Arbitr is an AI context platform. It gives your organization one governed memory, Cortex, holding your approved claims, terminology, rules and product knowledge. Arbitr’s agents and checks use that memory to create and review content, so what you publish stays accurate for every market and every language.
How Arbitr works
Cortex, your governed memory: The single source of what your organization has approved. Cortex grows as your team works, so each piece of content benefits from every decision made before it.
Models and agents: Arbitr’s own small language models and agents do the work — creating, localizing and checking content — guided by what Cortex knows.
People where they matter: You set the bar and decide what needs a human. Reviewers, inside or outside your organization, only see the content that needs a decision.
What you can do in Arbitr
The guides in this help center are grouped by what you are trying to do:
Localization workflows — Run multilingual campaigns, from upload and agent configuration through review and delivery.
Cortex — Manage your glossaries, terminology and other approved knowledge, import existing assets, and approve what Arbitr learns.
AI Visibility — See how AI assistants describe and cite your brand across languages. Guide coming soon
Account & administration — Team members, billing, security, API keys, integrations and reporting.
Who uses Arbitr?
Content and localization managers: Plan, launch and track content work across products, markets and languages.
Reviewers and subject matter experts: Review and sign off the content that needs a human decision, in a focused, side-by-side workspace.
Marketing, product and legal leads: Make sure published content matches approved claims, terminology and regional requirements.
Administrators: Configure single sign-on (SSO), multi-factor authentication (MFA), roles, API keys and credit allocations.
Access to Arbitr is secured by enterprise single sign-on (SSO), multi-factor authentication (MFA), and role-based access control. Depending on your organization’s security policy, you can sign in using corporate identity providers (Google Workspace or Microsoft Entra ID) or credentials with mandatory verification.
Signing In
Open your browser and navigate to the Arbitr login page (/login).
Enterprise Single Sign-On (SSO):
If your organization has enabled Google or Microsoft SSO, click Sign in with Google or Sign in with Microsoft.
If your company enforces SAML 2.0 / Okta SSO, enter your corporate email address and click Continue with SSO. You will be redirected to your company’s identity provider.
Password Sign-In:
Enter your registered corporate email and password.
Click Sign In.
Session Persistence
Active sessions remain authenticated across browser tabs. If you remain inactive past your organization’s session timeout threshold (configured under Security Settings), Arbitr will prompt you to re-authenticate before saving any in-flight changes.
Multi-Factor Authentication (MFA)
When MFA is enforced by your administrator or when accessing Arbitr from an unrecognized device:
After submitting your primary credentials, you will be redirected to the MFA verification screen (/sso-mfa).
Open your authenticator app (such as Google Authenticator, 1Password, or Microsoft Authenticator) or check your registered mobile number for an SMS security code.
Enter the 6-digit verification code.
Click Verify Code to access your dashboard.
Accepting an Invitation
When an administrator invites you to join an Arbitr organization:
You will receive an invitation email containing a secure link.
Click the link to open the invitation acceptance screen (/invite/<token>).
If you do not have an existing account, enter your full name and choose a secure password.
If your organization enforces SSO, click Accept with SSO to bind your corporate identity to the organization.
Upon confirmation, you are routed directly to your assigned workspace.
Reviewer Access
If you were invited strictly as an Internal Reviewer, your account lands directly on the My Reviews queue (/reviews) rather than the administrative dashboard.
Resetting Your Password
If you forget your password and do not use corporate SSO:
Arbitr organizes localization operations into clear, dedicated surfaces. Depending on your assigned role and permissions, your top bar, navigation items, and landing views adjust to fit your responsibilities.
Top Header Bar
The sticky header bar sits at the top of every screen:
Logo & Home Link: Clicking the Arbitr logo returns you to your primary landing page — /dashboard for Admins, Owners, and Analysts; /reviews for Reviewers.
Organization Switcher / Badge: Displays your active organization context.
Language Picker: Select your preferred user interface display language (English, Japanese, etc.).
User Initials & Account Menu: Click your avatar or initials in the top right to access:
Your Profile & Role: View your current role (Owner, Admin, Analyst, Reviewer) and email.
Settings (/settings): Access organization configuration, team management, billing, and accessibility.
Agent Marketplace / Studio (/intelligence-marketplace): Browse available AI agents and language models.
Contact Support: Direct link to Straker technical assistance.
Sign Out: Safely terminate your active session.
Primary Navigation Bar
Directly beneath the header, the global navigation bar gives one-click access to core functional areas:
Dashboard (/dashboard): Your mission control. Highlights intelligence alerts, quick-start upload zones, recent translation projects, organizational health scores, and credit balances.
Projects (/projects): The campaign and job repository. Browse active, completed, and in-review translation projects, inspect deliverables, and check agent findings.
Cortex (/cortex): The central linguistic memory hub. Explore domain profiles, approved glossaries, translation memory segments, and learning contributions.
Governance (/governance): Comprehensive, tamper-evident audit logs documenting every automated agent action, human edit, and compliance certification. (Requires audit.read permission).
Reviews (/reviews): The dedicated task queue for human-in-the-loop linguists and internal reviewers. Highlights assigned translation batches, urgencies, and due dates.
Role-Based Landing Views
Arbitr customizes the initial view based on what you need to accomplish:
Focused Review Queue & CAT Workspace (Admin nav collapsed)
Responsive Design & Full-Width Workspace
When working inside the Review Workspace (/projects/<id>/review/<batchId>), the interface automatically expands edge-to-edge and suppresses extraneous navigation elements so you can focus on segment-by-segment review.
Cortex: Your Governed Memory
Cortex is Arbitr’s governed memory. It holds what your organization has approved — claims, terminology, rules, product knowledge and verified past work — and every agent and check in Arbitr works from it.
For localization, Cortex turns your past translation work into an active layer of knowledge that guides each new campaign, rather than an archive of finished files.
Cortex for Localization
For localization work, Cortex organizes your linguistic knowledge into four connected components:
2Terminology & glossariesApproved, do-not-translate and forbidden terms
3Domains & contextThe right terms and rules for each content area
4Contributions & continuous learningApproved human edits flow back into all three
1. Translation Memory (TM)
A translation memory stores previously translated sentences (segments) as bilingual pairs (source and target). Whenever you submit new documents, Arbitr queries your TM for matches:
100% Match (Exact): The source sentence is identical to a previously translated and verified segment.
In-Context Exact (ICE / 101%): The sentence matches an existing segment and shares the exact same preceding and succeeding sentences, guaranteeing contextual fidelity.
Fuzzy Matches (70%–99%): Similar sentences with minor differences (such as dates, product names, or punctuation). The Auto Post-Editor agent uses fuzzy matches to draft high-quality suggestions.
Matches reduce both project turnaround time and cost — exact matches incur no translation charges.
2. Terminology & Glossaries
Glossaries define mandatory corporate vocabulary, product names, acronyms, and forbidden terms across all languages:
Approved Terms: Words or phrases with designated official translations (e.g., product feature names, legal definitions).
Do-Not-Translate (DNT) / Brand Terms: Terms that must remain in their original source form (e.g., trademarks, model numbers, software identifiers).
Forbidden / Deprecated Terms: Outdated terminology or competitor names that agents actively replace or flag.
Glossary terms take precedence over general neural machine translation defaults.
3. Domains & Context Profiles
Not all content within an enterprise shares the same vocabulary. A term in a software interface might have a completely different translation in a legal contract or a marketing brochure.
Cortex organizes knowledge into Domains (e.g., Legal & Compliance, Marketing & Brand, Technical Product Documentation, Investor Relations). When launching a project, assigning the appropriate domain ensures that translation models and agents query the correct sub-set of terminology and style rules.
4. Contributions & Continuous Learning
Cortex is not a static database; it is a learning system. Whenever an in-country reviewer, internal manager, or Straker professional editor adjusts a translated segment in the Review Workspace, that change is captured.
Automated Capture: Segment updates are staged as proposed Contributions.
Verification & Promotion: Verified edits are committed directly into your active translation memory.
Immediate Compounding: The next project that contains that sentence or term automatically benefits from the human edit, preventing repetitive mistakes and continuously elevating quality.
