Studio

Studio is WriterzRoom’s governed production workspace. It replaces prompt-first generation with a three-stage workflow that validates context, template structure, style behavior, and required inputs before execution.

​

Context

Each context states what it applies, its focus areas, and how many templates and styles it offers. Select General for unrestricted access, or an industry vertical to activate:

​

Workflow selection

The template defines the output contract: required fields, section order, structural rules, validation requirements, and agent instructions. The style profile defines voice, pacing, tone, complexity, and formatting behavior. The catalog is searchable and filterable by category. Selecting a workflow loads its full parameter schema and narrows the styles to those offered for it. Each style states why it is offered: recommended for this workflow, or compatible. The style list has three scopes: Template links from the catalog open Studio with the workflow preselected. Studio then loads the full template detail schema before allowing generation.

​

Production brief

The Production brief is generated from the selected template’s YAML schema. Field types can include text, long-form text, select, multiselect, checkbox, number, range, date, combobox, and file upload. Fields are grouped by purpose: Required first, then Shape the draft (optional choices that most change the result), Sources, and Advanced (collapsed, each field with a sensible default). Run settings hold the decisions that apply to the whole run: context, workflow, and style; document length; tier and its credit cost; and optional brand voice and audience. Length is one of the workflow’s three presets or a custom word range. A tier that cannot finish the chosen length within its time budget is refused before any credit is charged, with the reason and the tier that can. See Length decides the tier. Before you select Run workflow, WriterzRoom names any required inputs still missing.

​

Source files and coverage

Upload sources in the Production brief. Studio shows a short text preview and a separate extraction report: retained units, total units when known, worksheet coverage, unreadable pages and processing limits. The preview is not the full evidence payload. Original file bytes and extraction versions are saved privately to your account. The brief records the selected version. CSV and JSON retain records beyond the preview. Spreadsheets include visible and hidden sheets; Word extraction includes tables, nested tables, headers and footers. Spreadsheet formulas are retained as text, not calculated results. Charts, drawings, embedded objects and document notes may require a text export. Required source fields must contain a complete, readable upload. Selected files with incomplete extraction hold generation even when the field is optional. Remove or replace those files before continuing. Removing a file detaches it from the brief and clears its preview and coverage; it does not delete a stored extraction or earlier generation records. Each file is limited to 10MB and 80,000 retained extraction characters, with additional page, sheet and unit limits shown in its coverage report. There is one combined 60,000-character source context limit per generation, including source labels. If the sources exceed that limit, split the work or upload a smaller relevant selection. The system holds the request before charging rather than silently dropping source material. Earlier unsaved uploads must be uploaded again to use the stored-source contract.

​

Generation tiers

The selected tier controls workflow depth, timeout, model ceiling, and validation stages. It does not substitute a different template or style profile.

​

Input checks before a run

Studio validates the submission before starting the pipeline. If the backend refuses it with HTTP 422, the output area shows Input check and the field or admission reason. No workflow or draft was started, and no generation credit is charged for that rejection. Correct the brief and submit again. Accepted requests have a request ID and stage activity. A later execution failure is shown separately from an input rejection; keep its request ID when contacting support.

​

Agent activity

While a run is in progress, WriterzRoom reports the active stage, the step count, and elapsed time. Agent activity shows each stage as queued, running, or completed, with what it reported. Which stages run depends on the tier.

​

Output and readiness

The draft fills in as the writer produces it. If your connection is served by a different server than the one running the generation, the live preview is not available; Studio says so, keeps showing each stage’s activity, and shows the draft when the run finishes. A completed draft has copy, Markdown export and Review in library actions. After a completed run, the output section shows what was recorded for it:

• Quality gate — passed, or not recorded
• Verification — how many verifiers ran for this context, their coverage and any blocking findings, or that none is configured for the context; when verifiers ran, counts by verdict (verified, in violation, could not be checked, did not apply). “Could not be checked” and “did not apply” are never counted as verified
• Assurance — the level this run achieved, and whether it is below the level its tier declares

After a failed run, it shows the stage the run stopped in, the reason (time limit, too few sources, a check rejected the draft, citations could not be verified, or an internal error), and whether any check ran. A run that stopped before reaching a check shows its checks as not run, not failed. A missing value is shown as not recorded rather than as a pass or a score of zero. Planning and revision stages share evidence and prose instructions. Nonfiction scenes and quotations require supplied or attributable material; formatting and SEO preserve the edited voice. The final body is checked for specific stock scenes and narrator padding, with one directed revision before a surviving defect blocks completion. These checks do not prove authorship or factual accuracy.

​

Visual generation

Visual generation remains disabled in Studio. It will not be enabled until image quality, persistence, moderation, retries, and audit logging meet the same production standard as written output.

​

Responsive layout

On narrower screens, Run settings moves below the brief and the workflow detail moves below the catalog. On mobile, everything stacks vertically without removing any workflow controls.

​

Reference files

Templates that expose a file-upload field can parse PDF, DOCX, XLSX, CSV, JSON, TXT, and Markdown inputs. Original files and extraction versions are saved privately. Choose a saved version in a Studio file input, or open Sources to inspect locations, download originals, add versions and manage access. Briefs remain pinned to the selected version. In Sources, an owner can pin one version of a source as its default. Studio’s saved-source picker lists the pinned version first and labels it, and an API request that names the source without naming a version uses it; unpinned sources use the latest. It never moves evidence under work already generated — a brief that recorded an exact version keeps resolving to that version.

​

Saved evidence and access

Sources start private. A source owner can grant an active teammate read access or read-and-generation access for all versions, including future versions. Review assignment alone does not grant access. Reviews and the asset dossier can reopen an authorized source and its retained page, table or cell locations. Revoking a source holds further generation use and release. Revoking a grant or ending either workspace membership removes the recipient’s source access. This cannot recall downloads, completed provider calls or already published material. Older uploads without a retained original must be uploaded as a new version before reuse. Extraction coverage and location anchors do not themselves verify claims.

​

Dataset analysis

The Market Intelligence Report workflow requires an explicitly configured dataset analysis. Other templates retain their existing behavior. Choose an uploaded or saved CSV, JSON, or XLSX version, select a worksheet and header row where applicable, and review the complete column profile. Then choose columns, types, units, row ranges, and filters. Duplicate rows remain included; the profile shows missing and excluded rows. Add aggregates, endpoint growth, CAGR, ratios of sums, weighted scores, or rankings. Scoring requires explicit weights totaling one and explicit scales; ranking can use a calculated score. Choose an optional bar or line chart and Validate analysis. Editing a selection invalidates validation. Invalid or undefined calculations stop before paid generation; source access is checked again during the run. The result includes exact decimal values, source versions, formulas, source locations, missing-row exclusions, and charts derived from calculated metrics. Download the calculation evidence as JSON. Chart coordinates are visual approximations; the metric table and JSON preserve precision. Before publication, every metric must appear as an exact numeric reference. Unsupported numbers, altered references, or missing metrics trigger a directed repair and then stop publication if unresolved. This deliberately conservative check also rejects dates and recognized English cardinal and ordinal words outside metric references. The word detector is English-only. It establishes numeric reference integrity, not the truth of surrounding interpretations, causal claims, market coverage, or recommendations. Review those conclusions against the source data.