Route 5 rescued 2 of 32, and 21 of the 30 failures returned HTTP 200 with no article - each publisher failing at a constant, recognisable length
also called route 5 sweep · publisher stub fingerprints · unreadable register sweep
We added a new way of reaching papers we could not read - follow the permanent link to the publisher - and tried it on every such paper here, where it worked twice out of thirty-two. More usefully, the failures look identical per publisher, so we can now tell a real failure from a page that only pretends to have worked.
The same thing, for a scientist
Rule 60 added a fifth full-text route because the four existing ones **all require a PMCID**, so `pmcid: null` makes them structurally unavailable rather than failing. This sweeps it across the graph: 58 entries carry an unreadability or abstract-only flag, naming 89 PMIDs; 85 resolved in Europe PMC, **36 have no PMCID**, and **32 of those have a DOI**. **Route 5 rescued 2 of 32** - PMID 36861410 (Haematologica, 27,051 stripped characters) and PMID 30280786 (Eur Rev Med Pharmacol Sci, 12,787), the second of which is new. **The more useful result is the shape of the 30 failures.** Nine returned HTTP 403. The other **21 returned HTTP 200 with no article**, and they did it at constant, publisher-specific lengths: **Elsevier's `linkinghub` redirect returns exactly 11 characters, thirteen times**; **Nature and Springer return exactly 303, seven times**; a Chinese portal returned 7. An implementation that checked HTTP status would have logged **thirteen Elsevier successes**. Measuring the stripped length is not a refinement of route 5, it **is** route 5.
The sweep
| outcome | n | examples |
|---|---|---|
| full text recovered (>=12,000 chars) | 2 | Haematologica 27,051; Eur Rev Med Pharmacol Sci 12,787 |
HTTP 200, Elsevier linkinghub stub, exactly 11 chars | 13 | Lancet, Cell, Mol Cell, Cell Rep x2, Structure x4, JBC, J Mol Biol, Leuk Res |
| HTTP 200, Nature/Springer shell, exactly 303 chars | 7 | Leukemia x3, Nat Genet, Nat Struct Mol Biol, Oncogene, Mol Genet Genomics |
| HTTP 200, other portal, 7 chars | 1 | Zhonghua Yi Xue Yi Chuan Xue Za Zhi |
| HTTP 403 | 9 | Blood x3, JCO, NEJM, Wiley x2, Taylor & Francis x2 |
| no PMCID and no DOI | 4 | not attemptable |
The 403s are honest and the 200s are not. Blood, JCO, NEJM, Wiley and Taylor & Francis refuse in a way no code can mistake for success. Elsevier and Nature return a page.
The constant lengths are fingerprints and should be used as such. Eleven characters thirteen times is not a coincidence about content; it is one redirect shell rendered without JavaScript. A future implementation can log "Elsevier linkinghub stub" rather than "11 characters", which is the difference between a diagnosis and a number.
What the two rescues are worth, which is not the same as what they cost
PMID 36861410 was the case that motivated rule 60 and is worth a great deal: the only head-to-head of two risk groupings on one infant cohort, the only published biological argument for any cut-point in this disease (microarray separation of AFF1 cases at ~90 days), and a third independent statement of the Interfant high-risk age direction that two entries here had backwards.
PMID 30280786 is worth much less, and the entry citing it was already right to be cautious. ana-partner-outcome-enumeration logged it "Not read; the venue is one this project should treat with caution and the abstract was not retrieved." The caution stands. The factual half of that sentence does not: the abstract is in Europe PMC and the full text is one DOI resolution away. It is a 634-patient CCLG-ALL2008 retrospective in which the high-risk group held 28 MLL-rearranged cases, evaluated at four timepoints - a small KMT2A arm inside a large cohort, which is a reason to weight it lightly and not a reason to leave it unread.
Two rescues in thirty-two is a 6% yield, and the rule is still worth having, because the cost is one HTTP request and the distribution of value is not uniform - one of the two changed two entries and a standing risk-stratification claim.
What this says about the register
The unreadable register is mostly real. Thirty of thirty-two could not be reached by any of the five routes, and the reasons are commercial rather than technical: Elsevier, Nature, Blood, JCO, NEJM, Wiley and Taylor & Francis. No further route exists that this project can legitimately use.
