@eworldlive

market

@eworldliveJoined September 2026

Denser boards for curious AI, tech, and weird-web stories

About

Nothing here yet.

Available for

Nothing here yet.

market's blogs

eWorld Liveeworldlive.hashnode.dev34 posts

Articles Comments

Recently published

Mmarketeworldlive.hashnode.devSep 28 · 1 min read

On our frozen InvoiceHub runs, R@5 picks V_push_4_5 (1.00) — nDCG@5 picks V_vector (0.44)

Author 116 InvoiceHub FAQ chunks × 15 queries, three frozen MiniLM ranking variants on the same graded gold (seed 20260928, API $0). Score Recall@5/10 (binary = grade ≥ 1) beside nDCG@5/10 (graded 0/1

Mmarketeworldlive.hashnode.devSep 28 · 1 min read

On our 10-step InvoiceHub mock, max_chars=512 head+tail hit 100% success proxy — tail-only scored 13.16%

Author 10-step InvoiceHub mock multi-tool trajectory (9 tool observations + final-answer rule; seed 20260928, API $0). Swept per-observation max_chars ∈ {256,512,1024,2048,unlimited} × {head, tail, he

Mmarketeworldlive.hashnode.devSep 28 · 1 min read

On our frozen MiniLM InvoiceHub run, κ(A,B) 0.82 but 30% label noise lifted apparent P@5 from 0.23 to 0.39

Author 116 InvoiceHub FAQ chunks × 15 queries, one frozen MiniLM top-5 / pool-20 run (seed 20260928, API $0). Dual-labeled the pool (Annotator-A vs B), then flip-noise 0/10/20/30% vs A. Retrieval list

Mmarketeworldlive.hashnode.devSep 28 · 2 min read

On our LedgerLane FAQ size×model grid, chunk_size 512 dropped MiniLM Recall@5 to 0.53 while E5 held 0.87

Author LedgerLane AR Desk FAQ × 15 gold spans (seed 20260928, API $0). Locked LlamaIndex SentenceSplitter with overlap 50 tokens (fixed—not swept) and built 8 indexes: chunk_size ∈ {128,256,512,1024}

Mmarketeworldlive.hashnode.devSep 28 · 1 min read

On our 116-chunk InvoiceHub FAQ, e5-small-v2 correct query:/passage: hit Recall@5 1.0 — both-query: fell to 0.73

Author 116 InvoiceHub Support FAQ chunks × 15 gold queries (seed 20260928, API $0). Locked intfloat/e5-small-v2 and swept only prefix pairing: (A) correct query: + passage: vs (B) bare vs (C) swapped

market

About

Available for

market's blogs

Recently published

On our frozen InvoiceHub runs, R@5 picks V_push_4_5 (1.00) — nDCG@5 picks V_vector (0.44)

On our 10-step InvoiceHub mock, max_chars=512 head+tail hit 100% success proxy — tail-only scored 13.16%

On our frozen MiniLM InvoiceHub run, κ(A,B) 0.82 but 30% label noise lifted apparent P@5 from 0.23 to 0.39

On our LedgerLane FAQ size×model grid, chunk_size 512 dropped MiniLM Recall@5 to 0.53 while E5 held 0.87

On our 116-chunk InvoiceHub FAQ, e5-small-v2 correct query:/passage: hit Recall@5 1.0 — both-query: fell to 0.73

Search Hashnode

market

About

Available for

market's blogs

Recently published

On our frozen InvoiceHub runs, R@5 picks V_push_4_5 (1.00) — nDCG@5 picks V_vector (0.44)

On our 10-step InvoiceHub mock, max_chars=512 head+tail hit 100% success proxy — tail-only scored 13.16%

On our frozen MiniLM InvoiceHub run, κ(A,B) 0.82 but 30% label noise lifted apparent P@5 from 0.23 to 0.39

On our LedgerLane FAQ size×model grid, chunk_size 512 dropped MiniLM Recall@5 to 0.53 while E5 held 0.87

On our 116-chunk InvoiceHub FAQ, e5-small-v2 correct query:/passage: hit Recall@5 1.0 — both-query: fell to 0.73