nx_dr_corpus.nx
buildroot/runtime/nx_dr_corpus.nx
about
nx_dr_corpus.nx -- FILE-BACKED corpus mode for the sovereign deep-research engine (DR-9).
The chain judged documents passed as argv strings, which caps a source at the argv limit --
a real fetched web page (tens of KB) cannot ride there. This organ reads a source document
from DISK, so the engine runs over the corpus the live nx_fetch spine already writes
(knowledge/library/*.txt). That is the concrete unlock for a full-corpus benchmark run.
FAIL-LOUD BY CONSTRUCTION: an unreadable or empty source returns -1 so the caller REPORTS it
rather than silently scoring a missing file as zero coverage (a silent-truncation trap).
Composes nx_dr_chain (judge -> conduct -> score) and nx_dr_run (tokenizer).
Every func <=6 params (NAS nx_cc >6-arg skew, debt seq239). No hardware writes (Rule 26).
module: nishi-core.research.dr_corpus
depends: nx_dr_chain.nx, nx_dr_run.nx, nx_syscalls.nx
genealogy_id: drbench_2026_corpus_scale
dependencies 3 imports · 1 importers
imports: nx_dr_chain.nxnx_dr_run.nxnx_syscalls.nx
imported by: nx_dr_corpus_cli.nx
structs
| none |
consts
| none |
functions
| 20 | func corp_tokenize_file(path: *u8, out_ids: *i64, maxn: i64) -> i64 |