Tecnobosque Wildfire Black Box · Proposed evidence protocol · Research note · Prepared 5 November 2026 · Reviewed 6 November 2026
The problem: hindsight changes the story
An incident map downloaded today may describe a wildfire from several years ago. A field report may refer to a morning observation while the public press release containing it appeared the following afternoon. A later correction may replace a provisional incident acreage without creating any new burned area. Researchers who merge all these facts into one undated narrative risk hindsight bias: attributing knowledge to a decision-maker before that knowledge demonstrably existed.
Wildfire Black Box needs a stricter question: “What claim, supported by which version of which source, was publicly available by time T?” This is different from asking what the most accurate map shows today.
The five timestamps that matter
| Field | Meaning | What not to assume |
|---|---|---|
observed_at |
When an image, sensor or human observation was collected | The data were already public then |
source_published_at |
When the issuing organisation released the statement or file | The release’s date equals its observation time |
first_verified_public_at |
Earliest independently evidenced public availability (e.g. retained archive capture) | A later file timestamp proves earlier publication |
retrieved_at |
When Tecnobosque downloaded or read the source | Retrieval proves historic availability |
analysed_at |
When Tecnobosque computed or interpreted the evidence | The analysis was available to responders during the incident |
For revised resources, additionally retain source_updated_at, supersedes_version_id and a content checksum when the source can lawfully be stored. Use UTC ISO 8601 with offsets and do not silently replace unknown timestamps with midnight. For statements lacking a precise time, store the original date with a separate time_precision field (minute, hour, day, approximate, unknown).
Version rather than overwrite
One claim can legitimately have several states:
- C1/v1 — reported: “Area is approximately 1,200 ha,” issued on a particular day.
- C1/v2 — revised: “Reprocessed polygon measures 1,235 ha,” released later.
- C1/v3 — reconciled: “The difference is mainly a change in mapping method,” if a cited authority or reproducible comparison supports this interpretation.
The numbers are a synthetic teaching example, not a real incident. C1/v2 must not erase C1/v1: the earlier report remains historical evidence of the information environment at the earlier publication time. A revision must carry a reason, source URL and relation to the prior version.
Tecnobosque proposed evidence object
Store the following fields for each claim. It is a proposed house schema, not an ISO or emergency-service standard.
{
"claim_id": "TB-EXAMPLE-0001",
"claim_version": 2,
"incident_id": "SYNTHETIC-TRAINING-ONLY",
"claim_text": "A mapped area of 1235 ha was reported",
"measure_type": "mapped_operational_perimeter_area",
"value": 1235,
"unit": "ha",
"source_organisation": "Synthetic training source",
"source_url": null,
"observed_at": null,
"source_published_at": "2026-01-14T14:00:00Z",
"first_verified_public_at": null,
"retrieved_at": "2026-01-15T09:00:00Z",
"analysed_at": "2026-01-16T10:00:00Z",
"supersedes_version_id": "TB-EXAMPLE-0001/v1",
"time_precision": "minute",
"verification_status": "illustrative_not_real"
}
The known-at test: conservative rules
Given a target time T, a claim may be labelled demonstrably public by T only if an independently credible publication or archive record places that version at or before T. A publication date alone may support a reasonable inference, but a weakly dated, editable or retrospectively updated page is not enough for a confident historic availability claim. An archive captured after T proves the document existed by the capture time, not necessarily by T.
Use three status outputs:
- Confirmed by T: a dated, verifiable source snapshot or equivalent record supports availability by T.
- Likely but unconfirmed by T: publication metadata suggest availability, but there is no independently preserved contemporary version.
- Not established by T: later-only sources or insufficient timestamps. This is not proof the claim was unknown; it means the evidence does not establish availability.
Distinguish publicly known from known to a particular incident commander or agency: no public database alone establishes who read or acted on information. Operational decision reconstruction requires additional records and source-specific disclosure limits.
Source corrections and near-real-time products
Near-real-time satellite and incident datasets often change. NASA FIRMS documents replacement of near-real-time data by later standard science-quality products when available. The NIFC perimeter metadata expressly cautions that its datasets are dynamic and derived from sources of variable reliability. Consequently, a later reprocessed dataset should be stored as a new evidence version, not silently pasted into an older timeline. NASA FIRMS archive · NIFC metadata
Publication checklist for a Black Box timeline
Every timeline row should display the claim in neutral language; source organisation; date precision; observation versus publication dates; confidence in historical availability; whether later revisions exist; and an accessible source citation. Interpretations must be labelled as such, and lack of evidence must not be written as evidence of absence. If a source later changes, retain a revision note describing what changed and why it affects the reading of the timeline.
Original contribution: this five-timestamp, versioned claim approach is a Tecnobosque proposed research architecture. Its usefulness can be demonstrated in incident-level case studies, but it should not be represented as an adopted public-authority standard or a forensic certification.
Related Tecnobosque research: Perimeter discrepancies; QGIS comparison; Fire archive.