Review queue
Stage a batch of files here and work it top to bottom — every file is scored and triaged automatically, and each one you open is checked off so you can see what is left. Start with Return and Review. Reviewing against —.
How the queue works
Each file gets a disposition automatically:
- Return — a confident, deterministic problem; the file shouldn't pass as-is.
- Review — something was surfaced but not auto-decided, so it routes to a person.
- Clear — no issues found against your active checks.
Nothing is auto-rejected — LCS flags, a person decides. Click any row to open it in Review.
Your queue and your progress are saved on this device, so you can stop and pick up where you left off. Nothing is transmitted.
| File | Disposition | Score | Leading finding |
|---|
Activity
Every confirm and dismiss your team logs while reviewing — the audit trail, and the signal that tunes future checks. Stored on this device only.
| File | Finding | Severity | Action | Source | When |
|---|
Set up your rulebook
Pick a starting template, load your SOPs, and choose which checks run. LCS reviews every file against this.
How configuring works
Checks are deterministic — they test fixed facts: a required field, dates in the right order, a matching identifier, an attached exhibit. They give the same result every time. Anything that needs human judgment is flagged separately and never decided for you.
Every finding links to the exact SOP clause it came from, so a reviewer can check it in one click. Use Build checks below to create a starting set of rules from your SOPs and past returned files.
Starting template
Citations in every finding point back to the exact clause here, so a reviewer can verify the basis in one click.
Advanced — edit checks as JSON
Build checks from your SOPs & returned files
Upload a regulation or SOP (PDF, Word, or scanned) using the button above. LCS reads it on this device and, when you click Suggest checks, pulls out the rules it states: required content and attachments, value limits, banned wording, and filing deadlines. Review each one, fill in any blanks, and add the ones you want. Nothing leaves this device; only the rules you keep are saved.
Draft a writing-style guide
Turns the guide text above into advisory style rules. Draft on-device runs entirely in your browser — nothing leaves it. Draft with AI uses the AI provider your organization set up, which handles free-form guides better. Either way, you review and apply each rule; nothing is added for you.
The checks above run on this device. Publish to organization derives enforceable rules from your SOPs / writing guide and applies them to every reviewer in your department — only the rule configuration is sent, never any document.
Review a file
Paste or upload a document. Every finding cites the rule it came from — confirm or dismiss each one.
How review works
LCS scores the document, lists each finding with its severity and the cited SOP clause or check, and highlights the spot in the text. You decide: Confirm a real issue or Dismiss a false alarm — every decision is logged to this device with the file, the cited rule, your role, and a timestamp.
When you're done, download the audit record for the file — a complete, self-contained account of what was checked, what was found, the rules cited, and the decisions made. It carries a hash of the document for integrity and never leaves this device. The running decision log lives under Activity; replay past files under Backtest.
Routing preview under the editor shows how the same result reads to a Reviewer, Supervisor, or Auditor. It's a preview only — it doesn't change your permissions.
Shortcuts: ⌘/Ctrl + Enter runs a review. With a finding focused (Tab to it), press C to confirm or D to dismiss.
Reviewing against —. Findings cite their source; confirm or dismiss each one — those decisions feed your organization's check-calibration report.
Findings
Review JSON — the API contract
Review a case
Add the documents that make up one case. LCS reviews each, checks they agree on shared identifiers and dates, and confirms every required document is present — then recommends. Nothing leaves this device.
Prove it on past files
Replay files already decided in — and see how many returns LCS would have caught first.
How the backtest works
These files were already reviewed and tagged with what actually happened — approved or returned. LCS replays each and reports the returns it would have caught before a person opened the file, plus the time and cost that represents.
On the deterministic checks you configured, detection here is exact. Real-world recall depends on how completely your SOPs are encoded — so treat this as a floor you raise by adding checks.
Prediction vs. real outcome
| File | Outcome | Predicted | Score | Leading finding | Result |
|---|
Executive analytics
Organization-wide compliance posture, readiness, the most-confirmed checks, and calibration & training signals — aggregated review metadata only. Documents are never stored.
