Time Tracking

This glossary explains the key metrics, charts, and concepts used in the Time Tracking dashboard.

Summary Cards

KPI	Description	How it's measured
Total Time	The total estimated working time across all selected workers and modules in the selected date range.	Sum of all session durations. A session is a sequence of consecutive events where each gap is ≤ the session gap threshold (default: 5 min).
Active Workers	Number of distinct users who logged at least one event in the selected period.	Count of unique worker names found in the event data.
Avg. Time/Worker	Average total time per active worker.	`Total Time ÷ Active Workers`
Taiga Closed	Taiga issues closed in the period, credited to their assignee. Archived issues are shown beside the number, never inside it.	Issues whose status is flagged is_closed with a `finished_date` in the window. Credit follows the current assignee, so reassigning after closing moves it; reopening clears the date, so only the latest close counts.
Regions Annotated	Regions a person drew in the Image Annotator. The subtitle adds the two other things they did to the model's pre-annotation: regions deleted and corrections made.	One event per region created (`add_region` + `sam_segmentation`, which are disjoint). Deletions come from `user_data.del_region_count`, so they are regions rather than clicks. Corrections are `change_region_attribute` + `resize_region` — edits to regions the batch already had, i.e. YOLO's. Never summed together: they measure different things.
Time/Image	Average annotation time per image. Annotation only — Image QS time is excluded even when the QS module is selected, because a QS pass over a batch is a different activity from annotating it.	Image Annotator session seconds ÷ the images in the batches touched. A batch counts once however many people worked on it. Time comes from this counter's own fetch, never from the worker row (which may also hold cleaning time).
QS Cleaned	Regions removed by an Image QS save; reclassifications are shown beside it.	Sum of `user_data.count` on `qs_cleaning`, logged on save success — so it counts regions that really went, not clicks. Shows an em dash rather than 0 for a window before that event existed, because there is no backfill.
Items Cleaned	Placements a Cleaner deleted or renamed. Undeletes are reported beside the number, not inside it.	Sum of `user_data.count` on `log_cleaning` (logged on success), not a count of clicks — one click removes anywhere from one placement to several thousand. Bulk filter actions record no count at all, so the real volume is this number plus an unknown amount; they are surfaced as uncounted actions.
Activity flags	Suspicious activity — ≥ 600 events (Cleaner: ≥ 400) of one worker within 30 s either side of an event. No human clicks that fast for that long. The time is excluded from every hour figure on this page, not deleted from the log; workers with less than 5 minutes excluded are not badged. Low activity — fewer than 600 events per credited hour in the annotator (Cleaner 350, Image QS 100) over at least 2 hours. Slow is not fraud: nothing is subtracted, the row is marked so someone can look. Idle share = the part of the credited time in which nobody clicked for two minutes or more.	Left: excluded hours summed over the selected modules (`excluded_seconds`; Total Time already has them subtracted and says +X h excluded; under a customer filter a worker's whole exclusion counts once any of their jobs is that customer's — exclusions belong to a worker, not a job). Right: workers flagged low activity in any selected module, computed from the hours table itself. Click the card for one row per worker and module.

Sessions & Time Calculation

What is a session?

A session represents a continuous period of activity for a single worker in a single module. Events are sorted chronologically, and consecutive events with gaps smaller than the session gap threshold are merged into one session.

Session Gap Threshold: 5 minutes (configurable via server config SHIKENSO_TIME_TRACKING_SESSION_GAP_SECONDS).

Session Duration = timestamp of last event in session − timestamp of first event in session.

Example: If a worker triggers events at 09:00, 09:03, 09:07, and then the next event is at 09:25, the first three events form one session (7 min), and the 09:25 event starts a new session.

All dates and clock times on this page are UTC, so that every surface agrees and a number can be reproduced from the date range alone. A day runs 00:00–24:00 UTC, which means work done before 02:00 in Germany (08:00 in Manila) counts towards the previous day. Other Jobs Manager pages still show your local time, so the same event can carry two different clock readings.