Knowledge Coverage & Health Metrics
On your Dashboard and within the Cortex Overview, Arbitr displays real-time health metrics:
Total Intelligence Entries: Cumulative count of verified TM segments and glossary terms.
Active Coverage Rate: The percentage of incoming content volume that matches existing Cortex knowledge without requiring novel translation.
Learning Velocity: Number of new verified terms and segments ingested per week.
The Cortex Hub (/cortex) is where you manage your organization’s governed memory. It brings together translation memories, approved terminology, domain definitions, and continuous learning contributions into a single, searchable workspace.
Cortex Overview Cards
At the top of the Cortex dashboard, four key telemetry cards illustrate the strength and scale of your linguistic memory:
Total Intelligence Entries: Cumulative count of bilingual sentence pairs and verified glossary terms.
Knowledge Growth: Visual trend indicating new memory segments ingested over the last 30, 60, and 90 days.
Active Coverage: The percentage of incoming project volume that leverages existing memory without requiring net-new translation.
Learning Velocity: Recent verified contributions committed from human reviews.
Domains & Partitions
In the center of the screen, the Domains list displays all subject-matter categories configured for your organization:
Domain Cards: Click into any domain (e.g., Legal, Life Sciences, Marketing, Technical Hardware) to inspect its specific glossaries and translation memory partitions.
Quick Navigation
From the Cortex header, you can jump directly to specific asset managers:
Glossaries (/cortex/glossary): Browse, search, and edit approved and forbidden terminology.
Translation Memory (/cortex/memory): Search bilingual sentence pairs and verify match quality.
Import Assets (/cortex/import): Upload legacy TMX, CSV, or TBX files to seed your organization’s memory.
Contributions (/cortex/contributions): Review and commit pending learnings generated by human edits in the review workspace.
The Glossary Manager (/cortex/glossary) allows you to define, curate, and enforce corporate terminology, brand names, product identifiers, and prohibited vocabulary across all language pairs.
Term Types & Attributes
Each term in your glossary is configured with specific behavioral rules:
Approved Terms: Official translations that must be used whenever the source term appears.
Do-Not-Translate (DNT): Brand names, proprietary model codes, trademarks, or code identifiers that must remain in their original source form across all target languages.
Forbidden / Deprecated Terms: Outdated terminology, competitor names, or non-inclusive language. The Glossary Agent flags any occurrence of these words in raw machine translations and replaces them with approved alternatives.
Browsing and Searching Terms
Search Bar: Search by source term, target translation, or definition notes.
Language Pair Filter: Filter terms by specific source and target locales (e.g., English (US) → German (Germany)).
Domain Filter: Restrict terms to a designated domain (e.g., Clinical Trials vs. Consumer Marketing).
Adding a Single Term
In the top right corner, click Add Term.
Source Term: Enter the original phrase.
Target Translation: Enter the approved localized equivalent (leave identical for DNT terms).
Language Pair: Select the source and target locales, or check Apply to all target languages for global DNT terms.
Domain: Assign the term to one or more relevant domains.
Definition / Context Note: (Optional) Add context explaining when to use this term.
Click Save Term. The term takes effect immediately for all subsequent translation projects.
Editing & Deleting Terms
Edit: Click the edit pencil icon on any term row to update the translation or adjust its domain assignment.
Delete: Click the trash icon to remove a term. Once deleted, agents will no longer enforce or repair that term in future translations.
If your organization has accumulated linguistic assets in external translation tools, CAT software, or spreadsheets, you can bulk-import them into Cortex via the Import Assets Screen (/cortex/import).
Overwrite Existing: Replaces older Cortex translations with the newly imported version.
Append / Version: Creates an alternative translation variant for context matching.
Step 5: Ingestion & Validation
Click Start Import. Arbitr runs automated validation checks (verifying valid UTF-8 encoding, tag integrity, and ISO language codes). Once finished, a summary report displays the total count of ingested segments and terms.
The Contributions Queue (/cortex/contributions) is the gateway through which human expertise continuously enriches your organization’s AI models. Whenever a human linguist edits a segment in the review workspace or signs off on a Straker order, that correction is staged here for verification.
How Learning Contributions are Captured
Human Edit Trigger: An internal reviewer or Straker linguist modifies a machine-translated sentence in the Review Workspace.
Diff Extraction: Arbitr compares the original neural translation against the human-edited output to extract the corrected term or phrasing.
Staging: The new bilingual pair is staged as a Pending Contribution associated with the originating project, domain, and author.
Reviewing Pending Contributions
Administrators and Lead Linguists can inspect pending contributions:
Source Segment: The original sentence text.
Original Translation: What the machine translation engine originally generated.
Human Correction: The final approved translation authored by the linguist.
Detected Learning: Highlights whether this change represents a new terminology pair, a grammatical fix, or a stylistic improvement.
Contributor: Who made the edit (e.g., Sarah Jenkins, Marketing Reviewer).
Verification Actions
Approve & Commit: Approves the contribution and commits it immediately into your active translation memory and glossary. All subsequent projects will automatically apply this correction.
Edit Before Committing: Click the edit icon to refine the translation or adjust the domain categorization before committing.
Reject: Discards the contribution if the edit was a one-off situational exception not intended as a universal corporate rule.
Automated Promotion
Under Settings → Content Preferences, administrators can enable Auto-Commit for Certified Linguists. When enabled, edits authored by Straker Certified Linguists are committed directly to Cortex without requiring manual administrator approval.
How a Localization Campaign Works
Every localization campaign in Arbitr follows the same guided lifecycle. From the moment you drop a source document into the intake zone to the download of clean localized deliverables, each step is tracked, scored, and audited.
The 6-Stage Lifecycle
1
Intake & readingFiles are parsed and split into segments
2
Cortex match & pricingMemory matches found, cost calculated
3
Agent ensemble deployedAgents chosen for the content’s domain
6aMeets thresholdDeliverables auto-publish and are ready to download
6bBelow thresholdHuman review required. Choose a pathway:
Review it myself
Assign to team
Send to Straker
Stage 1: Document Intake & Extraction
You drop one or more files into the upload zone or initiate an API job. Arbitr supports common formats including Microsoft Office (Word, PowerPoint, Excel), InDesign (IDML), Adobe PDF, HTML, XML, Markdown, and standard localization files (JSON, PO, XLIFF).
Arbitr parses the document hierarchy, strips markup into protected inline tags, and segments the text into translatable linguistic units while preserving formatting.
Stage 2: Cortex TM Match & Cost Analysis
Before a single token is submitted to a neural model, Arbitr queries your organization’s Cortex translation memory:
Exact & In-Context Matches (100% / ICE): Previously translated and verified sentences are pulled directly from memory. These segments require zero translation and are heavily discounted or free.
Repetitions: Identical segments appearing multiple times within the same document are identified so you only pay for the first occurrence.
Volume & Credit Estimation: The system calculates the net billable character count, matches against your credit balance, and generates an exact cost preview before launch.
Stage 3: Agent Ensemble Deployment
Arbitr evaluates the document’s classified domain (e.g., Financial Earnings, Legal Contract, Clinical Trial, Technical Software) and recommends a specialized Agent Ensemble.
These agents prepare prompt contexts, inject approved terminology lists from Cortex, and configure specialized system constraints for the translation engine.
Stage 4: Neural Translation & Automated Repair
The segments are translated via high-performance neural machine translation (NMT) and large language model engines.
During and immediately following generation, active agents perform real-time verification:
Glossary Agent: Verifies that mandatory terminology was strictly applied. If an approved term was omitted or mistranslated, the agent repairs the segment automatically.
Style & Tone Agent: Calibrates phrasing against your organization’s configured tone (e.g., Formal, Professional, Conversational).
Arbitr assigns an objective Trust Score (0–100%) to the completed translation. This score synthesizes:
Lexical consistency with Cortex translation memory.
Compliance with mandatory glossary terms.
Language model perplexity and fluency evaluation.
Agent findings and severity of any flagged anomalies.
Stage 6: Delivery or Human Review
Based on the calculated Trust Score and your organization’s Auto-Publish Threshold:
Auto-Publish: If the Trust Score meets or exceeds your threshold, the final documents are compiled and made available immediately in the Deliverables tab.
