Scaled Legal product development lab
Built on my work
Scaled Legal products point the same checked, logged agent work at a client's problem.
In development, in closed beta. These products are not generally available yet.
Ask about the closed betaThe public work underneath: research, method and plain-language notes.
California law · 5 checks
Compliance screens
A screen loads a site's homepage in a real browser, watches for eight seconds without clicking anything, respects robots.txt and hashes what it records as evidence. It looks for five gaps under California law:
- Trackers before consent
Third-party requests that fire before a visitor agrees, under the California Invasion of Privacy Act.
- Do Not Sell or Share
The opt-out link required by Civil Code § 1798.135.
- Auto-renewal
Disclosure and consent terms under Business and Professions Code § 17602.
- Data-broker registration
Registration under Civil Code § 1798.99.82.
- Accessibility
Automated axe checks on the homepage.
In a 20-site pipeline test, 12 sites could be assessed. On 7 of those 12, trackers fired before any interaction (Wilson 95% CI 32% to 81%). That tests the pipeline. It is not a prevalence estimate.
Offered as a free checklist, a fixed-fee remediation or ongoing monitoring.
| Check | Result | Evidence recorded |
|---|---|---|
| Trackers before consent | Flag | 3 analytics requests in the first 8 s, before any banner interaction |
| Do Not Sell or Share link | Flag | No matching link text on the homepage |
| Auto-renewal terms | Not applicable | No subscription offer found |
| Data-broker registration | Not applicable | Not a data broker on the facts observed |
| Accessibility (axe) | Flag | 4 violations: color-contrast, missing form label |
confidential · public data
Whistleblower screening
A structured first look at whether a lead built from public data could support a claim, written for attorney review. Screens are confidential and never published. Nothing is filed, submitted or sent to an agency without an attorney's decision.
- Confidentiality first: the lead stays off public channels.
- Vehicle: the federal False Claims Act (relator share 15% to 25% when the government intervenes, 25% to 30% when it does not), the California False Claims Act (15% to 33%, or 25% to 50%), and the SEC, CFTC (10% to 30%) and IRS (15% to 30%) programs.
- The public-disclosure bar and whether the client could be an original source.
- A provenance log for every data point.
- A memo for attorney review.
| Step | Fictional entry |
|---|---|
| Confidentiality | Kept off public channels; no outreach to the source |
| Vehicle | Federal FCA and California FCA; SEC program not applicable |
| Public-disclosure bar | Allegations partly in a 2024 news report; original-source question open |
| Provenance | 11 data points, each with source URL, fetch time and sha256 |
| To attorney | Memo to the reviewing attorney with three open questions |
routing model
Legal department as a service
Custom and post-trained models and systems engineering, in partnership with Litil.
Most legal spend sits in the middle: generalist hours on work that is either routine or needs real judgment. This model splits it. Routine requests run on playbooks the client controls, with my work underneath and attorney review on top. Questions that need judgment go to senior counsel with a clean file and one clear question, so the client pays for the judgment and not for counsel catching up. The in-house team keeps the context.
On a $3M annual legal spend, the working model returns about half. Treat that as an illustrative model, not a result.
| Line | Today | With routing |
|---|---|---|
| In-house team | $1.5M | $337.5K core team |
| Outside counsel | $1.14M | $360K senior counsel |
| Tools | $360K | $129K product |
| Agent work with attorney review | $507K | |
| Forward-deployed legal engineer | $166.5K | |
| Total | $3M | $1.5M |
| Measure | Fictional value |
|---|---|
| Requests received | 412 |
| Closed on a playbook | 71% |
| Sent for attorney review | 38 |
| Sent to senior counsel | 9 |