Platform updates and legal-monitor detections, in reverse chronological order. Each entry has its own structured-data anchor so search engines and AI assistants can cite specific changes by date.
Legal-entity fragments of the same company (e.g. a parent corporation plus its many LLC subsidiaries) now roll up to one canonical employer page, and each employer carries a normalized industry from a fixed 25-category taxonomy instead of free-text.
Court records name the same employer dozens of ways across filings. We built a canonical employer graph that collapses those variants — deterministic matching for exact duplicates, then a model pass (with web verification) for cross-name corporate families — so one company has one page that aggregates its full litigation footprint. Industries are now mapped to a controlled vocabulary (retail, healthcare, finance, transportation, manufacturing, government, and ~19 more) rather than the raw strings extracted from individual opinions, which makes industry filtering and comparisons reliable. Low-confidence or ambiguous merges (e.g. distinct institutions that merely share a place name) are deliberately left un-merged.
Employer pages now show a single list of federal cases — court dockets and published rulings merged into one view — with a top filter for all cases / decided / open dockets. An outcome badge appears only when a case reached a published ruling.
Previously an employer page stacked two separate tables: federal dockets and analyzed court rulings. They are now one list joined on the docket-to-opinion relationship, so each row is one case and the outcome (employer win, employee win, settlement, dismissed, mixed, remanded) shows when the case produced a published decision. Filters let you narrow to decided cases or to open dockets without a ruling yet. The data is built on primary/foreign-key relationships between employers, dockets, and opinions rather than text matching.
Court dockets that produced a published opinion are now connected to that ruling, so a case can show its outcome and a ruling can surface its full docket (parties, defense counsel, disposition).
Most lawsuits settle or are dismissed without a published opinion — which is exactly why docket data surfaces employment cases the opinion corpus alone never shows. Where a docket and an opinion are the same case, they are now bridged on court and normalized docket number, letting outcomes flow onto the case list and connecting the structured docket record to the analyzed decision.
Attorneys in the directory are matched to the federal cases they appear on, and firm pages aggregate their attorneys' cases — a factual record of appearances, with plaintiff names redacted.
Tens of thousands of attorney-on-case records from federal dockets were matched to the verified attorney directory by name and location. Attorney and firm pages surface those appearances factually — caption, court, year, nature of suit, side — without editorial judgments. Cross-links connect a case's employer, its attorneys, and their firm.
Case captions drawn from federal dockets now render the individual plaintiff as "Employee" (e.g. "Employee v. [Employer]"), keeping the employer visible while protecting the worker's identity. Government and agency plaintiffs are preserved.
The platform is plaintiff-protective by design. Docket captions, which name individual workers, are redacted at display time to a generic "Employee" while keeping the defendant employer — the entity the page is about. Agency enforcement actions (e.g. EEOC) keep the agency name. The goal is to surface employer litigation patterns, not to index private individuals.
Employers that appear in federal dockets but have no published opinion now render a complete profile of their docket footprint, instead of returning a not-found page.
The earlier site only built employer pages from analyzed opinions. With the docket layer added, roughly a hundred thousand additional employers — many of whom never produced a published opinion — now have pages showing their federal case footprint. These were previously advertised in the sitemap but returned 404s; they now serve real content.
Employer profile pages now serve a stable, self-canonical URL. A class of profiles that previously redirected off their own listed URL — scattering search signals — now resolve correctly.
When two distinct companies normalized to the same URL slug, the disambiguated variant could redirect away from the exact URL the sitemap advertised, sending crawlers in circles and splitting ranking signals. Employer pages now treat their stored slug as canonical and only redirect known aliases, so each profile consolidates its authority on one URL.
The employer directory is now a top-level nav item, and the /employers index shows clear corpus totals — profiled employers, analyzed court rulings, and federal dockets — over an A–Z browse.
Employer litigation history is one of the most useful things a worker can check before acting, so it now sits in the primary navigation. The index page headline numbers were reconciled to a single source so the totals (employers, rulings, dockets, jurisdictions) are internally consistent.