Human Review: If the score falls below threshold or if auto-publish is toggled off, the project is gated. You can:
Open the Review Workspace to review and edit flagged segments yourself.
Assign the batch to an internal team linguist via the Reviewer Queue.
Dispatch the project to Straker Professional Linguists for certified translation, second-linguist editing, or desktop publishing (DTP).
The Continuous Learning Feedback Loop
Whenever a human reviewer edits a translation segment — whether an internal team member or a Straker linguist — the corrected source-to-target pair is returned to Cortex as a Learning Contribution. Once approved, this newly verified knowledge powers all future translation projects, ensuring the same mistake is never repeated.
Localization Agents & the Ensemble
In traditional machine translation, a monolithic model receives source text and produces target text with no opportunity for specialized critique or post-processing. Arbitr replaces this with an Agent Ensemble — a collaborative group of specialized AI agents working together to evaluate, correct, and certify each translated document.
What is an Agent Ensemble?
An ensemble is a curated group of autonomous agents assigned to a translation campaign. While the underlying neural translation engine translates segments, the ensemble performs parallel validation, domain adaptation, and linguistic repair.
Each agent has a specific mandate, specialized prompt constraints, and distinct validation criteria.
The Core System Agents
Arbitr includes several purpose-built agents that integrate directly with your Cortex knowledge base:
1. Glossary Agent & Terminology Repair
Purpose: Enforces 100% adherence to your approved domain and brand glossaries.
Behavior: Scans source segments for terms present in your active termbases. If the neural translation omits the approved term, mistranslates it, or substitutes a synonym, the Glossary Agent flags the anomaly and automatically injects the verified term back into the segment.
Reporting: Logs whether terms were honored, repaired, or flagged for manual review.
2. Auto Post-Editor (APE)
Purpose: Polishes raw machine translation into natural, fluent target prose using fuzzy translation memory context.
Behavior: Examines historical 75%–99% fuzzy matches in Cortex. Instead of requiring a human linguist to adapt the fuzzy match, the APE agent synthesizes the historical translation with the new source segment, producing high-confidence fluent rewrites.
Note: APE rewrites are applied directly to the delivered translation; it does not produce per-segment diagnostic flags.
3. Style & Tone Calibrator
Purpose: Aligns voice and register with your organizational content preferences.
Behavior: Evaluates sentence complexity, honorifics (e.g., Japanese Keigo, German Sie vs. du), and emotional resonance against your selected tone (Formal, Corporate, Friendly, Technical).
4. Compliance & Regulatory Agent
Purpose: Ensures legal disclaimers, mandatory disclosures, numerical formatting, and regional requirements are preserved.
Behavior: Verifies that currency symbols, date formats, copyright notices, and mandatory contractual clauses match regional statutory requirements.
Understanding Agent Findings & Flags
When viewing project results or inspecting segments in the review workspace, agent findings are color-coded and classified with clear verdicts:
Finding badge
Meaning
Action taken
ACCURATE
Approved terminology was correctly identified and used.
Verified as-is.
REPAIRED
The raw translation missed an approved term; the agent repaired it.
Corrected automatically in the output.
IMPROVED
The agent substituted a more natural or brand-aligned phrasing.
Applied to final translation.
GLOSSARY CONFLICT
An agent suggested a phrasing that conflicted with a locked glossary term.
Locked glossary term was preserved; suggestion suppressed.
NEEDS REFINEMENT
The approved term was used, but the surrounding sentence syntax is awkward.
Flagged for human review.
INACCURATE
A critical term or number is missing and could not be safely repaired.
Flagged as a priority issue for human review.
Finding Severities
Agent findings carry one of three severity levels to help prioritize human triage:
Critical: Missing negations, untranslated core terminology, corrupted numbers or currency values, or missing regulatory disclaimers. Gating for auto-publish.
Major: Awkward phrasing, incorrect grammatical gender/number, or minor tone mismatch.
Minor: Minor stylistic variance or alternative synonym preference.
Agent Marketplace & Custom Ensembles
Under the Account Menu → Agent Marketplace (/intelligence-marketplace), organizations can enable additional specialized agents designed for specific industries (e.g., Life Sciences, Financial Reporting, Automotive Engineering, Software UI Strings). Custom agents can be configured under Model Settings with custom prompt instructions and specialized terminology sets.
Trust Scores & Quality Baselines
Arbitr evaluates translations using objective, multi-factor quality scoring. Rather than forcing your team to manually inspect every document or blindly trusting unverified machine output, the platform calculates a Trust Score for every completed project and file.
What is a Trust Score?
The Trust Score is a composite metric expressed as a percentage (0% to 100%) that measures the reliability, accuracy, and brand alignment of translated output.
A higher score indicates that the content strongly aligns with your approved linguistic assets and exhibits high linguistic fluency with no detected critical errors.
Factor
Evaluated by
Glossary Compliance
Glossary Agent & Termbase checks
Translation Memory Match
Cortex Exact / ICE match ratio
Linguistic Fluency
Language Model Perplexity & NMT
Agent Finding Penalties
Severity of unhandled issues
How Scoring Factors Work
Terminology Fidelity (35% weight):
Evaluates whether all terms identified in your source documents were translated using your approved Cortex glossary definitions.
Deductions occur if terms were missing, translated to unapproved synonyms, or conflicted with forbidden term lists.
Memory Consistency (25% weight):
Scores alignment with historical translation memory segments. If sentences closely match established patterns in your TM, confidence increases.
Fluency & Grammatical Integrity (20% weight):
Measures target language fluency, syntax, grammatical agreement, and absence of hallucinations or awkward literal translations.
Major Issue: -5 points per occurrence (e.g., awkward phrasing, wrong grammatical register).
Minor Issue: -1 point per occurrence (e.g., slight stylistic variation).
The Auto-Publish Threshold
Under Settings → Content Preferences (/settings/content-preferences), administrators configure your organization’s Auto-Publish Threshold:
Auto-Publish OFF (Default for high-stakes enterprise accounts):
Every project stops at the Results screen with status Review Required.
Deliverables cannot be published or downloaded until an authorized reviewer approves the batch.
Auto-Publish ON (e.g., Threshold set to 85%):
Files scoring equal to or above 85% are marked Published automatically. Clean deliverable files are ready for instant download without human intervention.
Files scoring below 85% are routed to Review Required. Notifications are dispatched to your review team or project manager to resolve the flagged segments.
Tuning by Content Type
You can establish different auto-publish thresholds based on document category. For example, internal knowledge base articles might auto-publish at 80%, while customer-facing contracts or financial earnings releases can require 100% human sign-off.
Org Health & Quality Baselines
Your Dashboard displays your organization’s long-term Quality Baseline:
Average Quality Score: Rolling 30-day average of Trust Scores across all completed campaigns.
Domain Precision: Accuracy rate within specialized terminology subsets.
Escalation Rate: Percentage of files that fail the auto-publish threshold and require human review.
A declining escalation rate indicates that your Cortex knowledge base is maturing and successfully automating translation quality.
Credits & Costs for Localization
Arbitr uses a transparent, usage-based model powered by Credits (also referred to as Intelligence Credits or IC). You only consume credits for the net novel translation work required — meaning the larger and more mature your Cortex memory becomes, the less you spend on every project.
How Volume is Measured
Translation volume in Arbitr is calculated using Character Count (or word count depending on the script, e.g., CJK characters):
Total Characters: The full volume of translatable text extracted from your uploaded files.
Memory Matches: Characters that match existing 100% or In-Context Exact (ICE) segments in your Cortex translation memory. These are discounted 100% (zero translation charge).
Internal Repetitions: Identical sentences that appear multiple times within the same project. The first occurrence is billed at the standard rate; all subsequent occurrences receive a 75% repetition discount (charged at 25% of the base rate).
Before you launch any campaign, Arbitr displays a real-time cost estimate in the campaign drawer:
Base Neural Translation: Credits allocated for model inference and neural generation across all selected target languages.
Agent Ensemble Operations: Credits for specialized agent tasks (Glossary verification, APE rewrite, tone alignment).
Current Balance & Balance After Launch: Displays your remaining organization credit balance and warns you if additional credits are needed before launching.
