Glossary
Alphabetical reference for the vocabulary used across ContentsVision and this handbook. For how the core entities nest, see concepts.
Adjuster id
The Adjust Square user id that a submitted claim is attributed to. It is the signed-in user's external_id. See Adjust Square.
Adjust Square
The external claims platform ContentsVision submits finished inventories to. Distinct from AdjustSquare SSO, the identity provider used for login.
Approved
An item explicitly cleared for export and submission (approved = true). Only approved items are sent when "approved only" is selected. Set manually, in bulk, or by auto-approval.
Auto-approval
Configurable rules that approve items automatically when they pass a confidence threshold and condition filter. See automation and prompts.
Chunk (time chunk)
A fixed-length slice of a video (default 30 seconds) analyzed by one AI call. Chunks run in parallel. See the AI pipeline.
Claim
The top-level unit of work, mirroring an insurance claim. Owns rooms; carries the metadata and, after submission, the Adjust Square claim key.
Confidence
A 0.0-1.0 score the AI assigns each item by how clearly it was seen and whether it was named in narration. Items below 0.6 are flagged.
Credits
The unit customers are billed in. Video and audio are charged per 20 second block (rounded up); photos per AI request. Separate from the internal dollar cost. See billing.
Cross-chunk dedup
The step that merges the same item reported by two adjacent video chunks near their shared boundary, so one physical object is not listed twice.
Flagged
A boolean marking an item as needing attention. Set automatically for confidence below 0.6, and respected by auto-approval (skip_flagged).
Guided Review
A focused, one-item-at-a-time review mode (ReviewMode.tsx) with keyboard shortcuts; for audio items it can replay the relevant slice of the recording.
Item (inventory item)
One line of the inventory: name, quantity, condition, age, brand, evidence, photos, and a claim-unique item number.
Item number
A sequential number unique within a claim, continuing across all jobs in that claim. AI items are numbered in capture order; manual inserts can shift later numbers.
Item photo
A photo attached to a finished item (the AI crop, the source photo, extra angles), stored in item_photos. The source of truth for item imagery.
Job (processing job)
One upload and its processing run, with a source_type of video, photo, or audio, and a status that advances through the pipeline.
Loss type / peril
The cause of loss (Fire, Flood, Tornado, etc.). Mapped to an Adjust Square peril code on submission.
OpenRouter
The routing layer ContentsVision calls for Claude vision and analysis (default model anthropic/claude-opus-4-7).
Pipeline
The staged AI process that turns an upload into items. Three variants: video, photo (grouped), and audio. See the AI pipeline.
PostHog
The optional product-analytics service. When configured, it captures page views, session replays, and a server-side event for every user action and job outcome, keyed on the same user and team identity on both sides. Off by default. See analytics.
Request (photo request)
One AI call in the photo pipeline (one per group). The billing unit for photos.
Room
A labeled subdivision of a claim ("Kitchen"). Media is uploaded into a room.
Source type
Whether an item or job came from video, photo, or audio. Drives badges, billing, and which pipeline runs.
SSE (Server-Sent Events)
The one-way stream the frontend uses to receive live job progress, logs, and final metrics from GET /api/jobs/{id}/stream.
PDF recipe
The rules for reading one vendor's PDF page layout: which of the two layout families it belongs to (one photo per item, or a heading owning a run of photos), which side of a photo the description sits on, what the item label looks like, what a room heading looks like, what page furniture to ignore. Written by the AI once per format from a small sample, saved against a fingerprint of the layout in pdf_templates, then executed by the server over every page. A format we have seen before costs no AI call at all. See PDF import.
Source note
The wording that came with a photo from outside the app: printed beside it in a source PDF (e.g. "Charred barber products"), or typed against it by the homeowner on an insured inventory link (e.g. "the TV is behind the boxes on the left"). Kept on the staged photo verbatim, handed to the AI as a hint when that photo is analyzed, and saved onto the finished item as a note. It is never used as the item's name: the AI is told to identify what it can see FIRST and use the note only to confirm, because a note read the other way round makes a model report things that are not in the picture.
Insured
The homeowner or policyholder who suffered the loss. Not a user of ContentsVision and never will be: they have no account. Distinct from the adjuster, who is the user.
Insured inventory link
A tokenized link an adjuster emails to the insured so they can build their own inventory for one claim, room by room, on their phone. The app's only public write surface. Photos and videos they add wait for the adjuster to press Process, so nothing they do spends a credit; items they type in cost nothing at all, because they named them. See Insured inventory link.
Staged photo
A raw uploaded photo waiting to be grouped before AI runs (staged_photos). The user merges shots of the same item into groups.
Synthesize
The final pipeline stage: sort, crop photos, number, and save items to the database.
Team
An organization. Every record is scoped to a team; users only see their own team's data.
Team token
A team's Adjust Square API token (team.team_token), captured from the SSO login token and used as the bearer token for claim submission.
Transcription
Speech-to-text, done by Deepgram Nova-2. Drives item naming for video and audio.
Reviewed
A boolean marking that a human looked at an item. Set when an item is edited.
Submitted
A boolean marking that an item has been sent to Adjust Square (submitted). Submitted items are read-only and excluded from later batches.