Charts

Chart	Description
Time by Worker	Doughnut chart showing the top 32 workers by total time. Remaining workers are grouped as "Others". Click a segment to toggle that worker as a filter.
Daily Trend	Shows time-per-day over the selected range. Display modes: Stacked by Worker – Area chart with one layer per worker (top 10 or selected) Total Hours – Single line showing aggregate hours per day Bar Chart – One bar per day for total hours Combined – Bar for total daily hours + line per worker overlay
Time by Module	Horizontal bar chart showing hours per module (e.g. Image Annotator, Image QS, Cleaner). Stacked by job category when available (Social Media / Newspaper / Other).
Time by Customer	All customers by total time. Can be stacked by category or by module. Shows which customers consume the most annotation/QS effort.

Modules

Module	Description
Image Annotator	Time spent creating and editing bounding-box annotations on images for YOLO model training.
Image QS	Time spent on quality assurance (QS) of image annotations – reviewing and correcting annotated data.
Cleaner (Chat)	Time spent in the Chat Cleaners (ChatQS + ChatQSBot), reviewing and cleaning chat message detections and bot-detection.
Cleaner (Video)	Time spent in the Video Cleaners (VideoQS + VideoQSFinal), cleaning up and performing final review of video detection results.
Cleaner (VoiceQS)	Time spent in the Voice QS Cleaner, reviewing and cleaning voice analysis results.
Customer Dashboard (Prod)	Time spent on customer-facing setup tasks on production (shikenso.com).
Customer Dashboard (Dev)	Time spent on customer-facing setup tasks on the dev environment.
Jobs Manager	Time spent managing jobs (creating, archiving, prioritizing, uploading).

Job Categories

Category	Description
Social Media	Jobs whose name starts with `social_media_`. Typically social media content analysis.
Newspaper	Jobs whose name starts with `newspaper_` or `newspapers_`. Print/online media analysis.
Other	All remaining jobs (typically livestream/VoD analysis).

Worker Details Table

Column	Description
Worker	The username of the person who performed the actions.
Total Time	Sum of all session durations for this worker in the selected period and filters.
Sessions	Number of distinct sessions detected for this worker.
Events	Raw count of individual tracked events (clicks, saves, page loads, etc.).
Δ Change	Only with comparison enabled: this worker’s time difference and percentage change against the previous period.
Taiga Closed	Taiga issues closed in the period, credited to the current assignee. See the Summary Cards table above.
Regions	Regions drawn in the Image Annotator, with regions deleted shown after it in grey. Hover for the full split: drawn, via SAM, deleted, relabelled, resized, and arrow-key nudges. A worker with 0 regions and a large deletion count did QC work, not nothing. Arrow-key nudges are a box shifted with the arrow keys, logged once per keypress — shown for context but not scored, since counting them would rank a keyboard user an order of magnitude above a mouse user for the same work.
Time/Image	This worker’s annotation seconds per image. Blank when none of their batches has a known image count.
QS Cleaned	Regions removed by their Image QS saves, with reclassifications after it in grey.
Cleaned	Placements they deleted or renamed in a Cleaner. Hover for undeletes and any bulk actions whose volume the log does not record.
Excluded	Auto-clicker time subtracted from this worker’s Total Time. ⚠ from 5 minutes up (hover: bot minutes, peak events per minute, last seen); slow when the worker is low activity in a selected module (hover: events per hour against the module’s floor, and idle share). See Activity flags above.
Time Distribution	Visual bar showing this worker's share of total time as a percentage.
Timeline	Opens a detailed chronological view of the worker's sessions and events, grouped by day.

Period-over-Period Comparison

Enable the Compare with previous period checkbox to overlay a previous time range on all charts, KPI cards, and worker tables.

Auto mode (default): The comparison range is computed automatically as the period of equal length immediately before your selected start date.

Manual mode: Click the Manual button to pick custom comparison dates.

Element	How comparison is shown
Summary Cards	A +12.3% or -5.2% badge appears below each KPI value.
Daily Trend	A dashed line/bar overlay shows the previous period’s values aligned day-by-day.
Module & Customer Bars	Translucent dashed-border bars appear next to the current period bars.
Workers Table	A “Δ Change” column shows the time difference and percentage change per worker. Workers present only in the previous period appear in a separate section at the bottom.
CSV Exports	When comparison is active, exported CSVs include previous-period values and delta columns.