Campaign cost estimate
Total Source Volume
12,450 characters
Leveraged from Cortex TM (100% Match)
- 4,200 characters
Internal Repetitions (75% off)
- 1,800 characters
Net Billable Volume
6,450 characters
Cost in Credits
645 IC
Current Credit Balance
5,000 IC
Balance After Launch
4,355 IC
What Happens if Memory Matches are Sent for Human Review?
If you decide to send a project to Straker Professional Linguists for human translation or post-editing:
Standard Policy: Human linguists review and verify the entire document context.
Fully Matched Batches: If you explicitly choose to include 100% TM-matched batches in a human review order, the platform displays a confirmation modal notifying you that human verification is billed on full character counts without the automated memory discount.
Managing and Topping Up Credits
Purchasing Credits: Organization Owners and Admins can purchase credit packages or subscribe to monthly tiers under Settings → Billing (/settings/billing).
Low Balance Alerts: When your organization balance drops below 20% of your average monthly usage, banner notifications appear on the dashboard and email alerts are sent to billing admins.
Credit Expiration: Purchased Credits remain valid for 12 months from the date of purchase.
Creating a translation campaign in Arbitr begins by uploading source documents, selecting target languages, choosing domain profiles, and reviewing volume and pricing estimates.
Step 1: Accessing the Upload Screen
You can initiate a new project from multiple points in the application:
On your Dashboard, click Upload documents or drop files directly into the central drop zone.
In the top navigation, click Projects → New Project (/upload).
Step 2: Adding Source Files
Drag and drop files into the upload target or click Browse Files to open your native file picker:
Supported Office Formats: Microsoft Word (.docx), Excel (.xlsx), PowerPoint (.pptx).
Design & Publishing Formats: Adobe InDesign (.idml), Adobe PDF (.pdf).
Web & Documentation: HTML (.html, .htm), Markdown (.md), XML (.xml), Plain Text (.txt).
You can upload multiple files simultaneously. Arbitr automatically groups them under a single campaign while preserving individual document structures and file names.
Step 3: Configuring Campaign Parameters
Once files are uploaded, configure the project settings in the setup panel:
Campaign Name: Enter a recognizable project title (e.g., Q3 Financial Disclosures or Mobile App v2.4 Localization).
Source Language: Select the language your documents are written in (auto-detected by default).
Target Languages: Choose one or more languages to translate into. You can select individual languages or pre-configured regional bundles (e.g., Top European Tier, APAC Core, LATAM).
Domain: Pick the subject-matter domain from your Cortex repository (e.g., Financial, Legal, Medical, Marketing, Technical). This determines which glossaries and translation memory partitions are prioritized.
Brand Tone: Select the preferred tone of voice (Professional, Formal, Conversational, Authoritative).
Step 4: Reviewing Volume & Cost
As soon as your files are parsed, Arbitr analyzes your text against your Cortex translation memory:
Total Characters & Words: The gross volume of content extracted from the files.
Leveraged from Memory: Words or characters identified as 100% or ICE matches that will be translated instantly without charges.
Estimated Credits: The net credit deduction required to process the campaign.
Balance Verification: If your organization’s available credits are insufficient, an alert appears with a link to Top up credits.
Step 5: Launching the Campaign
Click Configure Agents or Start Campaign:
If you click Start Campaign, Arbitr deploys the default agent ensemble for your domain and begins processing immediately.
If you click Configure Agents, you advance to the Agent Configuration screen to customize active agents before translation begins.
Configuring & Deploying Agents
Before translation processing begins, you have the option to inspect and customize the Agent Ensemble assigned to your campaign. This allows you to tailor automated validation rules specifically to the nuances of your content.
The Reading & Agent Selection Screen
During the initial reading stage (/projects/<id>/reading or /projects/<id>/agents), Arbitr displays:
Document Categorization: Shows the detected file types, layout complexity, and semantic classification.
Recommended Agents: The default suite of agents selected by Arbitr’s orchestrator based on your chosen domain and file structure.
Agent Cards: Each active agent card displays:
Specialization: What the agent monitors (e.g., Glossary Repair, Tone Alignment, Compliance Check).
Confidence Score: Predicted impact on translation accuracy.
Status Pill:Deployed (active) or Available (inactive).
Modifying the Ensemble
To customize your active agents:
Enable / Disable Agents: Toggle the checkbox on any recommended agent card to include or exclude it from this campaign.
Add from Marketplace: Click Hire from Agent Marketplace (/intelligence-marketplace) to browse additional industry-specific agents (e.g., Medical Device Regulatory Agent or Software String Variable Protector).
Inspect Constraints: Click an agent card to view its active rule set, linked Cortex glossaries, and prompt instructions.
Launching the Campaign
Once you are satisfied with the active ensemble:
Confirm the summary showing the total number of deployed agents (e.g., 3 agents deployed).
Click Start Campaign (or Deploy).
Arbitr deducts the estimated Credits and transitions the project to the Processing stage.
Monitoring Processing
While your campaign is in progress, the Processing Screen (/projects/<id>/processing) provides real-time telemetry into the extraction, neural translation, and agent validation pipeline.
The Four Processing Stages
Arbitr visualizes project advancement across four pipeline stages:
Queued: Files are queued in the distributed processing cluster, memory caches are warmed, and active glossaries are loaded.
Reading & Extracting: Documents are parsed, layout geometries are indexed, translatable text is separated from formatting tags, and TM matches are pre-populated.
Translating: Segments are submitted to neural machine translation and language model endpoints across all requested language pairs in parallel.
Reviewing & Repairing: The deployed agent ensemble runs validation passes, resolves glossary deviations, repairs sentences, and computes quality trust scores.
Operations Control Room View
For large-scale enterprise campaigns with dozens of files and target languages, Arbitr displays the Operations Control Room:
Overall Progress Percentage: Real-time completion progress across the entire campaign.
Job Matrix Table: Displays every target language batch with its respective status:
Extracting: Preparing document structure.
In Progress: Active translation and agent review.
Complete: Ready for results.
Needs Review: Gated due to quality score falling below threshold.
Filter Bar: Filter jobs by target locale, file type, or processing status to isolate bottlenecks.
Background Processing
You do not need to keep the browser window open. Arbitr executes translation jobs asynchronously. You will receive an in-app notification and email confirmation once processing completes.
What Happens on Failure?
If an unrecoverable error occurs (such as a corrupted PDF file or an invalid document password):
The affected file is marked with a red Failed badge.
A descriptive error banner explains the failure reason (e.g., File is encrypted and requires password unlock).
Non-affected files and language pairs continue processing to completion.
You can click Back to Upload to re-upload a clean version of the failed document.
Inspecting Results & Deliverables
When processing completes, Arbitr directs you to the Results Screen (/projects/<id>/results). This view consolidates quality evaluations, agent diagnostic findings, and final downloadable deliverables.
The Results Header & Trust Score
At the top of the screen, the primary metric card displays:
Trust Score: The composite accuracy score (e.g., 94%).
Target Threshold: Your organization’s configured auto-publish threshold (e.g., 85%).
Status Pill:
Published (Green): The score met or exceeded the threshold. Deliverables are verified and ready.
Review Required (Amber): The score fell below the threshold or auto-publish was turned off. Human sign-off is required before files are published.
Deliverables & Downloading Files
Click the Deliverables tab to access the final translated files:
Download All (ZIP): Packages all localized documents into a single archive matching your original directory structure.
Per-File Download: Download specific language outputs individually by clicking the download icon next to any document row.
Partial Downloads: If certain files are still being processed or pending human review, clicking download provides all currently available files and includes a _missing_files.txt manifest explaining what remains pending.
Deliverables Locked When Review is Required
If your project is in Review Required status and your organization policy blocks unverified content, final deliverable downloads remain gated until approved.
Agent Activity & Findings Breakdown
The Agent Activity section details exactly what your agent ensemble analyzed, flagged, or repaired:
Filter Controls
Filter by Agent: Focus on specific agents (e.g., Glossary Agent or Compliance Agent).
Filter by Verdict: Show only INACCURATE, REPAIRED, IMPROVED, or GLOSSARY CONFLICT.
Filter by File & Language: Narrow down findings to a single document or target locale.