The read-only Model Context Protocol server — which lets AI assistants query the corpus — now reports the canonical employer graph and the federal docket corpus alongside ruling counts.
As discovery shifts toward AI assistants, the platform exposes its data through an MCP server so models can ground answers in real employment-law records. Its corpus-stats tool now includes the number of profiled employers and the size of the federal docket corpus, not just the rulings count.
Before shipping the employer/docket work, a multi-agent adversarial review verified each change against live data — catching and fixing canonical, redaction, and data-integrity issues prior to release.
New features were reviewed by independent agents across SEO, data-integrity, privacy, and correctness dimensions, with each finding adversarially verified against the production database before it was accepted. Confirmed issues — a self-redirect class on collision slugs, a docket-only 404 class, and several edge cases — were fixed in the same pass.
A persistent employers table — one row per real company, deduplicated and junk-filtered — now anchors the platform, joined to a large federal court-docket corpus alongside the existing analyzed opinions.
This is the foundation for employer litigation track records: a canonical entity for each company that both court opinions and federal dockets attach to, with consistent trust filters (hide junk, drop low-confidence extractions). It replaced an ad-hoc, per-request name aggregation that drifted as the corpus grew.
When admin approves a detected legal change with field updates, the laws table now updates accordingly — closing the loop between the cron monitor and the plaintiff-facing rights engine.
The legal-monitor cron has been detecting law amendments for months, but the "approve" action in the admin review queue only flipped the change-row status without patching the laws table. That meant detected changes didn't flow through to plaintiff-facing rights advice — only the cron-detected timestamp updated. The review API now wires the apply path end-to-end via applyChangeToLaw(), and the admin UI surfaces three distinct actions: "Apply patch to <lawId>" (real DB change), "Mark applied (no DB change)" (informational ack), and "Dismiss". Reviews happen on a bi-weekly cadence.
Eliminated a class of sitemap duplicate-URL bugs by switching from offset-based pagination to deterministic ID-range buckets. A ruling's sitemap page is now determined by its own ID — stable across cache cycles and ingest churn.
A Screaming Frog scan flagged 9 ruling URLs appearing in two different sub-sitemaps each. Root cause: offset-based pagination shifted ruling positions whenever rows were hidden between cache renders, so the same URL could land in two pages over a 24-hour window. The fix partitions rulings into fixed 500K-wide ID buckets — a ruling's bucket is determined by its own ID and never moves. As a side effect, sub-sitemaps now grow uniformly with the corpus instead of always emitting exactly 10K URLs.
Variants like "/rulings/claim/Disability%20Discrimination" and "/rulings/claim/disability_discrimination" now redirect to a single canonical hub. Non-employment claims that were leaking into the URL space (Bivens action, antitrust) now 404 instead of rendering empty pages.
Ruling detail pages, employer profiles, state hubs, and industry hubs were all building claim hub URLs with raw, unsanitized claim_types strings from Supabase. A new claimSlug() helper normalizes case, punctuation, and aliases (e.g., "FLSA" → "wage_theft", "Title VII" → "discrimination") and the route handler 308-redirects non-canonical inbound URLs. SEO authority that was fragmented across thousands of near-duplicates now consolidates onto the canonical hubs.
The "Similar Rulings" widget on every ruling detail page now applies the same trust filters the hub pages use: extraction confidence ≥ 0.4, known outcome, and at least one strong-employment claim type. Non-employment rulings (immigration, antitrust) no longer surface as suggestions.
Similar Rulings is the largest single source of internal links on the site (5 per ruling × ~146K rulings ≈ 700K links), so even a small fraction of non-employment rulings here distorts Google's topical understanding of the domain. One immigration case had attracted 2.7K cross-links before this filter shipped. The new filter aligns Similar Rulings with the trust standard used elsewhere — when an attorney reviews a ruling page, the suggestions they see are now on-topic.