Filters

Filter	Description
Date Range	Start and end dates, interpreted as UTC calendar days — so are the preset buttons (This Week, Last Week, This Month, Last Month). Only events within this range are included.
Modules	Filter by one or more modules. Selecting a single module shows a detailed view with daily trends and per-job breakdowns. Selecting none shows a summary across all modules.
Workers	Filter to specific workers. Supports free-text tags and multi-select. Leave empty for all workers.
Customers	Filter by GPM customer name. Only events associated with jobs belonging to these customers are included.

Tracked Events by Module

Every user interaction listed below is recorded as an event and sent to the central logger. Time tracking works by grouping consecutive events from the same worker into sessions. The event names below are the raw values stored in the database.

Image Annotator `image_annotator`

Event	Description
`open`	User opens a job in the Image Annotator. Logged once per job load.
`save`	User saves annotations for the current batch. Triggers data write to MongoDB via Jobs Manager.
`add_region`	A new bounding-box or polygon region is drawn on the current image.
`resize_region`	An existing region is resized (dragging a handle).
`move_region`	An existing region is nudged with the arrow keys — once per keypress, so shifting one box ten pixels logs ten events. Shown in the Regions tooltip for context but deliberately excluded from the correction count, which would otherwise rank a keyboard user an order of magnitude above a mouse user for the same work.
`del_sel_regions`	One or more selected regions are deleted.
`change_region_attribute`	A region’s label or other attribute is changed (e.g. assigning a class name).
`sam_segmentation`	The SAM AI segmentation model is invoked to auto-generate polygon masks from bounding boxes.
`jump_to_image`	User navigates directly to a specific image by index or thumbnail click.
`move_to_prev_image`	User navigates to the previous image in the batch.
`move_to_next_image`	User navigates to the next image in the batch.
`auto_label_suggested`	The auto-label AI suggested a label for a new region based on nearby existing annotations.
`auto_label_no_match`	The auto-label AI could not find a matching label for the region (no nearby annotations or low confidence).
`auto_label_corrected`	User manually corrected or overrode an auto-label suggestion by changing the region’s attribute. Logged from inside `change_region_attribute`, so it is a strict subset of it — adding the two together double-counts. It concerns the embedding label suggester, not the YOLO pre-annotation, which is why the Regions counter does not use it.

Image QS `image_qs`

Event	Description
`open`	User opens a QS (quality sample) job for review.
`save`	User saves QS review changes. The regions marked for removal are applied here.
`delete`	User toggles a region’s removal marker — and it fires again when they un-toggle it. It carries no count and the removal itself happens on save, so this event does not measure deletions and is deliberately not read by the QS Cleaned counter.
`qs_cleaning`	Logged on save success, once per type, with the number of regions that save really removed (`type: delete`) or reclassified (`type: rename`) in `user_data.count`. This is what QS Cleaned counts.

Cleaner `cleaner`

Includes Video Cleaners (VideoQS + VideoQSFinal), Chat Cleaners (ChatQS + ChatQSBot), VoiceQS Cleaner, and Cluster Cleaner variants. All share the cleaner module name. In the dashboard, Chat and Video variants are merged into Cleaner (Chat) and Cleaner (Video).

Event	Description
`next_asset`	User navigates to the next asset (brand/logo) in the cleaning queue.
`select_click`	User clicks on a detection item to select it for action.
`delete_click`	User marks a single detection item for deletion.
`rename_click`	User marks a detection item for renaming (label correction).
`delete_execute`	User executes all pending deletions (batch commit).
`rename_execute`	User executes all pending renames (batch commit).
`log_cleaning`	Summary event logged when a cleaning action completes. Contains counts of deleted/renamed items in `user_data`.
`page_change`	User navigates to a different page in the detection gallery.
`filter_change`	User changes a filter setting (deleted mode, sort order, date range, channel, event).
`filter_apply`	User applies range-slider filters (size, duration, confidence, movement).
`filter_reset`	User resets all filters to their defaults.
`toggle_cluster`	User toggles the cluster grouping mode on or off.
`toggle_inverse`	User toggles inverse selection mode (select everything except the current selection).
`mode_switch`	User switches between cleaning and renaming mode.
`clear_selection`	User clears the current item selection.
`undelete_click`	User restores a single previously-deleted item.
`undelete_all_asset`	User restores all deleted items for the current asset.
`delete_all_cluster`	User deletes an entire cluster of similar detections at once (Cluster Cleaner).
`delete_chat`	User deletes a single chat message (Chat QS Cleaner).
`delete_chats_multiple`	User deletes multiple chat messages at once (Chat QS Cleaner).
`undelete_chat`	User restores a previously-deleted chat message (Chat QS Cleaner).
`undelete_chats_multiple`	User restores multiple previously-deleted chat messages (Chat QS Cleaner).
`set_sentiment`	User sets the sentiment label (positive/neutral/negative) on a chat message (Chat QS Cleaner).
`set_source`	User sets the source type (organic/bot) on a chat message (Chat QS Bot Cleaner).
`select_sender`	User selects a chat sender to filter or inspect their messages (Chat QS Cleaner).