Finding Cards
Each flagged segment displays a detailed comparative breakdown:
Segment Reference: The paragraph or sentence number in the document.
Source Text: The original text string.
Original Machine Translation: What the raw translation model originally produced.
Suggested / Refined Translation: What the agent repaired or suggested.
Approved Terms: The official Cortex glossary terms identified and enforced in this segment.
Verdict & Reason: The justification explaining why the agent acted (e.g., Approved term was missing from the machine translation — your agent put it back).
Taking Action: Getting Reviewed
If your results require human verification or if you want certified linguistic sign-off:
Click the prominent Get Reviewed button at the top right. This opens your review options:
Review it myself: Immediately opens the Review Workspace.
Assign to team: Dispatches the batch to an internal colleague’s Reviewer Queue.
Send to Straker: Hands the project off to Straker Professional Linguists.
Choosing a Review Pathway
When a translation batch requires human verification — or when you want an extra layer of human sign-off on sensitive enterprise content — clicking Get Reviewed on the results screen presents three flexible pathways.
The Three Review Pathways
Get Reviewed
1Review it myselfEdit straight away in the browser workspace
2Assign to teamA colleague’s review queue, with a due date
3Send to StrakerProfessional linguists and desktop publishing
1. Review It Myself
Best for: Fast, agile review by content owners or bilingual staff who want to make quick edits without waiting for task assignments.
How it works: Clicking Review it myself launches the full-screen Review Workspace instantly.
Workflow: You navigate sentence-by-sentence, edit target text directly, inspect translation memory suggestions and glossary terms, and click Approve & Sign Off when done.
Turnaround: Immediate.
2. Assign to Your Team
Best for: Distributed enterprise organizations with dedicated in-country marketing reviewers, subject matter experts (SMEs), or internal localization teams.
How it works:
Select the language batches to assign.
Choose an internal team member from the reviewer dropdown. The selector highlights team members with matching language pair skills and availability (Available, Away, Offline).
Set an Urgency level (Normal or Urgent) and an optional Due Date.
Enter a summary note providing context or instructions.
Click Assign to [Name].
Workflow: The assigned reviewer receives an email notification and finds the batch waiting in their personal Reviewer Queue.
Cost: No credit charge — internal team labor.
3. Send to Straker Professional Linguists
Best for: Regulated documentation, high-visibility marketing launches, legal contracts, complex multilingual desktop publishing (DTP), or when internal teams lack capacity.
How it works:
Select the batches and click Send to Arbitr / Straker.
Choose the required service tier:
Translation: Certified human translation.
Translation + Edit (TEP): A primary native linguist translates or post-edits, followed by a second independent specialist reviewer.
Desktop Publishing (DTP): Page-by-page layout reconstruction for InDesign, PowerPoint, Word, or PDF files.
Linguistic Sign-Off (LSO): Final proofing after DTP formatting is complete.
Enter delivery notes and specify required completion deadlines.
Confirm the Credit estimate and submit.
Workflow: The job is routed into Straker’s global production network managed through the Managr console. Once certified, final files return automatically to your deliverables tab.
Comparing Review Options
Factor
Review Myself
Assign to Team
Send to Straker
Linguist
You
Internal colleague
Certified Straker linguists
Setup Time
Instant (0 sec)
30 seconds
1 minute
Delivery Speed
At your pace
Dependent on team
Guaranteed SLA date
DTP Formatting
Text-only
Text-only
Full page-by-page layout
Billing
Zero credits
Zero credits
Credits deducted
Cortex Learning
Captured
Captured
Captured
The Review Workspace (CAT)
The Review Workspace (/projects/<id>/review/<batchId>) is Arbitr’s professional computer-assisted translation (CAT) environment. Designed for speed, precision, and focus, it expands full-width across your browser, minimizing distractions while putting translation memory, terminology, and agent diagnostics at your fingertips.
Workspace Layout
The workspace is organized into three primary sections:
Header Bar:
Project name, document name, and target language pair.
Progress meter showing percentage of verified segments.
Batch status and urgency badge.
Complete & Sign Off button.
Segment Editor (Main Grid):
Left Column: Source segment text, protected tags, and identified glossary terms.
Right Column: Editable target translation, segment status, and agent findings.
Inspector Sidebar (Right):
Translation Memory: Concordance matches from Cortex with similarity percentages (100%, 95%, 80%).
Terminology: Approved definitions and forbidden terms detected in the active segment.
Agent Diagnostics: Explanations for any automated corrections or flagged anomalies.
Navigating and Editing Segments
Click to Edit: Click directly on any target segment text to activate the inline rich-text editor.
Keyboard Shortcuts:
Ctrl + Enter (or Cmd + Enter on macOS): Confirm segment, save changes, and advance to the next unverified segment.
Ctrl + Down / Ctrl + Up: Move between segments without modifying status.
Ctrl + 1, Ctrl + 2: Insert top translation memory or glossary suggestion directly into the target segment.
Auto-Save: Changes are saved automatically as you type. A green Saved badge confirms synchronization with the server.
Segment Statuses
Each segment row displays an indicator reflecting its verification state:
Unreviewed / Machine Translated (Grey): Raw machine translation or agent repair not yet inspected by a human.
Flagged (Amber / Red): An agent identified a potential quality issue requiring attention (e.g., missing approved term, number discrepancy).
Approved / Verified (Green): The linguist has confirmed the translation is accurate and ready for publishing.
TM Exact (Blue): Pre-populated from Cortex 100% or ICE translation memory.
Handling Protected Tags & Formatting
When translating structured documents (Word, HTML, InDesign), bolding, hyperlinks, and layout codes appear as protected inline tags (e.g., <1>, </1>):
Never delete or alter tag numbers unless intentionally removing the corresponding styling.
Keep tag pairs around the exact equivalent words in the target language.
If a tag is accidentally deleted, click Re-insert Tags in the segment toolbar to restore the original tag hierarchy.
Completing & Signing Off
Once all segments have been reviewed and verified:
Check the progress meter in the header to confirm 100% Complete.
Click Complete & Sign Off.
A confirmation dialog summarizes the session:
Total segments reviewed.
Number of manual edits made.
Proposed contributions ready for ingestion into Cortex.
Click Confirm Sign-Off. The batch status updates to Submitted, the deliverable documents are regenerated with your latest edits, and status transitions to Published.
The My Reviews queue (/reviews) is the dedicated task list for human reviewers. Internal linguists, regional reviewers, and bilingual team members use this inbox to discover assigned translation batches, monitor deadlines, and launch the review workspace.
The Reviewer Experience
For users assigned the Reviewer role, /reviews serves as their primary landing page. To minimize operational friction, all administrative tabs (such as billing and organization settings) are hidden, presenting a clean, task-oriented workspace.
Administrators and Project Owners can also access /reviews to oversee all pending review batches across the entire organization.
Queue Columns & Metadata
Each row in the queue represents a specific document batch assigned for linguistic review:
File: The name of the source document and campaign reference.
Locale: The source and target language pair (e.g., EN-US → DE-DE).
Reviewer: The assigned internal team member (or Unassigned).
Urgency:Normal or Urgent (highlighted in red for accelerated deadlines).
Due Date: Target delivery deadline. Overdue items display a warning icon.
Status:
Available for review: Ready to be picked up by any authorized linguist.
In progress: A reviewer is actively working on the batch.
Submitted: The review has been signed off and completed.
Filtering and Searching Reviews
Above the queue table, quick filters allow you to prioritize your workload:
Search Bar: Search by file name, language code, due date, or reviewer name.
Status Filter: Toggle between All statuses, In progress, Available, and Submitted.
Urgency Filter: Filter by Urgent to address high-priority items first.
Opening a Batch
To begin working on an assigned task:
Locate the batch in your queue.
Click anywhere on the table row or click Open Review.
Arbitr loads the document into the full-screen Review Workspace.
Once you submit and sign off, the batch is automatically removed from your active queue and marked Submitted.
Sending to Straker Linguists & DTP
When projects require professional certification, second-linguist editorial review, or complex desktop publishing (DTP), you can dispatch batches directly to Straker Professional Linguists from the Send to Straker screen (/projects/<id>/send-to-straker).
