Home/ User Guide/ Operator guide/ Media & sources

11 — Media & sources for incidents ⭐

(New, August 2026.) Photos, videos, news articles, satellite products and official social posts about a fire — collected, rights-checked and reviewed in one place, then attached to the incidents they document.

This article covers the Media & sources workspace, the review workflow, the two agents that propose new material (agency social pages, and web sites / news feeds per area), and the rules about what may and may not be downloaded.

The one rule to remember. The cockpit stores links and metadata. It downloads a media file only when the item's rights policy allows it, or when someone has recorded a written permission for that specific item. Nothing here scrapes a platform.


Why media attach to episodes, not directly to incidents

The press, agencies and satellite services publish about a fire episode — "the Karst fire of July 2022", "Padriciano, 1 August 2026". The incident registry, by contrast, records one row per municipality affected by that fire, because that is how the authorities report it.

So the cockpit keeps a fire episode in the middle:

episode  ──────  linked to several incidents  ──────  incident (municipality)
   │
   └── media items (photo, video, article, satellite product, social post)

A media item is linked to the episode; an incident then shows everything its episodes carry. Practically, this means:

  • One article about "the Karst fire" appears on every municipality incident belonging to that episode — you file it once.
  • An incident that belongs to no episode yet shows an empty gallery with the hint "Sources attach to fire episodes; link this incident to an episode…".
  • Some links are proposals the importer could not confirm automatically. They are shown as "Proposed link — to confirm" and wait for a human.

Where to find it

ScreenRouteWhat you get
Incident list/incident/indexa Media column with per-class chips (Video · Images · Audio · Publications · Satellite products · Social) and a Models badge (N runs, with IoU when available). Click the chips to open the gallery.
Incident form/incident/update?id=…a Media & sources card with the totals and "N awaiting review"
Media & sources workspace/incident/media?id=…the filter rail, the gallery and the inspector — the screen this article is about
Discovery sources/media-discovery-source/indexthe web sites, news feeds and searches the daily agent walks
Agency social profiles/operational-registry/index?entity=agency-social-profilesthe official pages of registry agencies that the hourly monitor reads

If /incident/media answers "Media gallery is not enabled on this host", the feature's database tables have not been installed on your deployment. Ask an administrator — see 08 — Administration.


The workspace, pane by pane

The rail (left) — filters

Four groups, each with live counts: Type · Platform · Rights · Review status. They combine, so "Video + Social — manual + Pending" is one click away.

Cards grouped by media class, in a fixed order: Video · Images · Audio · Publications · Satellite products · Social (reference only). Each card shows its rights pill and review pill, so you can scan a queue without opening anything.

Social cards state "Social reference — no asset is copied", and view-only cards state "Preview at the source only (platform)" — they are deliberately not previewable inside the cockpit.

The inspector (right)

Select a card to see the item's platform, publisher type, original type, content group (the axis that ties the same content across publications), source reliability, whether the URL has been verified, and the link basis (why this item is attached to this episode).

From here you open the source in a new tab and set the review status. The pill, the card and the rail counters update immediately — no page reload — and every change is written to the media audit trail.

About the map point. When an item is shown on a map, the position is inherited from the linked episode — it is not the camera's geotag. The inspector says so explicitly. Do not read it as "where the photo was taken".


Rights — what you may actually do with an item

Decisions are made on one field, the rights policy. (The free-text rights status you see in the inspector is information, never a permission.)

Policy shown asWhat it means for you
Open licenceThe asset may be fetched and stored.
Official productOfficial/institutional product — may be fetched and stored.
View onlyLook at it at the source. No copy is made.
Social — manualReference only. No asset is ever copied, whatever the platform shows.

The exception that makes copying legitimate. An individual item can carry a recorded permission: its status, its scope, and a link to the evidence (the mail, the written grant, the licence). That per-item record — not a general assumption, and never an automated download — is what allows the bytes of a specific press or social item to be ingested.


The review workflow

Every incoming item starts as Pending. In the inspector, set it to:

StatusUse it when
ApprovedThe item is genuine, relevant and its rights are clear.
Needs permissionWorth keeping, but a written grant is required before touching the file.
DuplicateThe same content already exists (typically under another publisher — check the content group).
RejectedNot relevant, not this fire, or unusable.

Alongside the status you can:

  • Link an item to an episode, or unlink it (a wrong automatic proposal).
  • Cluster related sources — groups the episode's items by content group, so the same photo republished by four outlets is visible as one story rather than four finds.

Every action is recorded with who did it and when.

Visual analysis is not available yet. The Analyze action deliberately answers that the visual-analysis pipeline is not connected, and records the refusal in the audit trail so the gap stays visible instead of failing quietly.


Where new material comes from

1. The seed import (one-off, administrator)

A curated corpus of episodes, media and links is imported once by an administrator from the project's data mount. It is strict on purpose: the importer verifies the expected number of episodes, media and links and stops if they differ, never creates incidents, and never downloads a file. Automatic confirmation happens only when a fire's native code and its start time match to the minute; everything else is filed as a proposal for you to confirm.