Jobs Manager `jobs_manager`

Event	Description
`create_yolo_batch`	User creates a new YOLO training batch from job videos (frame extraction + S3 upload).
`toggle_ready_for_process`	User toggles the “ready for process” flag on a job.
`update_customer`	User changes the customer assignment of a job.
`toggle_voice_job`	User toggles the voice/audio job flag.
`set_priority`	User changes the priority level of a job.
`reset_job`	User resets a job to its initial state.
`archive_job`	User archives a completed or inactive job.
`unarchive_job`	User restores an archived job.
`delete_job`	User deletes a job (soft delete).
`update_job`	User updates job metadata (e.g. video list, settings).
`create_job`	User creates a new job, specifying model and voice job flag.
`clear_video_errors`	User clears broken/errored video entries for a job.
`clear_audio_errors`	User clears broken/errored audio entries for a job.
`recreate_tsv`	User triggers recreation of the TSV output file for a job.
`delete_batch`	User deletes a YOLO training batch from a job.
`upload_batch`	User uploads a YOLO training batch to a job.
`skip_train`	User skips YOLO model training for a job.
`page_open`	User opens the Time Tracking page (logged with job `test_henlein`).
`export_events_csv`	User exports the raw events table as CSV (Time Tracking page).
`export_workers_csv`	User exports the workers summary table as CSV (Time Tracking page).
`export_chart_csv`	User exports a chart’s data as CSV (Time Tracking page).
`reset_filters`	User resets all filters on the Time Tracking page.
`apply_filters`	User applies date/module/worker/customer filters on the Time Tracking page.
`view_worker_timeline`	User opens the detailed timeline view for a specific worker (Time Tracking page).

Customer Dashboard `customer`

Events from the customer-facing dashboards (Summary, Video, Channel, Voice, Social, Articles, etc.). Job field contains the customer name (kunde).

Event	Description
`dashboard_enter`	User opens a customer dashboard page.
`dashboard_filtered`	User applies or changes filter settings on the dashboard.
`ui_view_dropdown`	User selects an option from a chart/view dropdown (e.g. switching chart type or metric).
`ui_table_pagination`	User navigates to a different page in a data table.
`ui_filter_column_toggle`	User expands or collapses a filter column in the side panel.
`ui_filter_items_collapse_expand`	User collapses or expands filter items within a filter group.
`ui_filter_sort`	User sorts filter items alphabetically or numerically.
`ui_nav_dropdown`	User selects an item from the main navigation dropdown.
`ui_nav_dropdown_download`	User clicks the download option in the navigation.
`ui_nav_quickfilter_dropdown`	User uses the quick-filter dropdown in the side navigation.
`ui_nav_archive`	User switches between archive years or ongoing view.
`ui_nav_currency`	User changes the currency setting.
`ui_nav_faq`	User opens the FAQ page.
`ui_nav_spm`	User navigates to the SPM (Sponsorship Performance Metric) section.
`ui_nav_additional_tabs`	User clicks additional tab links in the side navigation footer.
`ui_nav_notifications`	User opens the notifications panel.
`ui_nav_notifications_feature`	User clicks on a feature notification to see the feature intro.
`voice_play_audio`	User plays an audio snippet in the Voice dashboard.
`voice_show_chats`	User opens the chat transcript view in the Voice dashboard.
`voice_open_vod`	User clicks to open the VoD source (Twitch/YouTube) for an audio snippet.

Total Time

Active Workers

Avg. Time/Worker

Taiga

Activity flags

Regions Annotated

Time/Image

QS Cleaned

Items Cleaned

Time by Worker (Top 25)

Daily Trend (UTC days)

Time by Module

Time by Customer

Worker Details

Time Tracking KPI Glossary