Service Tiers
When configuring an order for Straker services, choose the service tier that matches your quality and compliance requirements:
1. Professional Translation
Scope: A certified native specialist linguist translates, post-edits, and verifies every segment against your Cortex terminology and style rules.
Best For: High-volume marketing copy, internal communications, training manuals, and product catalogs.
2. Translation + Edit (TEP)
Scope: Four-eyes principle. A primary specialist translates the content, and an independent senior editor reviews, cross-checks, and signs off on every sentence.
Best For: Regulated medical documentation, legal contracts, investor disclosures, and high-impact brand campaigns.
3. Desktop Publishing (DTP)
Scope: Graphic specialists rebuild your document layouts page-by-page to resolve text expansion, font reflow, hyphenation breaks, and image alignment.
Supported Formats: Adobe InDesign (.idml), Adobe PDF (.pdf), Microsoft PowerPoint (.pptx), and Word (.docx).
Right-to-Left (RTL) Layouts: For languages written right-to-left (such as Arabic and Hebrew), layouts must be mirrored. A standard 2× page rate applies to account for bidirectional restructuring.
4. Linguistic Sign-Off (LSO)
Scope: A native linguist conducts a final proofreading pass after the desktop publishing team completes layout reconstruction, verifying that no headers were truncated or captions displaced.
Volume & Pricing Calculation
On the right side of the screen, the order estimate panel calculates costs in real-time:
Characters & Words: Gross count of translatable text across the selected batches.
Memory Match Policy:
By default, human translation orders cover the full context of the document.
If you select batches that are already 100% matched by Cortex TM, a confirmation modal explains that human review of pre-translated segments is billed at the full character count.
Repetition Discounts: Repeated segments inside the order are billed at a discounted 25% rate.
Required Credits: The exact number of credits deducted upon order placement.
Setting Deadlines & Delivery Dates
Delivery Date Selector: Choose your desired completion date.
Earliest Due Estimate: Arbitr calculates the minimum required turnaround time based on volume, service tier, and linguist capacity (e.g., Earliest delivery about Thursday, 3:00 PM).
Weekend Auto-Adjustment: If your requested delivery date lands on a weekend or local public holiday, the platform automatically adjusts the target date to the next business day and displays a notice.
Note to Reviewers: Add optional guidance for the assigned linguists regarding brand tone, preferred terminology, or background context.
Tracking Order Progress
Once you click Send to Straker:
The required Credits are deducted from your balance.
The project is routed into Straker’s global production operations managed through the Managr console.
In your project workspace, the batches display a purple Sent · [Reference] badge.
When Straker linguists complete and certify the work, the finished deliverables return automatically to your project’s Deliverables tab, and verified segment updates are contributed back into Cortex.
The Translation Memory (TM) Explorer (/cortex/memory) gives you direct visibility into the bilingual sentence pairs stored in your organization’s memory. You can search historical translations, verify consistency, and inspect match performance.
Concordance Search
Use the concordance search bar to look up how specific phrases or sentences were translated in past projects:
Query Text: Enter a word or sentence fragment in either the source or target language.
Language Filter: Select the desired source and target language pair.
Domain Filter: Restrict the search to a specific domain context.
Results: The grid displays all historical bilingual segments containing your search terms, alongside metadata showing when the segment was approved, the originating project reference, and the author (e.g., Straker Certified Linguist or Internal Team Reviewer).
Segment Metadata & Match Types
Each translation memory segment records critical provenance data:
Source Segment: The original sentence.
Target Translation: The approved localized sentence.
Match Level:
ICE (101%): In-Context Exact match with identical surrounding sentences.
100% Exact: Identical sentence string.
Fuzzy (70%–99%): High-similarity segment adapted by the Auto Post-Editor.
Usage Count: How many times this segment has been leveraged across subsequent projects.
Last Verified: Timestamp of the latest human verification.
Deleting or Correcting Stale Segments
If a historical translation memory segment contains an error or outdated product reference:
Locate the segment using Concordance Search.
Click Edit Segment to correct the target text in place.
Click Delete Segment to permanently purge it from memory, preventing future projects from referencing the erroneous translation.
SwiftBridge (/swiftbridge) is Arbitr’s integration and automation engine for continuous localization. It eliminates manual file uploads by connecting your content repositories, content management systems (CMS), and codebases directly into Arbitr’s translation and review pipeline.
What SwiftBridge Does
Automated Content Ingestion: Listens for webhooks, repository pull requests, or CMS status changes to ingest new or modified content automatically.
Continuous Localization: Feeds small, frequent content increments (such as release notes, UI string changes, or blog updates) through your Cortex memory and agent ensembles without human coordination.
Automated Delivery: Pushes verified, published deliverables directly back into the originating repository, CMS, or cloud bucket once quality thresholds are satisfied.
Cross-Platform Glossary Sync (/swiftbridge/glossary): Keeps glossaries synchronized between Arbitr and external termbase systems.
Creating a SwiftBridge Pipeline
In the account dropdown or dashboard teaser, open SwiftBridge and click New Pipeline (/swiftbridge/new).
Connector Type: Choose your source system (e.g., GitHub, GitLab, WordPress, Contentful, Zendesk, or Amazon S3).
Authentication: Authorize connection using API keys or OAuth credentials.
Trigger Rules: Specify when translation should trigger (e.g., When a markdown file in /content/en/ is merged to main).
Campaign Defaults: Assign the default source language, target languages, domain, and agent ensemble.
Delivery Action: Configure the destination branch, folder, or publishing state.
Monitoring Pipeline Health
The SwiftBridge dashboard displays:
Active Pipelines: Status, last sync timestamp, and total jobs dispatched.
Sync History: Log of past document transfers, payload sizes, and delivery receipts.
Error Alerts: Notifications if an external API token expires or if a repository branch is unreachable.
AI Visibility Guide coming soon
AI Visibility is live in Arbitr. It shows how AI assistants describe and cite your brand, across the languages you publish in.
We’re preparing a step-by-step guide covering setup, reading your results and acting on what you find. In the meantime, you can open AI Visibility from the Arbitr app.
The Organization Settings (/settings/organization) and Content Preferences (/settings/content-preferences) screens allow administrators to manage your company identity and establish platform-wide defaults for translation campaigns.
Configure the quality gate for automated delivery:
Slider (0% to 100%): Set the minimum Trust Score required for files to publish automatically (e.g., 85%).
Toggle Off: Disables automated publishing entirely, forcing all campaigns to require human reviewer approval.
3. Continuous Learning Automation
Auto-Commit Certified Edits: When enabled, post-edits authored by Straker Certified Linguists are committed directly into your Cortex translation memory without requiring administrator sign-off.
Reviewer Contribution Moderation: When enabled, edits made by internal team members are staged in the Contributions Queue for administrator approval.
The Team Management screen (/settings/team) is where administrators invite colleagues, assign role-based permissions, configure language specializations, and manage user access.
The Four User Roles
Arbitr enforces strict role-based access control (RBAC) across all platform surfaces:
Role
Core responsibilities
Gated permissions
Owner
Full organization ownership
Billing, plan management, company deletion, all admin capabilities.
Admin
Daily operations & asset management
Create projects, manage Cortex assets, invite team members, configure integrations.
Analyst
Governance & executive oversight
Read-only access to Analytics, Governance audit trail, and project reports.
Reviewer
In-country & linguistic review
Access restricted to the Review Queue (/reviews) and CAT workspace (/projects/<id>/review/<batchId>).
Reviewer Isolation
Users with the Reviewer role cannot access administrative settings, team management, or billing. Their global navigation collapses to show only their active review tasks.
Inviting a New Team Member
In the top right corner of the Team screen, click Invite Member.
Email Address: Enter the colleague’s corporate email.
Assigned Role: Select Owner, Admin, Analyst, or Reviewer.
Language Pairs (for Reviewers):
If assigning the Reviewer role, select the specific source and target languages they are qualified to review (e.g., English → French (Canada)).
This metadata enables Arbitr to recommend this reviewer when assigning batches during the review workflow.
Click Send Invitation. An email containing an onboarding link (/invite/<token>) is dispatched immediately.