2. Official agency pages — hourly

Agencies in the operational registry can register their official social pages. Once a human has marked a page as verified and switched monitoring on, an hourly job proposes new incident-related posts into the Pending queue as Social — manual references.

Boundaries, by construction:

  • Only official pages of registry agencies — never private profiles.
  • Read only through the platform's approved interface. If a deployment has no approved access, the platform is simply reported as unconfigured and skipped. There is no scraping fallback.
  • Linking a proposal to an episode stays a human act.

3. Web sites, local outlets and news feeds — daily

/media-discovery-source/index holds the sources a daily agent walks for each area of interest: news feeds, local outlets, agency web sites, and social profiles (which are handed to the agency lane above rather than fetched here). Saved Google searches are retired — their result links hide the publisher behind an opaque token, so nothing usable ever came back; add the outlet's own feed instead. Each source can be enabled or disabled from the list, and shows when it last ran and what went wrong if anything did.

The agent proposes links and metadata only, always as Pending and "view only", and silently ignores anything already captured. Search-result and profile pages are refused as sources — the same rule the manual "add source" path applies.

A known limitation, stated honestly. News-aggregator results now hide the publisher's real address behind an opaque token, so that lane often resolves to nothing and reports it on the source row. Outlet and feed sources carry discovery in the meantime — prefer them when adding sources.


The wildfire diary — /incident/diary?id=… (August–September 2026)

Every incident that belongs to a fire episode has a diary: the book icon in the incident list's row actions, or the Wildfire diary link in the gallery. It is the fire told in order, built deterministically from the records — nothing in it is invented, and every entry cites its row.

  • Summary — the stats line (days of coverage, sources, media by class) and a deterministic summary of what the records say.
  • Generated report — the model-written narrative, when one exists. It is shown with its review state; an unreviewed narrative is named but withheld from exports. Generate report from sources is available when at least one source has retrieved text and no accepted narrative exists yet.
  • Timeline — one merged stream, by day (phase 6, September 2026): incident and episode milestones, detections (satellite hotspots and other hazard events linked to the fire — same-minute pixels from one source collapse into one entry), dispatches (rolled / on scene / returned / cancelled, with vehicle counts and a link to the run), simulation runs, perimeter updates and emissions estimates, and every media item with its stored excerpt. Undated sources are listed last.
  • Intervention cost — the dispatch hours priced against an approved cost factor set (the same versioned, named-reviewer mechanism the CO₂ inventory uses). No set is approved on most deployments yet, so the block reads "No approved cost factor set — the intervention cost is not estimated rather than invented"; closed dispatches only, billable hours floored at the set's minimum.
  • Exports — JSON ↓ for the raw diary and PDF ↓ (September 2026) for a document through the same generator as the incident report; the PDF carries the narrative only when it is accepted, because it is a document people may cite.

Rights approvals — /rights-approval/index (September 2026)

The written side of the rights rules above. Two kinds of approval live here: inbound — may we hold and use this publisher's material? — and outbound — may this researcher receive it?. Each moves draft → requested → approved / refused, and an approved one can later be revoked or expire; every terminal state can be requested again (a publisher who refused in June may agree in September). An approval is only ever approved with documentation attached (the scope and the evidence — an e-mail, a licence page — uploaded to the record), which is what lets an item's bytes be exported at all.

  • Researchers ask through Request access; the request lands in the outbound queue for an operator to decide.
  • Datasets — an outbound approval names a published dataset version; the holder downloads the manifest or single assets from the gated download page. Every download is logged, no administrator bypass exists, and an item whose publisher grant no longer allows bytes stops downloading immediately while the manifest row stays as history.
  • Inbound approvals are what the workspace's Needs permission → Permission requested → granted / refused pills reflect.

  1. Open /incident/index and look at the Media chips — an incident with a pending count is one that needs attention.
  2. In the workspace, filter Review status → Pending.
  3. Work top-down: for each item, open the source, decide, set the status.
  4. Use Cluster related sources before hunting duplicates by hand.
  5. Confirm or unlink the "Proposed link — to confirm" items — those are the automatic matches nobody has checked yet.
  6. Anything you want the actual file of, but may not take: mark Needs permission and start the written request. Record the evidence link when it arrives.

Troubleshooting

SymptomLikely cause / fix
"Media gallery is not enabled on this host"The tables are not installed on this deployment — administrator task.
The gallery is empty but the fire is documentedThe incident belongs to no episode yet. Link it to one.
An item shows no previewExpected for View only and Social — manual: nothing is copied into the cockpit. Use Open source.
The same photo appears several timesDifferent publishers of the same content. Use Cluster related sources, then mark the extras Duplicate.
A social page yields nothingThe page is unverified, monitoring is off, or the deployment has no approved platform access.
A discovery source shows an error about unresolvable linksThe aggregator lane limitation above. Add the outlet's own feed instead.
Analyze refusesCorrect — visual analysis is not wired yet.
The diary's Intervention cost says no approved cost factor setExpected until an administrator imports and approves a cost pack; the estimate is withheld, not zero.
A diary PDF has no narrativeThe narrative is not accepted yet — review it in the diary first.