Before any assessment method is applied, each panorama gets the same universal record. That record is the base truth: a lens joins it and adds its own reading, but never rewrites it. This is what stops two different assessments of the same corridor from producing two contradictory accounts of what is physically there.
Every object read out of a scene is anchored to the pixels it came from. The compass bearing and elevation angle are then computed from panorama geometry — never estimated. If a corridor's north reference is recalibrated later, each observation moves with it, because it is still attached to the same pixels.
Signs, markers, gauge plates, and regulatory postings are transcribed and flagged. Clear means it was legible in the imagery — a verified fact. Partial means it was only half-readable — an unverified read, carried as such. The distinction is stored, not smoothed over.
Surface, water character, the season as it appears, how open or enclosed the view is, whether people are present. Corridor-level facts that hold regardless of which assessment method you later apply.
Streams, summits, crossings, and populated places from the national geographic names record, matched to the scenes that can actually see them, distance- and bearing-checked, and surfaced as located reference points.
Where a corridor has been captured more than once, repeat captures of the same reach are linked and walkable by date, so a condition can be compared against its own baseline rather than against a recollection.
A reviewer can correct a position, hide an observation, or fix a transcription. Those edits are marked as human-owned and survive every subsequent re-read of the corridor. A search index over this record is in progress and can be walked through on request.
License a corridor once and read it through more than one assessment method, without standing up a second viewer or forking the scene record. Each lens is versioned: the rules that produced a result are recorded alongside the result, so a finding can be re-run, compared, or defended two years later. 2 lenses are live today.
The universal read. Every scene in the corridor gets it: located objects with true bearings, transcribed on-scene text, surface and water character, named places, and the capture record itself.
The foundation every paid protocol lens is joined to.
River Access Planning Guidelines applied to a captured access site — launch profile, safety and condition observations, accessibility observations, visitor-management signals, experience indicators, and recreation profile.
Delivered to a client corridor and published. Your measured access classification stays authoritative: the lens supplies observations, not a ruling.
NPS Visual Resource and MS4 Stream Assessment are declared names in our theme catalog. They are not offered, not quotable, and not runnable until the protocol work behind them is built and reviewed. We list them here because a name on a roadmap is not a capability, and you should be able to tell the difference from the outside.
An observation is only placed on a scene as a precise marker if its geometry holds up. Everything else is kept — it stays in the record and stays reviewable — but it is not rendered, and it is never delivered as though it were located. A five-tier ladder decides, and the tier travels with the observation.
| Tier | What it means | What happens to it |
|---|---|---|
| verified | A reviewer marked the object in the source pixels. | Placed as a precise marker. |
| reliable | Valid source face and anchor, grounding confidence at or above 0.75. | Placed as a precise marker. |
| approximate | Valid geometry, confidence between 0.55 and 0.749. | Retained for review. Not drawn as a precise marker. |
| unreliable | Missing or invalid geometry, a mismatch, or confidence below 0.55. | Retained for review. Not drawn. |
| legacy | An older result whose direction was estimated rather than computed from pixels. | Labelled legacy. Never silently reclassified as reliable. |
The mechanism is not a clever one, and that is the point: the reading identifies pixels, and deterministic code turns pixels into a compass bearing. Nothing is ever asked to guess a direction. It is also why a delivery can honestly report how much of what it read it chose not to show you.
The RAPG Access Sites lens shipped on a real river corridor in northwest Montana. These are the release's own audit figures, not a tidied summary of them.
The universal pass recorded 3,897 observations across the corridor. 1,860 met the bar to be placed as precise markers; 1,357 were held as approximate and 680 as unreliable — kept in the record, reviewable, and not drawn. Under half of what was read was judged good enough to put on a map, and the delivery says so.
A read-only audit validated scene and result counts, ready and error totals, theme and version identity, confidence and bearing ranges, matching source face and image index, and public-trail state. It found zero face or index mismatches, zero coordinate-range errors, zero renderable observations missing coordinates, and zero overlap with excluded trails. The published run passed with no warnings and no errors.
$19.14 is the total provider spend for the grounded universal and RAPG passes over all 400 panoramas. Including a separate assessment of 18 field composite sheets, the delivery represents $19.75 in provider spend. That is analysis cost, not project price — capture, review, integration, and delivery are the work. We publish it because this line item is normally hidden, and an institutional buyer is entitled to see it.
One site in the release carries 18 panoramas; another carries three and is labelled limited coverage. Three named sites were excluded outright, with zero scene overlap, rather than being padded out. The client's own field measurements remain authoritative throughout, and their access classification was deliberately left out of what we published rather than being overridden by a reading.
Figures as recorded in the release audit for the published run, 29 July 2026. A walkthrough of the live corridor is available on request.
The register is deliberate. Scene Intelligence is built for work that somebody else reviews — a permit file, a grant packet, a design memo, a council presentation, a claim.
Located, dated observations across a whole corridor feed workflows that already exist in your organization. What follows is what the record supports — joined to your own measurements and field evidence, which stay authoritative.
Surface the corridor locations where condition observations cluster, each with coordinates, a bearing to the object, and the scene it was read from. A field crew gets a location list, not a folder of photographs.
A dated, corridor-wide baseline is the hardest part of most funding packets to produce. Repeat captures turn the same record into before-and-after evidence for the next cycle.
Aggregate corridor-level observations to argue for a program rather than a project. Find where an issue is systemic instead of anecdotal — and where the evidence is thin enough that somebody should go and look.
Hazard observations carry a date, a location, and the imagery they were read from. That combination is what makes a record of awareness defensible later, whether the outcome is a repair, a closure, or a claim.
The record notes what is visible in the imagery — surface, grade as it appears, obstructions, parking and approach conditions. It is an input to a transition plan and a site visit. It is never a compliance determination on its own.
After a flood, storm, or fire, re-capture the corridor and compare it against its own baseline instead of against recollection. The comparison is scene-for-scene, at the same viewpoints.
Give bidders the actual pre-construction condition of a corridor rather than a written description of it, then verify completed work against the same viewpoints.
Drainage patterns, existing conditions, and problem locations documented across the whole corridor before design starts — including the segments nobody walked because they were inconvenient to reach.
The second list is the more useful one. If a vendor cannot tell you where their evidence stops, you will find out yourself at the worst possible moment.
Terrain360 also runs an older, self-contained tool that assesses individual uploaded photographs — not captured corridors — against NPS Visual Resource Inventory, MS4 visual stream assessment, USDA erosion and sediment control, and RAPG river access protocols. It works and it is supported, and it is a different system: it does not read the captured corridor, and its output does not join the scene record described on this page. It is kept here for the customers already using it.
Sign in to the single-photo tool →Tell us the reach, the assessment method you work to, and what the record has to survive. We will scope the capture, the lens, and what the delivery can honestly claim.
Start a conversation →