Make field matching meaningful and add an eligibility evaluation harness

[ma7moudalysalem]

What this changes

The recommender and the eligibility checker both compared a student's field of study against a listing by exact string equality, across two vocabularies that never met. Listings are written from a fifteen-item canonical list; the seeder wrote fine-grained specialisms and the student profile takes free text. A student interested in Software Engineering could never match a listing open to Engineering.

Scoring

The match score is now the share of the attainable weight a listing earns across the criteria it actually constrains, instead of a fixed point sum. A criterion the listing leaves open drops out of both the numerator and the denominator, so an award open to every discipline is no longer punished for that openness — previously Gates Cambridge and Fulbright scored zero on the 40-point field term for being open to everyone.

Field comparison is graded: an exact match counts in full, a shared significant word counts as a half. The explanation now names the criteria that matched rather than restating the score band, and says plainly when nothing matched.

Eligibility

The field criterion could never report "not met" — its final branch returned partial, so a law student was told a computer science award was partially met. Unrelated fields now report not met, and "partially met" is reserved for a genuinely adjacent field.

Data

The seeder writes FieldsOfStudyJson on generated listings and states student preferences in the same canonical vocabulary.

	before	after
open listings carrying a field	0 of 336	733 of 763
(student, listing) pairs sharing a field	0.0%	12.2%

The curated external dataset carries fieldsOfStudy: five entries restrict discipline, nineteen are explicitly open to any field.

Evaluation harness

server/tools/ScholarPath.Eval runs the delivered checker over a seeded database with a fixed seed and reports per-criterion outcomes, so a figure can be reproduced:

dotnet run --project server/tools/ScholarPath.Eval -- --students 60 --listings 60 --seed 20260918

On 3,600 pairs:

field criterion	before	after
not met	0%	90.4%
partially met	91.9%	5.2%
met	8.1%	4.4%
unrelated field reported as partially met	93.5%	0%

Two claims made true in code

The SignalR Redis backplane is now wired when Redis is configured (AddStackExchangeRedis was never called), and WebKit and Edge are added to the Playwright projects.

Verification

• dotnet build — 0 warnings, 0 errors
• dotnet test — 951 passing, 18 new (10 scoring, 8 field matching). The scoring path had no test coverage at all before this.

Note for the reviewer

Program.cs in the working tree also carries an unrelated, unfinished "Live System Activity" feature whose files are untracked. That work is deliberately not included here — only the SignalR hunk is.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.