Add Entity Resolution case study; refresh EV Charging metadata and re-tier projects

[cbratkovics]

Motivation

• Add the Entity Resolution independent project as a featured card and a case-study route, with every published number traced to a committed v0.1.0 artifact key, and coverage reported separately from accuracy.
• Close the Codex finding on PR Feature EV charging data case study and surface projects earlier #13: page-level openGraph and twitter objects replaced the root ones, so case-study routes rendered without og:image / twitter:image.
• Re-tier the project grid to match the resume's Selected Independent Projects lineup, and align three EV Charging statements with the EV repository's docs on main.

Description

• portfolio/components/case-study/: EvidenceLinks, Section, and FindingBlock extracted from the EV page (identical markup) plus a caseStudyMetadata() helper that builds route metadata with the site's social image in both objects. Both case-study routes use them.
• portfolio/data/projects.ts: new optional, data-only citations: ArtifactCitation[] on Project; new entity-resolution entry (featured: true) with 21 citations pinned to entity-resolution commit f50d764 (tag v0.1.0); grid re-tiered by data fields and order only: EV Charging (flagship) → Entity Resolution → Fantasy Football → NBA (featured) → SQL Genius → AI Chat → Document Intelligence (additional).
• portfolio/app/projects/entity-resolution/page.tsx: new route mirroring the EV page (canonical URL, Open Graph and Twitter metadata with image, skip link, "Back to projects", eyebrow, "As of v0.1.0" scope line, evidence links top and bottom, one /#work cross-link). Registered in app/sitemap.ts.
• EV card and page: three meaning-level alignments listed under Claims narrowed. No EV numbers changed. No v0.1.0 GitHub Release exists for either repository, so no Release links were added. No screenshots were added: the two candidate images in demo_images are generated infographics rather than product captures, and each carries a figure the text does not.
• Tests: content-contract assertions now encode the new tier contract (4 featured / 3 additional, new order, 13 narratives), check the Entity Resolution hedges, and require every cited figure to appear in the card or page copy. E2E covers the new route, the homepage tier order, the sitemap entry, and og:image / twitter:image on both routes.

Deliberate assertion changes: featured: false count 2 → 3, ordered title list and repository list extended, narrative counts 12 → 13, plus the new entity-resolution content-contract and e2e tests. No assertion was loosened.

Artifact citations (Entity Resolution, v0.1.0, commit f50d764)

Figure on site	Artifact and key	Value
482,514 / 482K sampled release groups	artifacts/manifest.json#counts.musicbrainz.sampled	482514
241,752 in-scope truth pairs	artifacts/blocking_report.json#pair_completeness.truth_pairs	241752
96.3% before cap	artifacts/blocking_report.json#pair_completeness.union	0.963355
96.2% after cap	artifacts/blocking_report.json#pair_completeness.after_cap	0.962362
cap of 200	artifacts/blocking_report.json#candidate_cap_per_a	200
240 truth pairs lost	artifacts/blocking_report.json#pair_completeness.truth_pairs_lost_to_cap	240
7.6M candidate pairs	artifacts/blocking_report.json#candidate_pairs_after_cap	7595477
rules precision 0.9946	artifacts/eval_rules_v1.json#metrics.at_auto_accept.precision	0.994615
rules recall 0.931	artifacts/eval_rules_v1.json#metrics.at_auto_accept.recall_labelled	0.930917
rules F1 0.962	artifacts/eval_rules_v1.json#metrics.at_auto_accept.f1	0.961712
classifier precision 0.9993	artifacts/eval_learned_v1.json#metrics.at_auto_accept.precision	0.999344
classifier recall 0.850	artifacts/eval_learned_v1.json#metrics.at_auto_accept.recall_labelled	0.849952
classifier F1 0.919	artifacts/eval_learned_v1.json#metrics.at_auto_accept.f1	0.918614
review queue 25,076	artifacts/eval_rules_v1.json#metrics.ambiguity_rule.review_queue	25076
review queue 6,016	artifacts/eval_learned_v1.json#metrics.ambiguity_rule.review_queue	6016
4,501 unverified accepts	artifacts/eval_rules_v1.json#metrics.unverified_accepts.count	4501
0.50 unlinked share per side	artifacts/manifest.json#sample.thresholds.a.unlinked_share_actual	0.5
~13.8 GiB data peak	artifacts/manifest.json#runtime.data_dir_bytes_peak	14784312081 bytes
242,542 linked album release groups	artifacts/profile/musicbrainz.json#truth.a_records_with_link_by_primary_type.Album	242542
2,324,821 album release groups	artifacts/profile/musicbrainz.json#records.albums	2324821
decision-level calibration overconfident	artifacts/eval_learned_v1.json#metrics.calibration.decision_level.ece vs pair_level.ece	0.047858 vs 0.000147