Ask your policies
Answers are drawn only from your organization's approved policy versions, with citations, and the assistant says so when your policies don't cover the question. Your question is masked on this device and sent browser-direct to your organization's AI - never through LCS. This is informational guidance about your own written policy, not legal advice or a determination.
Coverage & obligations report
An examiner-ready summary built from your organization's tamper-evident review trail: files reviewed, findings worked, and statutory deadlines met or missed. Documents are never stored — only review metadata.
Deployment & data handling
Control exactly what — if anything — is allowed to leave this device.
What these settings do
Checks run entirely in your browser. The default, Restricted, makes no network connections at all — documents are reviewed in memory, and nothing is sent or saved anywhere else. Use it for CJI, CUI, and law-enforcement-sensitive material.
Standard adds an optional AI judgment pass. De-identified, on-device-masked text goes browser-direct to your organization’s AI — Anthropic (your organization’s shared key or your own) or an endpoint you control — never through LCS. The deterministic engine still runs entirely on-device, and the AI pass is advisory.
Data sensitivity mode
Data minimization
Information sharing & approvals CHECKS & BALANCES
Every file carries an exposure score — the value if its information were leaked or shared outside. Releasing a higher-exposure file (its audit record carries the full document) needs sign-off from a higher role. Set the minimum role that may release each tier; a request below it routes to that role for approval and is written to your tamper-evident trail.
AI judgment pass OPTIONAL · ADVISORY
The on-device engine does most of the review without sending anything anywhere. The AI pass adds judgment about meaning — for example, whether a narrative actually supports its conclusion. It never decides on its own and never changes the score; it only adds notes for a human to confirm. On each review it masks personal information on-device and runs automatically — structured identifiers (SSNs, card numbers, emails, phones) are masked first; names and narrative are not, so this is de-identification, not anonymisation. A report that still carries high-sensitivity markers after masking is held back and never sent, and the exposure panel shows what was masked. Your organization brings its own key — LCS holds no AI key, and the de-identified text goes directly from this browser to the provider you choose, never through LCS. AI Disclosure.
Stored only in this browser, on this device, and sent directly to Anthropic with each pass — never to LCS, never to any LCS server. Use a key scoped to this purpose; clearing this browser's data removes it. A key held in a browser is a tradeoff Anthropic flags — if you'd rather keep it on a server, choose the self-hosted endpoint instead.
Two-factor authentication ACCOUNT SECURITY
Protect your account with a one-time code from an authenticator app (Google Authenticator, 1Password, Authy, Microsoft Authenticator). Applies to your email/password sign-in. Many security reviews require it.
Change password ACCOUNT SECURITY
Update the password for your email & password sign-in. Choose something long and unique.
Organization Anthropic key YOUR ANTHROPIC KEY
One key for your whole organization. It is stored encrypted on the server and powers Draft with AI in the Rulebook and — if you choose the shared organization key mode for the judgment pass — that pass too. For drafting, the key is used server-side only and never returned to the browser. For the shared-key judgment pass, it is delivered to each signed-in reviewer's browser at review time so the browser can call Anthropic directly — your case documents still never go to LCS. Prefer the key never reach a browser? Use a self-hosted endpoint instead. Only an administrator can set or remove it; the on-device drafter needs no key. AI Disclosure.
Organization Voyage key SEMANTIC GROUNDING — OPTIONAL
Optional. When set, the judgment pass is grounded against your own policy library using semantic retrieval (Voyage embeddings) on top of the always-on keyword match. The key is stored encrypted on the server and, at review time, delivered to the signed-in reviewer's browser so the browser embeds a de-identified query directly with Voyage — your case documents never go to LCS. Leave it unset and grounding still runs in keyword-only mode with zero added egress. Only an administrator can set or remove it. AI Disclosure.
