Files
troupe-standup/wizard-work/sphinx/longctx-archive-2026-10-26/EVIDENCE.md
T

5.6 KiB

Evidence log

Generated by build_evidence.py on 2026-10-26. Everything below is captured tool output, not prose.

1. Fetch (live arXiv API)

query search_query entries bytes sha256
q1_longctx_multimodal.xml all:"long context" AND all:multimodal 30 129936 931310036cc108bec3d6bcc073c3f890518d0829fc8b7416876cd09c120462cc
q2_long_video.xml all:"long video" AND all:understanding 30 82212 494e1d801c62f333b13ccdbb4d323c3fabd8b0284f5c4f3920bd5031732b4acd
q3_longctx_cv.xml all:"long-context" AND cat:cs.CV 30 79247 68e2322eea52398d101a489583cb98a7c7dfb06a417a274598131335c040a6b3
q4_long_document.xml all:"long document" AND all:multimodal 30 78052 835383b5ed3865888777873d7e79941343ce6e190bb90d905170b4ddd01e7384
q5_videolang_long.xml all:"video-language" AND all:long 30 79242 86214b1fc776c9761f011c55f00e9d9f2d8763f1598311757b1d3b75d25ab148

De-duplicated corpus: 134 papers, 228277 chars of abstracts, dates 2025-11-28 .. 2026-10-23.

2. python3 verify.py — exit code 0

PASS 1a unique base_id across 134 records
PASS 1b all 134 records have title, abstract, published
PASS 1c all published timestamps ISO-8601
PASS 1d papers.jsonl count 134 == build_stats.papers
PASS 1e every abstract >= 200 chars (min 914)
PASS 2 all 5 raw Atom responses match manifest sha256 + byte length
PASS 3 all 150 raw entries re-parse to identical title/abstract/date in papers.jsonl
PASS 3b raw base-id set == archive base-id set (134 ids)
PASS 4 CORPUS.txt 228277 chars == build_stats.corpus_chars
PASS 4b all 134 papers appear as headers in CORPUS.txt
PASS 5 C01 stat papers = 134 (expected 134)
PASS 5 C02 stat corpus_chars = 228277 (expected 228277)
PASS 5 C03 stat date_min = '2025-11-28' (expected '2025-11-28')
PASS 5 C04 stat date_max = '2026-10-23' (expected '2026-10-23')
PASS 5 C05 category_count cs.CV = 96 (expected 96)
PASS 5 C06 term_count 'video' = 71 (expected 71)
PASS 5 C07 term_count 'memory' = 44 (expected 44)
PASS 5 C08 term_count 'retrieval' = 45 (expected 45)
PASS 5 C09 term_count 'benchmark' = 32 (expected 32)
PASS 5 C10 term_count 'long-context' = 42 (expected 42)
PASS 5 C11 term_count 'document' = 30 (expected 30)
PASS 5 C12 term_count 'agent' = 21 (expected 21)
PASS 5 C13 term_count 'KV cache' = 7 (expected 7)
PASS 5 C14 term_count 'RAG' = 14 (expected 14)
PASS 5 C15 term_count 'frame selection' = 4 (expected 4)
PASS 5 C16 cooccur memory+retrieval = 15 (expected 15)
PASS 5 C17 cooccur video+memory = 28 (expected 28)
PASS 5 C18 cooccur token+compression = 11 (expected 11)
PASS 5 C19 month_count 2026-09 = 33 (expected 33)
PASS 5 C20 month_count 2026-10 = 38 (expected 38)
PASS 5 C21 stat multi_query_papers = 15 (expected 15)
PASS 5 C22 verbatim quote present in 2610.16118 abstract
PASS 5 C23 verbatim quote present in 2610.16118 abstract
PASS 5 C24 verbatim quote present in 2610.05940 abstract
PASS 5 C25 verbatim quote present in 2609.32182 abstract
PASS 5 C26 verbatim quote present in 2610.30089 abstract
PASS 5 C27 verbatim quote present in 2610.26910 abstract
PASS 5 C28 verbatim quote present in 2610.20964 abstract
PASS 5 C29 verbatim quote present in 2610.20964 abstract
PASS 5 C30 verbatim quote present in 2605.08888 abstract
PASS 5 C31 verbatim quote present in 2605.08888 abstract
PASS 5 C32 verbatim quote present in 2609.34314 abstract
PASS 5 C33 verbatim quote present in 2609.34314 abstract
PASS 5 C34 verbatim quote present in 2609.19969 abstract
PASS 5 C35 verbatim quote present in 2609.17248 abstract
PASS 5 C36 verbatim quote present in 2610.31385 abstract
PASS 5 C37 verbatim quote present in 2610.23975 abstract
PASS 5 all 37 claims recomputed to their stated values
PASS 6a ANALYSIS.md is 10877 chars
PASS 6b ANALYSIS.md cites all 37 claim ids
PASS 6c ANALYSIS.md cites no claim id absent from the ledger
PASS 6d every arXiv id mentioned in ANALYSIS.md (12) exists in the archive
PASS 7a clean copy of the archive verifies (control)
PASS 7b verifier rejects a falsified claim value (C01 expected+1)
PASS 7c verifier rejects a tampered abstract (raw Atom mismatch)
PASS 7d verifier rejects a corrupted raw response (sha256)

56 checks, 0 failures
ARCHIVE VERIFIED

3. Artifact hashes

file bytes sha256
ANALYSIS.md 10941 dc7a3a006009baa45f56a69bb1573ac26282247feac19dfd0c1f4b6eaa2cebcb
CORPUS.txt 228277 94590a67d5fd5a3368e174e0693e9a02fbbb0226668b0189944e2dfd4cd3c767
README.md 2125 2e8ab9e230ad3fd9c8951a83b42d7ee5a68a17863532e8d17d7c0f6dafcb8d66
build_stats.json 150 66e01a6b4543791a125f89ec27adf42a319d78bbe8a6437208a369243abd0273
claims.jsonl 8021 27f9e5dcaa19cf8a34b940c1e4e5a9f8fc70759ec23ebc9f0259780990885cf8
papers.jsonl 284018 e4934840c2a720a9af9547c882e37174b2e5359c3b2a5139b1a1af78c26bfc82
raw/q1_longctx_multimodal.xml 129936 931310036cc108bec3d6bcc073c3f890518d0829fc8b7416876cd09c120462cc
raw/q2_long_video.xml 82212 494e1d801c62f333b13ccdbb4d323c3fabd8b0284f5c4f3920bd5031732b4acd
raw/q3_longctx_cv.xml 79247 68e2322eea52398d101a489583cb98a7c7dfb06a417a274598131335c040a6b3
raw/q4_long_document.xml 78052 835383b5ed3865888777873d7e79941343ce6e190bb90d905170b4ddd01e7384
raw/q5_videolang_long.xml 79242 86214b1fc776c9761f011c55f00e9d9f2d8763f1598311757b1d3b75d25ab148
raw_manifest.json 1979 a9b7218fe2b13de5a8160594145ad0917b0a98937bcd441c280cfc55a657b977
verify.py 12915 89ddedf5f8a41456f83ec8ece441d6f3c8f7a963178a3a1b014d6b82843c7716

4. Claim ledger coverage

37 claims: category_count=1, cooccur=3, month_count=2, quote=16, stat=5, term_count=10