Probe exact LWBM publication targets from CI

[GerhardBalz-Admin]

Summary

Completes the evidence-recovery work for SKE #17 by probing the exact original Lehigh LWBM publication targets from a clean GitHub Actions runner and recording the recovered payload hashes plus one benchmark-author result oracle suitable for narrowly scoped executable reuse.

Live recovery result

The probe starts from the original HTTP targets linked by the Lehigh LWBM page, follows publication redirects, rewrites nothing, and records status/effective URL/media type/size/SHA-256.

All seven targets returned HTTP 200 and resolved to their corresponding HTTPS paths:

Wine ontology      55,705 bytes  30da3cd5f8c3df59c83cbc309750292ed83e990157028f044be347b5240d1775
4k data           309,178 bytes  df22414d20d97937b84bce63665df791720025276350a64fa97e7b37db723b71
SPARQL workload     2,210 bytes  f981639d0257a7a96c2e475902969322b0943ddf0a4b3d5405aeb6eacb9f1428
RDQL workload       1,808 bytes  e355e3bb67ad022cb7cbd48f8b279dcb39c71448dfb229cecedae54e944becf3
RQL workload        1,390 bytes  3667600c639fed7c9efb087096bf6cc7ac453c2c200588f1a0435a99cb37752e
4k result table    10,612 bytes  7496ef4db76e91228dac73314227817f10345402a5f55305fa38ce303b520c11
10k result table    7,592 bytes  b4d9fe32aaeeba0ddda7791f670a7c67069c8333d8f83db46f9851e03bc4375a

Two independent CI recoveries reproduced every payload digest exactly.

Recovered third-party payloads remain short-lived workflow artifacts and are not committed into SKE.

Exact workload

The recovered Lehigh SPARQL page contains nine historical queries. It uses tuple-style pre-SPARQL-1.0 WHERE syntax; this PR preserves that text as historical evidence rather than silently modernizing it.

For selected Query 6:

PREFIX rdf: <http://www.w3.org/1999/02/22-rdf-syntax-ns#>
PREFIX wine: <http://swat.cse.lehigh.edu/onto/wine.owl#>

SELECT ?X

WHERE (?X, rdf:type, wine:Wine) ,
 (?X, wine:locatedIn, wine:CaliforniaRegion)

Independent oracle

The recovered Lehigh 4k result table reports Query 6 as:

OWLim   result 23   completeness 100
Jena    result 23   completeness 100
Pellet  result 23   completeness 100

The table itself is pinned by SHA-256 7496ef4db76e91228dac73314227817f10345402a5f55305fa38ce303b520c11.

This gives an independent benchmark-author-published target of 23 results, rather than a count derived by SKE/ESKA.

Executable-reuse gate

The selected 4k / Query 6 case now satisfies the three-part gate:

Lehigh 4k data + exact Lehigh Query 6 + independent Lehigh result oracle

Any ESKA implementation must still:

• verify these governed hashes before execution;
• preserve the historical query text separately;
• declare any SPARQL 1.1 form as an explicit syntax transformation;
• require exactly 23 normalized answers;
• keep retrieval/tooling non-authoritative;
• avoid LWBM-specific ESKA vocabulary unless a real gap appears.

Files

• .github/workflows/lwbm-network-recovery.yml
• scripts/recover-lwbm-network.sh
• landscape/lwbm-recovery.md
• landscape/lwbm-artifact-register.json

Boundary

No Lehigh payload bytes are committed. No identifier is rewritten. No mirror is promoted to authority. No new SKE/SMO/ESKA vocabulary and no separate LWBM repository.

Closes #17 after review.