Managing Active Members
Change Role: Click the role dropdown on any active member row to upgrade or downgrade their permission level.
Update Languages: Edit a reviewer’s language pairs to reflect new language capabilities.
Resend Invite: For pending invitations, click Resend to send a fresh activation link.
Revoke Access / Remove: Click the trash icon to revoke a user’s access. The user’s active sessions are terminated immediately.
The Billing & Credits screen (/settings/billing) allows Organization Owners and Admins to manage Credit balances, purchase packages, review transaction history, and download corporate invoices.
Current Balance & Usage Overview
At the top of the billing screen:
Available Credits: Your organization’s current active Credit balance.
Estimated Run Rate: Projected monthly credit consumption based on your last 30 days of campaign activity.
Monthly Usage Chart: Historical graph showing credit expenditure broken down by automated translation, agent operations, and professional Straker services.
Purchasing Credits
Click Add Credits or Purchase Package.
Select a credit tier (e.g., Starter 5,000 IC, Scale 25,000 IC, Enterprise 100,000 IC). Volume discounts are automatically applied to higher tiers.
Stripe Checkout:
You are redirected to a secure Stripe payment portal.
Enter your corporate credit card or corporate purchasing details.
The Security Settings screen (/settings/security) allows administrators to enforce enterprise identity standards, mandate multi-factor authentication (MFA), configure session lifetimes, and monitor active user logins.
Single Sign-On (SSO) Configuration
Arbitr supports SAML 2.0 and OpenID Connect (OIDC) integrations with enterprise identity providers including Okta, Microsoft Entra ID (Azure AD), Google Workspace, and Ping Identity:
Enable SSO: Toggle on Enterprise Single Sign-On.
Metadata Exchange:
Copy Arbitr’s Assertion Consumer Service (ACS) URL and Entity ID / Audience URI into your identity provider.
Enter your identity provider’s SAML 2.0 Endpoint (SSO URL) and upload your public X.509 Certificate.
Enforce SSO Login: Check Enforce SSO Only to disable standard email/password authentication for all users with your corporate email domain.
Multi-Factor Authentication (MFA) Policy
Optional (Default): Users can independently link authenticator apps (TOTP) from their personal profile.
Enforced: Mandates that all team members configure an authenticator app (Google Authenticator, Microsoft Authenticator, 1Password) upon their next login. Users cannot access any workspace features until MFA setup is verified.
Session Timeout & Idle Policies
Idle Session Timeout: Specify the duration of inactivity (e.g., 15 minutes, 1 hour, 8 hours) before an active session is automatically locked.
Absolute Session Lifetime: Define the maximum duration before users are required to re-authenticate completely.
Active Sessions & Device Management
At the bottom of the security screen, the Active Sessions table displays:
Device & Browser: OS and browser user-agent.
IP Address & Location: Approximate geographical origin.
Last Active: Timestamp of latest API request.
Revoke Session: Click Revoke on any row to terminate that specific session immediately. Click Revoke All Other Sessions to force re-authentication across all active devices.
The API Keys & Webhooks screen (/settings/api-keys) enables developers and engineering teams to integrate Arbitr’s translation pipeline, Cortex knowledge base, and governance auditing into automated CI/CD and enterprise software systems.
Managing API Keys
API keys provide programmatic access to the Arbitr REST API. Requests must include the key in the HTTP Authorization header:
Key Name: Enter a descriptive label (e.g., CI/CD Localization Pipeline or Zendesk Sync).
Permissions / Scopes:
Projects: Write: Upload documents and launch translation campaigns.
Projects: Read: Retrieve project status, trust scores, and deliverable files.
Cortex: Read/Write: Query translation memory and update glossaries.
Governance: Read: Access audit logs.
Expiration: Choose an expiry window (30 days, 90 days, 1 year, or Never).
Click Generate Key.
Secure Your Secret Key
The full secret key is displayed only once upon creation. Store it in a secure password manager or environment variable vault. If lost, the key must be revoked and regenerated.
Revoking an API Key
To immediately invalidate an active key:
Locate the key in your API Keys table.
Click the trash icon or Revoke.
Confirm revocation. All subsequent requests authenticated with that key are rejected immediately with HTTP 401 Unauthorized.
Webhooks
Webhooks notify your external services in real-time when key lifecycle events occur in Arbitr.
Adding a Webhook Endpoint
In the Webhooks section, click Add Endpoint.
Endpoint URL: Enter your secure HTTPS listener URL (e.g., https://api.yourcompany.com/webhooks/arbitr).
Subscribed Events:
project.processing.completed: Triggered when translation finishes.
project.published: Triggered when deliverables are compiled and ready.
project.review_required: Triggered when trust score falls below threshold.
batch.submitted: Triggered when a human reviewer signs off on a batch.
Signing Secret: Copy the generated webhook secret to verify payload HMAC SHA-256 signatures on your server.
The Integrations Screen (/integrations) allows administrators to connect Arbitr to corporate cloud storage providers, content management systems, and productivity suites.
Supported Integration Categories
Cloud Storage:
Google Drive, Microsoft OneDrive / SharePoint, Dropbox, Box.
Allows users to select files directly from corporate storage during upload and save published deliverables back to shared drives.
Content Management Systems (CMS):
WordPress, Contentful, Strapi, Drupal.
Ingests draft articles and exports localized variants directly to CMS staging environments.
The Governance Dashboard (/governance) provides an immutable, tamper-evident audit trail of every automated decision, agent modification, human approval, and compliance certification executed across the Arbitr platform.
The Audit Trail Table
Every operational event in Arbitr generates an audit record:
Timestamp: Precise ISO timestamp of the action.
Project & File Reference: The specific document and campaign ID.
Event Category:
Published: Translation verified and released for distribution.
Approved: Human reviewer verified segments without edits.
Approved with Edits: Human reviewer modified segments prior to sign-off.
Escalated: Quality score fell below threshold; routed for human review.
Agent Repair: Automated intervention by a specialized agent (e.g., Glossary Agent).
Rejected: Translation rejected by reviewer and returned for reprocessing.
Actor: The entity responsible for the decision (e.g., System Auto-Publish, Glossary Agent, or an authenticated user name and email).
Target Locale: Language pair involved.
Filtering and Exporting Records
Enterprise compliance officers can filter and export audit data:
Filter by Decision: Isolate escalations or human edits across a specific date window.
Filter by Actor: Audit actions performed by specific team members or AI agents.
Export (CSV / JSON): Click Export Audit Log to download certified compliance logs for external auditing or SOC 2 reporting.
Enterprise Compliance Certifications
At the bottom of the Governance screen, Arbitr displays active organizational compliance certifications:
SOC 2 Type II: Certified controls governing security, availability, and confidentiality.
AES-256 Encryption: All documents, memory segments, and audit logs are encrypted in transit (TLS 1.3) and at rest (AES-256).
GDPR & Regional Processing: Guarantees that data processing occurs within your designated geographic boundary (e.g., EU-only processing nodes).
The Analytics Dashboard (/analytics) provides high-level business intelligence into your organization’s localization velocity, translation quality, escalation frequency, and financial savings.
Executive KPI Cards
Across the top of the analytics dashboard, four primary metric cards summarize overall performance:
Jobs Processed: Total campaigns and document batches completed within the selected timeframe, accompanied by a percentage delta compared to the prior period.
Average Quality Score: The aggregate Trust Score across all completed projects. Tracks whether translation accuracy is improving as your Cortex memory matures.
Escalation Rate: The percentage of jobs that fell below the auto-publish threshold and required human intervention:
Within Target (Green): Low escalation frequency; automated agents and TM are handling the vast majority of volume.
Arbitr is built to meet WCAG 2.1 Level AA accessibility standards. The Accessibility Settings screen (/settings/accessibility) provides personal display controls that allow each user to tailor typography, visual contrast, motion, and interaction targets to their preferences.
Accessibility Preferences
Changes configured here apply immediately to your active session and persist across all browser logins:
1. High Contrast Mode
Setting: Toggle on High Contrast.
Effect: Increases visual contrast between background surfaces and text elements, sharpens border hairlines, and enforces distinct focus rings around interactive buttons and form inputs.
2. Dyslexia-Friendly Typography
Setting: Toggle on Dyslexia Font.
