Glossary
Terms used throughout the Sentinel v2 documentation, in alphabetical order.
Terms
| Term | Meaning |
|---|---|
| Access level | What an API key may call: public (widget endpoints), restricted (server-side endpoints), or full (server-side endpoints, plus the account-management endpoints its security group's role grants). It is set by the key's security group, not the key. See API keys. |
| Account | Tenant that owns API keys, security groups, and training data; users get access to it through roles. See Accounts and users. |
| Admin session | A signed-in user's session token (JWT), sent as a Bearer token or jwt cookie; required by the admin endpoints and never accepted where an API key is expected. Account-management endpoints also accept a full API key whose security group has a role. See Sign-in. |
| AI decider | Optional classification category (ai) that asks a remote decision model yes/no questions about a text; it can only raise the score. See AI decider. |
| Alert | A stored notification Sentinel raises about logins, verifications, API key use, replicas, remote files, storage, or releases, optionally delivered to webhooks or push. See Alerts. |
| API key | Credential used by the widget or your backend; its access level and rules come from its security group. See API keys. |
| Base URL | The public URL a listener is reached at. [server] base_url prefixes the URLs handed to widgets (verifyUrl, audio); [auth] base_url or [server.admin_ui] base_url sets the sign-in and passkey URL. See Listeners. |
| Campaign | The same text, or a near-duplicate, sent by at least min_sources distinct sources within a time window; scored by the CAMPAIGN rule. See Similarity. |
| Category | A group of classifier rules scored together, such as device, email, ip, location, text, similarity, his, or ai. See Classifier rules. |
| Classification | See Verdict. |
| Classification score | The top-level score: the mean of the categories that ran, floored to 2 decimals, then raised to minimums by strong signals (floors). Below 1 is GOOD, below 2 NEUTRAL, otherwise BAD. See Scoring. |
| Code challenge | An image or audio code the user must type, shown when the security group's rules require it. See Code challenge. |
| Doctor | Diagnostic report of config, storage, network, license, and cluster checks, available as altcha-sentinel doctor and through the API. See CLI reference and Troubleshooting. |
| EDK | Ephemeral Device Key: a daily device identifier derived from the client IP, User-Agent, and Accept-Language, rolling over at midnight in [edk] time_zone. |
| Floor | A minimum the classification score is raised to when a strong signal fires (for example a triggered injection rule raises it to at least 2, so BAD). See Scoring. |
| HIS | Human Interaction Signature: a bot probability (0.01–1) computed from the pointer, touch, scroll, focus, key, and click samples the widget collects. See Human Interaction Signature. |
| Instance slot | A license-counted lease each running process holds in the database (30-second lease, renewed every 10 seconds); a process over the license's instances limit gets 503 LICENSE_INSTANCE_LIMIT. See Clustering. |
| KV store | Key-value store for replay protection, rate limits, and penalties: embedded by default, or Redis-compatible via [redis]. See Databases. |
| Listener | One HTTP server Sentinel binds: the main API ([server]), and optionally the widget-only public API ([server.public_api]) and the admin UI ([server.admin_ui]). See Listeners. |
| Monitor mode | A rule that is evaluated and logged (in context.monitor) but has no effect on the request. See Security rules. |
| Near-duplicate | A text that matches a stored sample or recent message almost exactly (similarity score 1); scored by the NEAR_DUPLICATE rule. See Similarity. |
| Passkey | A WebAuthn credential a user registers as a second factor or to sign in. See Sign-in. |
| Penalty | Stateful 0–10 score per account and client IP, raised by misbehavior (failed verifications, rate-limit pressure, penalty.add rules) and expiring [penalty] ttl after its last increase. See Scoring. |
| Preset | Admin UI template that writes named rules (preset:<id>:<part>) into a security group. See Presets. |
| Remote file | A list or database (classifier lists, GeoIP .mmdb, threat lists, models) downloaded and refreshed on a schedule. See Remote files. |
| Replica | One running Sentinel process; replicas sharing a database form a cluster. See Clustering. |
| Request log | Per-request record (classification, triggered rules, context) of API keys with logging enabled, stored in the database or ClickHouse. See Request logs. |
| Risk | Stateless 0–10 score from the device, IP, and location rules, returned by GET /v2/inspect and usable in the risk rule condition. See Scoring. |
| Role / membership | A user's permissions within one account. Root can also give a full security group a role, which grants its API keys those permissions in the group's account. See Accounts and users. |
| Rule | One entry in a security group's ordered rule list, with action allow, deny, accept, or set. See Rules reference. |
| Security group | Access level plus an ordered list of rules applied to every request made with its API keys. See Security groups. |
| Snapshot | Export of the whole database as a ZIP file, optionally age-encrypted, restorable with POST /v2/snapshots/_restore. See Snapshots. |
| Step-up verification | A fresh second-factor check that returns a short-lived step-up token (x-step-up-token), required for sensitive account changes. See Sign-in. |
| Threat entry | A manual or automatic record for an IP or network with a threat verdict. See Threat intelligence. |
| Threat source | A remote IP list Sentinel downloads and matches client IPs against. See Threat intelligence. |
| Threat verdict | The outcome of threat matching for an IP: allow or block. See Threat intelligence. |
| Training group / sample | Labeled example texts (samples) organized in groups, used for similarity matching. See Similarity. |
| Training suggestion | A text the text rules found spam, queued for review to accept as a training sample. See Similarity. |
| User / root | An admin identity that signs in; root is a flag granting full access to every account, not a role. See Accounts and users. |
| verificationData | The signed, URL-encoded string POST /v2/verify returns with the classification, score, and signals, for your backend to check and act on. See Verification. |
| Verdict | The classification of a request: GOOD, NEUTRAL, or BAD. See Scoring. |
| Web Bot Auth | Signed HTTP requests by which AI agents and crawlers identify themselves; matched with the agent rule condition. See Bots and agents. |