From claiming a task to submitting it for review, from checking pre-check results to verifying reward details, everything is done in the browser with no client installation. The workspace keeps annotation specifications, tools and progress in one place, so you can focus on the judgement itself.
The workspace avoids 'feature stacking': each area solves the problems of one stage only, reducing irrelevant noise.
Filter tasks by data modality, difficulty level, unit price and remaining slots; each task shows its acceptance criteria, estimated hours and pricing method before you claim it, so you can review an example before deciding.
Imaging outlining, waveform segments and text structuring share one set of interaction habits: shortcuts, layer management, undo/redo and annotation notes. Progress is saved automatically and can be completed in stages.
Submitting triggers an automatic pre-check that flags suspected errors for you to fix; review verdicts and rejection reasons are listed item by item, and recurring mistakes become personal tips for you.
Valid workload, pass rate, level progress and settlement details are shown together. Every reward can be traced back to a specific task number and review verdict.
Task cards state the modality, sample size, acceptance criteria, estimated hours, pricing method and required level up front —— no need to claim a task first to learn the details.
* The table above illustrates the interface structure to show how task information is presented; actual task types, unit prices and batches are subject to what is published in the workspace.
Tools switch automatically with the task type, and specification prompts sit beside the annotation area, so you never flip back and forth through documents.
Polygons, freehand drawing and smart snapping, with multi-frame interpolation and structure copying to reduce repetitive work.
Diameter lines, angles, area and volume measurements, with values filled automatically into structured fields to avoid manual transcription.
Enumerated options and illustrated notes follow the specification; verdicts are linked automatically to specification clauses for easy review.
A millisecond-level timeline, synchronised multi-channel zoom and alignment cursors make boundary positioning more accurate.
Drag to create time intervals, place event points with one click, and copy similar events in bulk.
Waveforms are shown side by side with medical record fields and treatment parameters, making it easier to judge causation and temporal correspondence.
Selecting text annotates an entity, and dragging establishes relations; semantic qualifiers such as negation and suspicion are expressed with markers.
Fill in each item following the structured template; missing fields are explicitly marked as 'not mentioned' rather than left blank.
Built-in terminology sets and search suggestions normalise terminology as you annotate, reducing later cleaning costs.
Multiple annotations of the same task are compared automatically, with differences highlighted for quick decisions.
Disputed samples automatically enter the dispute pool, where a higher-level reviewer gives a verdict that is captured as a rule patch.
Pass rate, distribution of rejection reasons and common error types are visualised, helping you improve continuously rather than repeating the same mistakes.
The three stages divide the work as follows: 'machines detect errors, peers judge agreement, experts settle conclusions'; only annotations that pass all three count as valid workload.
A rule engine validates field completeness, outlining range, time boundaries and logical consistency, flagging suspected issues before submission.
Blind review by participants of the same or a higher level; annotations pass when agreement meets the threshold, and disagreements enter the dispute pool.
Specialist and high-risk tasks undergo final review by clinicians, with full re-check at the specified proportion, forming the final quality control conclusion.
Pricing rules, settlement cadence and dispute handling are open and transparent, so you never find out the unit price only after finishing the work.
Yes. Tasks that have not started or are only partially complete can be returned voluntarily; the unsubmitted part is not counted as workload and does not affect your pass rate. To avoid holding batch slots, the system automatically reclaims tasks not submitted within the agreed time limit.
Priced on the corrected result. If the issue is a misunderstanding, the platform provides the specific specification clauses and examples; if you disagree with the conclusion, you may file an appeal within the specified period, to be re-judged by a dispute pool reviewer.
Text structuring and speech transcription tasks can be done in a tablet browser; imaging outlining and waveform segments demand precise operation, so a desktop browser with a mouse is recommended for better accuracy and efficiency.
Levels are determined by long-term quality performance and can rise or fall. If a number of annotations needing correction occur in succession, your level may be lowered and the range of tasks you can take on adjusted accordingly; once you meet the standard again, you can be promoted again.
No. Task data is masked and displayed in a controlled environment; screenshots, exports, transcription, distribution or use for any other purpose are prohibited. The platform logs all operations, and any breach of the confidentiality commitment will result in account termination and legal liability.
Once identity verification is complete you can enter the Task Hall. If you would like to understand the platform's overall mechanism and participant roles first, return to the Crowd Platform section page.