Local data on this device
Your edited SOPs, checks, assumptions, and reviewer decisions are saved in this browser only — never sent anywhere. Export the decision log to use in your own training pipeline, or erase everything from this device.
Team
Manage the members of your organization. New accounts require your organization's principal to approve them before they can sign in.
Add a member
Members you add stay pending — unable to sign in — until your organization's principal approves them. You'll get a one-time temporary password to hand them.
Members
| Member | Role | Status | |
|---|---|---|---|
| Loading… | |||
Policies
Your versioned policy library across the regulatory hierarchy. Every edit is a new immutable version with an effective date and integrity hash — roll back or archive without losing history.
Add a policy
A policy belongs to one layer of the hierarchy (federal → team). Saving creates version 1; later edits add new versions, each with its own effective date.
Policy analysis
Compare the enforceable checks on your active policy versions and flag where the same requirement is defined at more than one layer, with a different severity, threshold, or parameter. Higher layers govern (federal → team). This surfaces overlaps and potential conflicts for a human to resolve — it is not a legal determination.
Version history
Add a new version
Audit trail
A tamper-evident record of who changed what in your organization — member, policy, and system-of-record actions. Metadata only; no document content is ever logged.
What’s in the audit trail
| When | Actor | Role | Action | Target | Details |
|---|---|---|---|---|---|
| Loading… | |||||
Help & quick start
How LCS works, what each section does, and how to read a result.
Quick start · three steps
- Open your Rulebook. Setup → Rulebook holds your organization's SOPs and the checks documents are reviewed against. It comes preloaded for your workspace; an administrator or principal can edit it.
- Review a document. Review → Review. Paste text or upload a file (PDF, Word, or text), then Run review. LCS scores it and lists every issue, citing the exact rule or clause.
- Act on the findings. Confirm the ones that matter and dismiss the rest, then download the audit record for your file. Nothing is decided for you — LCS flags, a person decides.
Everything runs in your browser. By default no document ever leaves this device.
What each section does
- Queue — files already scored and triaged. Start with anything marked Return.
- Review — run a single document through your checks and work the findings.
- Activity — a log of the decisions made on this device, exportable for your records.
- Backtest / ROI — replay past files against the real outcomes to measure accuracy and time saved.
- Rulebook admin / principal — edit SOPs and checks, get check suggestions, and draft a writing-style guide.
- Settings admin — data-handling mode, PII masking, your organization's AI key, and local-data controls.
Reading a result
After a review, the panel shows a compliance score, an overall call, and a list of findings.
- Return send it back — a confident, significant rule failure needs correction before it proceeds.
- Review a person should look — there are findings, but none rise to an automatic return.
- Clear nothing flagged — no issues found against your checks.
Each finding cites the rule or clause behind it. Style notes show as writing guide · cites … and are advisory — they never change the call. Mark each finding Confirm or Dismiss, then use Download audit record for a self-contained, signed-by-hash log of that file and your decisions.
Drafting a writing-style guide
In the Rulebook, paste your style guide text, then choose how to turn it into advisory checks:
- Draft on-device — instant, needs no setup, and never leaves your browser. Best for clear rules like “use active voice” or “spell out acronyms.”
- Draft with AI — for free-form guides. Requires your organization's Anthropic key (Settings → admin). Only the guide text is sent — never case documents.
Either way you review the proposed rules and click Apply. Nothing is added until you do.
Keyboard shortcuts
- ⌘/Ctrl + Enter — run the review (from the document box)
- C — confirm the focused finding
- D — dismiss the focused finding
Troubleshooting & FAQ
- My review found nothing / scored 100. That is a clear result — nothing failed against your configured checks. The deterministic engine only tests what your Rulebook defines; if you expected a finding, confirm the relevant rule is enabled in Setup → Rulebook.
- I can’t save to the system of record. Logging completed reviews to your organization’s tamper-evident trail is an administrator setting, and it is off in demo workspaces. If the “Add to system of record” action is hidden, your workspace does not have records enabled — ask your administrator to turn it on.
