From ef457256b576275fa88f67f5ccb1ea0b66b55462 Mon Sep 17 00:00:00 2001 From: timmy Date: Thu, 24 Sep 2026 13:34:47 +0000 Subject: [PATCH] quote fidelity: 269/269 excerpts present in the live articles across 91 entries (264 strict, 5 resolved as markup artifacts); quote fidelity is not summary fidelity --- recoveries/quote-fidelity-live.md | 60 +++++++++++++++++++++++++++++++ 1 file changed, 60 insertions(+) create mode 100644 recoveries/quote-fidelity-live.md diff --git a/recoveries/quote-fidelity-live.md b/recoveries/quote-fidelity-live.md new file mode 100644 index 0000000..3f02cf2 --- /dev/null +++ b/recoveries/quote-fidelity-live.md @@ -0,0 +1,60 @@ +# Quote fidelity against the live articles — 91 entries, 269 excerpts + +**Date:** 2026-09-24 · **By:** Timmy · **Status:** Q3-leftover item closed on the side it could be closed. + +The Q3 audit tested sources resolving and dates agreeing. It did not test whether each magazine entry's +"verbatim excerpts" actually appear in the article they came from. This does that test, from the outside, +against the live articles rather than the archive's own captures. + +## Method + +1. Article URLs: the archive's own site map (`MAGAZINE-SITE-MAP.md`, 91 entries) → `recoveries/magazine-article-map.md`, + every URL fetched, **91/91 return 200**. +2. Excerpts: the `
` elements on each published entry page + (`/var/www/alexander-whitestone/stackchain/magazine-0NN/index.html`), stripped of tags. +3. Test: is the excerpt a substring of the fetched article text — (a) strict after whitespace/entity + normalisation, (b) after also discarding punctuation and inline-markup boundaries + (`[^a-z0-9]` removed from both sides), which is the honest form when a word can be wrapped in a link or a + ``. +4. Every residual was then re-fetched **sequentially** and inspected by hand, because parallel fetches under load + returned truncated bodies and produced false misses. + +Script: `quotefidelity.py` / `quotefidelity2.py` (run from a shell on the archive host, output kept at +`/tmp/quotefidelity2.json` for the run). + +## Result + +| test | excerpts found | +|---|---| +| strict substring | 264 / 269 | +| punctuation- and markup-insensitive | 265 / 269 | +| after sequential re-fetch of the residuals | **269 / 269** | + +The five strict misses, each resolved by inspection: + +| entry | excerpt | why it did not match strictly | +|---|---|---| +| `magazine-033` | "June 4th 2021 – Ron Paul kicks off the bitcoin conference in Miami." | the live page renders the ordinal as `4 th` (``) → tag boundary inside a token | +| `magazine-083` | "This article was written by Grepples, an internet frog selling seeds and member of Stackchain crew." | the byline sits inside an ``; stripping leaves "by Grepples , an" | +| `magazine-083` | "You can find Grepples on X @15Grepples and on Nostr Grepples@nostr.land." | same byline block, link boundaries | +| `magazine-084` | "This tutorial was written by HiQ, a memer, DSB bookie and Stackchain Magazine core member." | same byline block, link boundaries | +| `magazine-089` | "Bitcoin, yes, has won. But Bitcoin culture?" | the live page has "Bitcoin culture ?" — question mark inside a link | + +Four further entries (`003`, `049`, `081`, `089`) were reported missing by the punctuation-insensitive pass and +were present on sequential re-fetch; the misses were truncated responses under parallel load, not content +differences. Both passes and the re-fetch are recorded because the false misses are the kind of result that gets +quietly deleted otherwise. + +## What this settles, and what it does not + +- **Settles:** for all 91 published magazine entries, the quoted excerpts are *in* the live articles. No + fabricated quotation was found, and no excerpt was found to belong to a different piece. +- **Does not settle:** whether each entry's **machine-written summary** (the "What happened" / "In plain English" + prose) states only what the article supports. Quote fidelity is not summary fidelity. The worked counter-example + is `magazine-040`, where the summary says the piece closes "with Satoshi resting on the 97,250th block, a nod to + the last block he mined" — the article's own closing lines are a Genesis 2:2 rewrite ("By Block 97250, Satoshi had + finished the work over which he labored"), and no public analysis ties 97,250 to Satoshi's last mined block + (Patoshi analyses put that near block 54,316, May 2010). The quote is faithful; the interpretive clause is not + supported. Found by zephyrus-sparkborn in Q4 (`f50f052`), reproduced and extended here. +- Point-in-time 2026-09-24. A live article can be edited after the archive captured it; this test says the excerpts + match the articles **as they are today**.