Documented datasets
Two public historical-discussion datasets were described with schemas, provenance pointers, evaluation evidence, and explicit uncertainty.
NostalgAI made a dense dataset
easier to evaluate.
The site already documented two historical-discussion datasets, provenance, held-out evaluation, and uncertainty. The Fix Pack made that substance easier to identify, compare, and act on.
98/ 100 · VERIFIED PRODUCTION
49 → 98 (+49)
The same deterministic v2 method scored the preserved pre-intervention source and the verified public after-state. Every check passed except the deliberately deferred, experimental two-point /llms.txt check.
Technical depth without a concise evaluation path.
Bounded access
The site named a contact route and explained that access, rights, and research fit were reviewed case by case.
Purpose and fit took inference
The homepage did not answer who should use the datasets, when a simpler archive is enough, or who operates the work.
Discovery and policy routes were absent
Canonical/share metadata, entity and dataset markup, privacy, robots, and sitemap routes were missing.
Editorial clarity had to fit the actual runtime.
The page needed a compact research-fit answer path, while the technical recommendations had to be expressed as runtime behavior—not as disconnected files someone might upload mechanically.
M2-R01
Research-fit answer path
State what NostalgAI provides, who needs reconstructed context, when a simpler archive is enough, who operates it, and how case-by-case access works.
M2-R02
Supported metadata and dataset identity
Add canonical/Open Graph metadata and Organization, WebSite, Dataset, and FAQ structured data without inventing licenses, availability, prices, or outcomes.
M2-R03–04
Verified public routes
Implement robots, sitemap, and privacy through the real Sites build, then test status and content type in production.
M2-R05
Keep the weak signal optional
Treat /llms.txt as an experimental convenience, not a required tactic.
The operator adapted the pack instead of forcing it.
- The answer path and metadata/structured-data bundle were implemented as recommended.
- Robots and sitemap were implemented with a documented adaptation to the actual ChatGPT Sites build.
- The privacy route used narrower confirmed language, avoiding invented retention or legal promises.
- The optional
/llms.txtrecommendation was deliberately deferred.
One method, two frozen states.
| Field | Before | Verified production after |
|---|---|---|
| Method | Deterministic v2.0.0 | Deterministic v2.0.0 |
| Score | 49/100 | 98/100 |
| Homepage words | 583 | 717 |
| Robots and sitemap | Absent | HTTP 200; text/plain and application/xml; 2 sitemap URLs |
/llms.txt | Absent | HTTP 404; deliberately deferred |
Baseline provenance: the v2 before record is a frozen scan of the pre-intervention production source, not a raw v2 HTTP capture. A separate exact live v1 capture corroborates the same 583-word state, but its score is not used because the methodology differs. Hosting transforms delivered HTML bytes, so this is a same-methodology, source-lineage comparison—not a byte-identical HTTP comparison.
What the result means.
It establishes that NostalgAI now exposes its identity, evidence, operator, fit, access boundary, privacy information, discovery routes, metadata, and structured dataset description through verified public inputs.
It does not establish improved rankings, AI citations, traffic, leads, conversion, revenue, dataset demand, access eligibility, or model preference. The ModularDot score is a declared deterministic QA metric—not an external industry benchmark.
COMPARE THE OTHER REFERENCE
Open LossyInfo →See the evidence-led service case.
LossyInfo mainly needed existing tools, evidence, and paid creator work connected into one public answer.