- The AI second read / suggested edits aren’t appearing. The AI judgment pass is optional. It runs only when your workspace is in Standard mode (not Restricted) and your organization has an Anthropic key set (Settings → admin). With no key, review stays fully deterministic and on-device.
- What is “semantic grounding” / Voyage? When the AI pass runs, it can rank passages of your own policy library by meaning using Voyage embeddings — only if an organization Voyage key is set; otherwise it falls back to keyword matching. Neither affects the deterministic score, and your case document is never sent to Voyage.
- My file won’t upload or extract. LCS reads PDF, Word (.docx), and plain text. A scanned, image-only PDF has no text layer to read — paste the narrative text directly instead.
- Where is my document data? Documents are processed in this browser and are never transmitted to LCS. Your reviewer decisions and rulebook are saved in this browser only; clear them anytime in Settings.
- How do I change the checks? Only administrators and principals edit the Rulebook (Setup → Rulebook). Reviewers and supervisors work against it as configured — if you need a change, contact your administrator.
Getting help
- Rulebook, access keys, AI keys, records. These are configured per organization — your administrator sets them up and is the fastest path for access or configuration questions.
- Report a security issue. See the Security page and security.txt for responsible-disclosure contact.
- Data, privacy & compliance questions. Email support@lcs-global.com, or see Data & compliance.
Privacy & data handling
Document review is deterministic and runs entirely on this device. In Restricted mode (the default) nothing is transmitted — the right setting for CJI, CUI, and sensitive material. Reviewer decisions and your edited rulebook are saved in this browser only; clear them anytime in Settings. The only thing ever stored on the server is your organization's configuration (and, if an admin sets one, an encrypted AI key) — never member documents. If your administrator turns on the optional system of record, completed reviews log metadata (findings, decisions, deadline statuses, file hashes) to your organization's tamper-evident trail and Coverage report — the document itself still never leaves this device.
Privacy Policy · Terms of Service · Security & disclosure · AI Disclosure · Data & compliance · Accuracy Validator
Common questions
I uploaded a scanned PDF and got no text.
Scanned PDFs are images. LCS will offer to run on-device OCR to recover the text, then review it. OCR reads the words, not the layout.
“Draft with AI” says it isn't configured.
The AI path needs your organization's Anthropic key, added by an admin in Settings → AI writing-guide drafting. Until then, use Draft on-device — it needs no key.
Which file types can I review?
Plain text and Markdown, Word (.docx), and PDF — including scanned PDFs via OCR. You can also paste text directly.
Can I change my organization's rulebook?
Only administrators and principals can set up the Rulebook — add, edit, and remove rules and regulations, and upload source documents. Reviewers and supervisors work against the Rulebook as configured. If you need a change, contact your administrator.
Where do access keys come from?
Your administrator issues them. Each key sets your organization, workspace, and role (reviewer, supervisor, or admin). Sign out from the top-right menu to switch accounts.
Escalations
Files routed to you for sign-off. Uphold the reviewer's call, override it with a recorded reason, or send it further up the chain.
Nothing is waiting on you right now. When a reviewer routes a file to your level, it appears here.
Export approvals
Requests to release sensitive files outside the device. Each shows the file’s exposure — the value if leaked — and who asked. Approve to authorize the release, or deny it. Every decision is recorded.
No export requests are waiting on you. When someone requests to release a file above their level, it appears here.
Escalation routing
Decide how completed reviews move up the chain of authority — who they go to, and when. Routing directs who signs off next; it never decides the outcome.
The chain of authority
Work climbs this ladder. A reviewer starts every file; the rules below send it up to a higher level when needed.
Routing rules FIRST MATCH WINS
Checked top to bottom; the first rule that matches decides where the file goes. If none match, the reviewer's call stands.
Who can override
An override changes a file's disposition after it's been routed — because not every call is final. Anyone at or above this level (and at or above the file's current level) may override, and must give a reason that is kept in the record.