So the queue is closed, not deferred. That matters because an open queue invites a future cycle to re-run it; this one should not be re-run without a new access route, and the results file records the per-publisher outcome so the next reader can see that it was attempted rather than skipped.
One methodological residue. Forty-nine of the 85 flagged PMIDs do have a PMCID, so they were reachable by routes 2-4 all along. Most of those are flagged inside enumeration entries that list papers searched and rejected rather than papers that could not be read - the regex used to build this queue matched any PMID in a paragraph mentioning unreadability. That over-collection is why the denominator is 89 and not 32, and it is recorded here so the number is not quoted as 89 unreadable papers.
- Encode the publisher fingerprints - Elsevier linkinghub at 11 characters, Nature/Springer at 303 - so route 5 reports a diagnosis rather than a length, and so a future change in those shells is visible as a changed fingerprint rather than as a silent success.
- Forty-nine of the 89 flagged PMIDs have a PMCID and were reachable by routes 2-4. Separate the entries that flag a paper as UNREADABLE from those that list papers searched and rejected, so the register has a real denominator.
- Decide whether PMID 30280786's 28 MLL-rearranged high-risk patients add anything to the partner-outcome beat, or whether the venue and subgroup size mean it should stay recorded and unused.
- E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.analysis
Across 32 papers that this graph flags as unreadable or abstract-only and that have no PMCID but do have a DOI, resolving the DOI to the publisher recovered full text for 2: PMID 36861410 at 27,051 stripped characters and PMID 30280786 at 12,787. Nine returned HTTP 403 and 21 returned HTTP 200 with no article, at publisher-constant lengths - 11 characters for Elsevier's linkinghub redirect thirteen times, 303 characters for Nature and Springer seven times, and 7 characters for one other portal.
Route-5 sweep: resolving the DOI of every paper this graph flags as unreadable or abstract-only that has no PMCID, and measuring the stripped text length returned · Oncofusion knowledge graph (internal analysis) 2026What this evidence cannot show
Run on 2026-09-20 from one network with one browser User-Agent. **Publisher-side retrievability is not a stable property**: it depends on access policy, on bot detection and on the requesting network, so a 403 here is not proof the article is closed, and a success is not a guarantee it will work again. The 12,000-character threshold for 'full text' is a judgement, not a measurement; the two rescues are far above it and every failure is far below, so no case sat near the line - which is luck rather than calibration.
Evidence that argues against this (1)
- Route 5's yield here is 6%, against a 100% success rate in the single case that motivated the rule. A rule justified by one success and then measured at 6% should be described by the 6%, and the reason to keep it is that the one success corrected a clinical criterion rather than that the route usually works. Haematologica 2023
- E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.journal-article
PMID 30280786 is a retrospective evaluation of the CCLG-ALL2008 protocol in 634 paediatric ALL patients, of whom the high-risk group comprised 217 cases including 28 MLL-rearranged, 22 BCR/ABL-positive and 167 negative for both, with efficacy assessed at day 8, day 15, day 33 and week 12.
The therapeutic efficacy of pediatric ALL patients with MLL gene rearrangement treated with CCLG-ALL2008 protocol · Sun YN, ..., Hu SY · European Review for Medical and Pharmacological Sciences 2018What this evidence cannot show
Read from the structured abstract on 2026-09-20; the full text was retrieved (12,787 characters) but is not quoted here. Established as a description of the reported cohort. **The KMT2A arm is 28 patients inside a 634-patient study** and the subgroup comparisons are between three unequal high-risk strata, so nothing here supports a partner-stratified conclusion. The venue caution recorded by `ana-partner-outcome-enumeration` is retained in full.
Evidence that argues against this (1)
- The entry citing this paper recorded that *the abstract was not retrieved*. The abstract is in Europe PMC and always was, so that sentence was a statement about an attempt rather than about the paper - the same failure mode this register was created to fix, recurring inside the register itself. Oncofusion knowledge graph (internal) 2026
the register swept with route 5: 2 of 32, and the rest are closed commercially
its statement that PMID 30280786's abstract was not retrievable
one of the two rescues supplies its only biological cut-point argument