhyo grīṣme … bhavanty) and śleṣmikāṇāṃ (instead of ślaiṣmikāṇān). Further instances are noted below.

4. Claude included an additional passage in K that is not part of the original work and is marked as an insertion in the TEI code of the draft transcript). This passage does not fit the syntax of the sentence into which it was inserted. It is a clear example of Claude uncritically accepting the reading of K.

5. Other problems include incorrect word division (e.g., dvimāsike naivaṃ instead of dvimāsikenaivaṃ), unnecessary footnotes, and incorrect punctuation of the edited text (although punctuation may not have been explicitly mentioned in the prompt).

Significant differences in 1.6.1–20

Significant differences between Claude’s and Birch’s editions of 1.6.1–20 are listed below.

1.6.3

Claude and Birch both adopt saṅkalayati and both write footnotes on this sentence, which is a nirukti of kāla. Claude’s footnote simply points out that K preserves the correct form saṅkalayati (as opposed to saṅkālayati in N and H). Birch considered this obvious and instead commented in his footnote on Ḍalhaṇa’s understanding of kāla.

1.6.5

Claude adopts triṃśatmuhūrttam, which is incorrect sandhi. K and N actually read triṃśanmuhūrttam, but the draft transcripts used by Claude had triṃśatmuhūrttam for these witnesses.

1.6.7

Claude did not insert an avagraha in āpyāyate ’rkaḥ. Claude and Birch both accept balavattarā, the reading of K^pc and H, which is the lectio difficilior, and Claude adds a footnote explaining why this reading is plausible. Further on, however, Claude adopts uttarottaraś ca, K’s reading in the draft transcript, which does not make sense. In fact, K, H and the vulgate read uttarottarañ ca.

1.6.9

Claude adopts varṣasaṃvatsaraḥ, the reading of K^pc in the draft transcript, which is unsatisfactory. K^pc actually reads varṣaḥ saṃvatsaraḥ, which is better.

1.6.10

Claude reads dvimāsike naivaṃ instead of dvimāsikenaivaṃ, an incorrect word division. Further on, Claude adopts sa vidāhāt, the reading of K, H and N, even though these witnesses have a plural verb. Birch adopts the vulgate reading vidāhāt, which is better.

1.6.11

Claude incorrectly includes the passage tatra … uttarāyaṇe, which was added to K and breaks the syntax of the original sentence. Claude states in a footnote that the passage was added by a later hand, an observation probably based on the transcriber’s TEI markup rather than on direct examination of the manuscript. Further on, Claude adopts evauṣadhyau grīṣmaniḥsārā … bhavaty, K’s reading in the draft transcript, but K and H in fact read evauṣadhyo grīṣme … bhavanty, the correct reading, which Birch adopts. Claude’s footnote points out the problem with the singular verb, but explains that it understands evauṣadhyau as a dual subject.

1.6.13

Claude adopts śleṣmikāṇāṃ, an incorrect transcription of K, which (like N and the vulgate) reads ślaiṣmikāṇān. Claude also adopts sañcayaprakopaśamā, a reading wrongly attributed to K in the draft transcript, and then writes a footnote stating that the correct reading, sañcayaprakopopaśamā, is found in H and N. In fact, K has the same reading as H and N.

1.6.14

Claude adopts haimanam, the reading of K, and then writes an unnecessary footnote noting that H and N read hemantam and hemanam respectively.

1.6.15

Claude adopts tūpayayujyamānāḥ, an incorrect reading of K, but then states in a footnote that this reading is incorrect. The error is in the draft transcript of K. Birch corrected the transcript and adopted tūpayujyamānāḥ, the reading of K, N and H.

1.6.16

Claude adopts tāsām, the correct reading of K, and writes an unnecessary footnote explaining why.

1.6.19

Claude adopts K’s uṣasyante (a transcription error for udhvaṃsyante). The better reading is H’s udhvasyante.