EV Charging figures were re-verified against the EV v0.1.0 artifacts and are unchanged: row counts are sums over by_source.*.{bronze,accepted,quarantined} in artifacts/silver/silver-20260920T194003Z.json; 45.1%, 12.4%, 3.6×, 28% (0.276), 11.4%, 10.5–12.4%, 8.7–11.4%, and 48,619 / 20,699 / 17,677 / 9,258 all resolve to keys in artifacts/findings/findings-20260920T194023Z.json.

Testing

Run in portfolio/ in this order (Node 23.9 locally; engines says 22.x, CI uses .nvmrc):

npm run lint            → exit 0 (eslint .)
npm run type-check      → exit 0 (tsc --noEmit)
npm test                → 13 tests, pass 13, fail 0
npm run build           → ✓ Generating static pages (10/10); routes: /, /projects/ev-charging-data-unified-schema, /projects/entity-resolution, /sitemap.xml, /robots.txt, /opengraph-image
npm run test:e2e        → 4 tests, pass 4, fail 0
npm run check:publication        → Publication content check passed.
npm run check:publication:build  → Publication content check passed.

Built HTML (via next start):

/projects/ev-charging-data-unified-schema
  <meta property="og:image" content="https://cbratkovics.dev/opengraph-image"/>
  <meta name="twitter:image" content="https://cbratkovics.dev/opengraph-image"/>
/projects/entity-resolution
  <meta property="og:image" content="https://cbratkovics.dev/opengraph-image"/>
  <meta name="twitter:image" content="https://cbratkovics.dev/opengraph-image"/>
/sitemap.xml
  <loc>https://cbratkovics.dev</loc>
  <loc>https://cbratkovics.dev/projects/ev-charging-data-unified-schema</loc>
  <loc>https://cbratkovics.dev/projects/entity-resolution</loc>
homepage first-occurrence order: Flagship evidence < EV Charging < Entity Resolution < Fantasy Football < NBA Stat Predictor < Additional work < SQL Genius < AI Chat < Document Intelligence

EV case-study HTML diffed against main (scripts and asset hashes excluded): the only changes are the build-id comment, the five added og:image* / twitter:image meta tags, and the three approved copy changes below. 18 changed lines.

External URLs on the new card and page (repository, results site, Findings, Methods card, a pinned artifact blob, EV Findings) all returned HTTP 200 from this environment.

Claims narrowed

• Entity Resolution card and page describe three methods (exact-match rule, weighted rules baseline, calibrated classifier), not "two matchers"; the headline comparison remains rules vs. classifier.
• The draft's "135 cited figures across 12 files" was dropped: the checker reports 124 at v0.1.0, and the count is checker output, not an artifact. The page now says the checker fails the build when a cited figure lacks a matching key.
• "MusicBrainz editors have linked a large share of release groups" became "242,542 of its 2,324,821 album-type release groups", both cited.
• The card summary says the sampled release groups were "matched against" Discogs masters rather than "linked to", since half the sample is unlinked by design.
• "A 'How this was built' section in the README says so" became "the brief says so": the entity-resolution README on main no longer carries that sentence; docs/BRIEF.md does.
• The pipeline line places dbt bronze/silver/gold after the artifacts, matching the repository's architecture.
• EV page: "three public operators' charging-session logs" → "three public sources' charging-session logs" (the UK DfT publication is not an operator).
• EV card and page: the "pilot at multi-port stations around midday" recommendation is replaced by prioritizing late-morning-to-mid-afternoon investigation at multi-port stations, with port-inventory validation and queue or turned-away-driver evidence required before claiming constrained demand or choosing an intervention.
• EV card and page: added "Utilization figures are within-source only; the sources differ in operator, place, and period and are not compared."

🤖 Generated with Claude Code

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
cbratkovics-dev	Ready	Preview	Sep 25, 2026 12:28am UTC

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-09-24T23:58:56.717732Z	b2ba8d7	PR opened
🔒 Security Review	✅ Completed 2026-09-24T23:59:03.892571Z	b2ba8d7	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.