A column heading rarely tells the whole story. Owner can mean an assignee, a control owner, a process owner, or the organization that operates a system. Impact can mean severity, business impact, or a narrative description. Translation alone cannot solve that ambiguity. The voiqq importer combines language detection with the selected Program, Library, column values, and expected field contracts so it can preserve meaning instead of merely replacing words.
Detection happens before mapping
- The workbook reader extracts cached cell values without executing formulas and analyzes each sheet separately.
- The analyzer locates a likely regular header row and distinguishes structured project sheets from covers, dashboards, pivots, and notes pages.
- Language scoring checks professional header aliases and representative row text across the seven supported spreadsheet languages.
- The highest supported score becomes the default for that sheet, with a confidence value shown to the reviewer.
- The Program and Library then supply the valid canonical targets and required project settings.
- Deterministic matching compares normalized headers, synonyms, value shapes, and negative evidence before any optional AI review.
Signals used by the detector
Header signals receive the strongest weight because structured sheets repeat them consistently. The detector recognizes terms for summaries, descriptions, remediation, severity, workflow status, validation, assignments, dates, WCAG criteria, Trust Services Criteria, ASVS requirements, MASVS controls, AISVS requirements, evidence, affected components, environments, retest notes, and related fields. Representative cells provide supporting language words and domain shapes. A value such as CC6.1, 1.4.3, V7.3.1, or MASVS-STORAGE-1 does not identify a natural language by itself, but it strongly supports a specific requirement field after the Program is selected.
Every sheet keeps its own language
A multi-sheet workbook is not assigned one language globally. Each valid or possible project sheet is detected independently. The review screen displays its source name, row and column counts, detected header row, language, confidence, mappings, warnings, and proposed project name. If structured sheets belong to different Programs or Libraries, configure them separately. One selected valid sheet becomes one normal voiqq project, with the original sheet retained as a grid snapshot.
Deterministic mapping comes first
Clear columns are mapped without an AI request. The mapper uses Unicode-aware normalization, multilingual aliases, Program field definitions, option values, URL and date shapes, canonical requirement patterns, and contradictory evidence. It prevents known collisions such as Summary versus Actual Result, remediation versus description, workflow Status versus Validation, Evidence Status versus finding Status, Assignee versus Control Owner, and Page URL versus an evidence or documentation link. Source-specific columns remain additional data rather than being forced into an unrelated standard field.
AI is limited to genuine ambiguity
If a column remains uncertain, the server may send only its header, bounded examples, value profile, Program and Library context, and the allowed canonical targets to the existing AI fallback chain. The model must select one supplied target at high confidence or preserve the column. It cannot rewrite source values, create evidence, invent a field, choose the Program, or make a compliance conclusion. If AI is unavailable, rate limited, outside the allowance, or uncertain, deterministic mappings remain and the unresolved column is preserved for manual review.
Paid access is enforced on the source snapshot
Free workspaces can analyze and import English sheets. Pro and higher plans can import all seven languages. The final server transaction does not trust a browser field alone: it sanitizes the original grid, detects the language again from the stored headers and sample rows, resolves current organization entitlements, and rejects an unauthorized multilingual import before creating projects. The same boundary protects AI-assisted column matching. This also means a downgrade takes effect when the next import begins without changing existing project data.
When to override the detected language
- A workbook uses mostly English headings but the actual findings are written in another supported language.
- A client uses internal abbreviations that resemble terms from a different language.
- The detected header row is correct but contains too few meaningful words for high confidence.
- A sheet deliberately mixes languages, and one language best describes the workflow labels expected in its export.
- A fluent reviewer has checked the source and can explain why the automatic result is wrong.