Effect: Replaces the default UI body font with OpenDyslexic, featuring weighted bottoms and unique letterforms to reduce visual transposition and improve reading ease during long review sessions.
3. Text Size Adjustment
Setting: Choose between Default, Large, or Extra Large.
Effect: Scales interface typography, table rows, and editor segment sizes while maintaining layout responsiveness without horizontal scrollbars.
4. Underline Links
Setting: Toggle on Underline Links.
Effect: Forces persistent text underlines on all navigational and interactive text links, ensuring link affordance does not rely solely on color.
5. Reduced Motion
Setting: Toggle on Reduce Motion.
Effect: Suppresses interface animations, modal transitions, and progress bar pulsing in accordance with operating system prefers-reduced-motion guidelines.
Live Preview & Resetting Defaults
At the bottom of the accessibility screen, an Interactive Preview Card renders a sample paragraph, heading, and button reflecting your current selections in real time. If you ever wish to restore system defaults, click Reset to Defaults.
Platform Glossary
A comprehensive reference for terminology, acronyms, and concepts used throughout the Arbitr platform.
Arbitr
An AI context platform. Arbitr gives your organization one governed memory, Cortex, which Arbitr’s agents and checks use to create and review content that is accurate for every market and language.
AI Visibility
An Arbitr tool that shows how AI assistants describe and cite your brand across the languages you publish in.
Agent Ensemble
In localization campaigns, a curated group of specialized AI agents (e.g., Glossary Agent, Style & Tone Agent, Compliance Agent) deployed alongside the neural machine translation engine to validate and repair translated content in real-time.
Auto Post-Editor (APE)
A specialized AI agent that leverages fuzzy matches from Cortex translation memory to produce natural, fluent rewrites of machine-translated sentences without human post-editing.
Auto-Publish Threshold
The minimum Trust Score required for a translation project to automatically release downloadable deliverables without requiring human reviewer sign-off.
Batch
A discrete translation unit representing one source document paired with one specific target language. A campaign containing three files translated into four languages consists of twelve batches.
Campaign
In localization, a project containing one or more uploaded source documents, target languages, a designated domain profile, and an assigned agent ensemble.
CAT (Computer-Assisted Translation)
Software designed to assist human linguists during translation and editing. In Arbitr, the CAT interface is the in-browser Review Workspace.
Concordance Search
A feature within the Cortex Memory Explorer that searches historical bilingual sentence pairs for specific words or phrases to verify how terms were previously translated.
Cortex
Arbitr’s governed memory. It holds your organization’s approved knowledge — claims, terminology, rules and product knowledge — along with translation memories, glossaries and continuous learning contributions.
Desktop Publishing (DTP)
The process of restructuring and reformatting localized documents (InDesign, PDF, PowerPoint, Word) page-by-page to resolve text expansion, font reflow, and bidirectional layout requirements.
Domain
A categorical partition within Cortex (e.g., Legal, Financial, Marketing, Medical) that isolates specialized glossaries, tone calibrations, and translation memory segments for specific business contexts.
Do-Not-Translate (DNT)
A glossary designation for proprietary brand names, product identifiers, trademarks, or code variables that must remain in their original source form across all target languages.
Fuzzy Match
A segment in a new source document that closely resembles (70%–99% similarity) an existing segment in your translation memory, differing only by punctuation, numbers, or minor phrasing.
In-Context Exact (ICE / 101%) Match
A source segment that matches an existing translation memory segment identically and shares the exact same preceding and following sentences, guaranteeing contextual fidelity.
Linguistic Sign-Off (LSO)
A final proofreading pass conducted by a certified native linguist after desktop publishing (DTP) is completed, verifying that layout adjustments did not truncate text or disrupt meaning.
Automated translation generated by artificial intelligence and deep learning neural networks.
Repetition Discount
A pricing policy where duplicate sentences appearing multiple times within the same project receive a 75% character discount after the first occurrence.
Segment
A discrete grammatical unit (typically a single sentence, heading, or table cell) created when documents are parsed for translation.
SwiftBridge
Arbitr’s continuous localization engine that connects content repositories (GitHub, CMS platforms, cloud storage) directly to automated translation pipelines.
TEP (Translation, Edit, Proof)
An enterprise quality workflow adhering to the four-eyes principle, where a primary linguist translates the text and an independent senior editor reviews and verifies the output.
TMX (Translation Memory eXchange)
An industry-standard XML file format used for importing and exporting bilingual translation memory segments between localization platforms.
Translation Memory (TM)
A database of previously translated sentences stored as bilingual source-and-target pairs. 100% matches from TM are reused automatically at zero translation cost.
Credit (IC)
The unit of account used to measure and purchase translation and agent operations in Arbitr.
Trust Score
A composite quality metric (0% to 100%) calculated by Arbitr that evaluates glossary adherence, translation memory alignment, linguistic fluency, and unresolved agent findings.
Roles & Permissions
Arbitr uses a role-based permission system to ensure team members have the exact level of access necessary for their organizational responsibilities.
Under the hood, Arbitr enforces granular permission gates:
team.read: Authorizes access to the general administrative shell and team directory. When absent (as with the Reviewer role), the global navigation collapses to show only the review queue.
team.write: Authorizes inviting new users, updating roles, or revoking access.
audit.read: Authorizes access to the Governance (/governance) audit logs and Analytics (/analytics) dashboards. Held by Owners, Admins, and Analysts.
billing.manage: Restricted exclusively to Organization Owners. Authorizes purchasing credit packs, adding payment cards, and viewing tax invoices.
projects.manage: Authorizes creating new campaigns, changing ensemble agents, and placing orders for professional Straker linguist services.
FAQ & Troubleshooting
Answers to common questions and practical troubleshooting steps for Arbitr.
Projects & Processing
Why is my project stuck on “Review Required”?
A project enters Review Required if its calculated Trust Score falls below your organization’s configured auto-publish threshold, or if auto-publish is toggled off in Content Preferences. Click into the project and select Get Reviewed to verify flagged segments or order Straker review.
Why did my document extraction fail?
Document extraction failures typically stem from:
Password Protection: PDFs or Word documents secured with passwords cannot be parsed. Remove encryption before uploading.
Scanned Bitmaps: Scanned PDFs that contain flat images rather than selectable text layers cannot be segmented without OCR pre-processing.
Corrupted Markup: Malformed XML or HTML files containing unclosed tags.
Can I cancel a campaign while it is translating?
If processing is in the Queued stage, you can click Cancel in the operations room to cancel the job and refund all reserved credits. Once neural translation is actively in flight, processing cannot be cancelled.
Cortex & Terminology
How do I ensure my brand name is never translated?
Navigate to Cortex → Glossaries (/cortex/glossary), click Add Term, enter your brand name in both the source and target fields, and check Apply to all target languages. The Glossary Agent will enforce this as a Do-Not-Translate rule across all models.
Do human post-edits immediately update future translations?
Yes. Once an edit is committed in the Contributions Queue (or auto-committed if authored by Straker Certified Linguists), the updated translation unit is stored in your active translation memory. Future documents with identical sentences will automatically leverage the human-verified translation.
Review & Linguist Services
What is the turnaround time for Straker Professional Linguists?
Turnaround depends on volume, service tier (Translation vs. Translation + Edit), and whether desktop publishing (DTP) is included. When placing an order under /projects/<id>/send-to-straker, the platform calculates an exact guaranteed delivery deadline based on word count and linguist availability.
What happens if our delivery date lands on a weekend?
Arbitr automatically detects weekends and regional holidays. If an estimated turnaround finishes on a non-business day, the delivery date automatically rolls forward to the next business morning.
Accounts & Security
Why can’t our reviewers access the dashboard or billing?
Users assigned the Reviewer role are intentionally isolated from administrative, financial, and organizational settings. They land exclusively in their personal Reviewer Queue to focus on translation verification.
What should we do if an employee leaves the company?
An Administrator or Owner should navigate to Settings → Team (/settings/team) and click the trash icon next to the user’s name. This revokes their access and terminates all active browser sessions immediately. If SSO is enforced, deactivating the user in your corporate identity provider (Okta/Entra ID) will also prevent login.