Research Codebook
Version 2026-09-21.1; schema 1. Change history: initial public derived-record vocabulary, 18 September 2026; two hold classes added, 21 September 2026 - see the version register for why. Examples below are illustrations, not events.
Identifiers, UTC observation/capture times, references and content hashes are structural fields. Null means unrecorded or not applicable to that record kind, never zero. A hash of decision-time text is available only when the producer recorded it. Current cluster text is never substituted for missing historical text. These snapshots carry this codebook; they do not reconstruct historical taxonomy or source-registry versions that were never captured.
feature: archive-arrival-polls
Archived feed polling outcomes.
Assignment: Copied from the observation series that produced this coverage row.
Illustration: A successful feed check with no changed material is a polling observation, not a count of real-world events.
Limit: A feature states what was measured. Published reporting never measures all real-world activity. Category slugs do not reconstruct uncaptured historical definitions.
feature: receiver-contacts
Instrument contacts within recorded receiver coverage.
Assignment: Copied from the observation series that produced this coverage row.
Illustration: Aircraft position reports contribute contacts only inside a supported observed grid.
Limit: A feature states what was measured. Published reporting never measures all real-world activity. Category slugs do not reconstruct uncaptured historical definitions.
feature: published-category:<slug>
Published reporting in the named category and grid.
Assignment: Copied from the observation series that produced this coverage row.
Illustration: Separate category identifiers keep a fire reporting series distinct from a flood reporting series in the same cell.
Limit: A feature states what was measured. Published reporting never measures all real-world activity. Category slugs do not reconstruct uncaptured historical definitions.
holdClass: analysis
An explainer, an interview, an argument or a piece asking what something means. It concerns a real event without being one.
Assignment: Assigned by the existing public hold-reason classifier; unrecognised reasons remain other.
Illustration: An article explains the consequences of an earlier event without reporting a new occurrence: analysis.
Limit: A hold is not a finding that the underlying claim is false. Diagnostic decision arithmetic is excluded.
holdClass: planned
Announced, scheduled or being prepared. We record what has occurred.
Assignment: Assigned by the existing public hold-reason classifier; unrecognised reasons remain other.
Illustration: An announcement schedules an action for a future date: planned.
Limit: A hold is not a finding that the underlying claim is false. Diagnostic decision arithmetic is excluded.
holdClass: aftermath
Recovery, cleanup, a revised casualty count or a damage assessment. The report is recent; the event it follows is not.
Assignment: Assigned by the existing public hold-reason classifier; unrecognised reasons remain other.
Illustration: A cleanup report follows an already recorded incident: aftermath.
Limit: A hold is not a finding that the underlying claim is false. Diagnostic decision arithmetic is excluded.
holdClass: retrospective
An anniversary, a history or a feature. Real, and not news.
Assignment: Assigned by the existing public hold-reason classifier; unrecognised reasons remain other.
Illustration: An anniversary feature recounts an old incident: retrospective.
Limit: A hold is not a finding that the underlying claim is false. Diagnostic decision arithmetic is excluded.
holdClass: speculated
May, might, could, is expected to, analysts believe.
Assignment: Assigned by the existing public hold-reason classifier; unrecognised reasons remain other.
Illustration: An analyst says an action might happen, without an announcement or occurrence: speculated.
Limit: A hold is not a finding that the underlying claim is false. Diagnostic decision arithmetic is excluded.
holdClass: claimed
A claim we are not willing to carry as a record of what occurred.
Assignment: Assigned by the existing public hold-reason classifier; unrecognised reasons remain other.
Illustration: A party says an incident occurred and independent confirmation is absent: claimed.
Limit: A hold is not a finding that the underlying claim is false. Diagnostic decision arithmetic is excluded.
holdClass: denied
A denial is not a record of the thing being denied.
Assignment: Assigned by the existing public hold-reason classifier; unrecognised reasons remain other.
Illustration: A statement disputes an alleged incident; the statement is classified denied, not proof of non-occurrence.
Limit: A hold is not a finding that the underlying claim is false. Diagnostic decision arithmetic is excluded.
holdClass: roundup
A digest, a daily briefing or a live blog. Not one occurrence.
Assignment: Assigned by the existing public hold-reason classifier; unrecognised reasons remain other.
Illustration: A daily digest combines unrelated incidents rather than describing one occurrence: roundup.
Limit: A hold is not a finding that the underlying claim is false. Diagnostic decision arithmetic is excluded.
holdClass: stale
The reports carry a date well in the past. Usually an anniversary piece or an archive item resurfacing, occasionally a model calling a history feature the news - the date settles it either way.
Assignment: Assigned by the existing public hold-reason classifier; unrecognised reasons remain other.
Illustration: A newly collected article describes an occurrence outside the publication recency window: stale.
Limit: A hold is not a finding that the underlying claim is false. Diagnostic decision arithmetic is excluded.
holdClass: uncovered
It happened, and it is not one of the kinds of event this service records. A drug seizure is real; it is not on this map.
Assignment: Assigned by the existing public hold-reason classifier; unrecognised reasons remain other.
Illustration: The classifier finds no matching category in the service taxonomy: uncovered.
Limit: A hold is not a finding that the underlying claim is false. Diagnostic decision arithmetic is excluded.
holdClass: uncorroborated
We publish what more than one independent source reports, with a narrow exception for official wires filing on their own account.
Assignment: Assigned by the existing public hold-reason classifier; unrecognised reasons remain other.
Illustration: Independent corroboration is absent and the official-wire exception does not apply: uncorroborated.
Limit: A hold is not a finding that the underlying claim is false. Diagnostic decision arithmetic is excluded.
holdClass: within-reach
Reported, and close to the level we require before we will carry an event of its kind. A further independent account would have been enough.
Assignment: Assigned by the existing public hold-reason classifier; unrecognised reasons remain other.
Illustration: A stored hold reason records that the decision fell short of the threshold for its kind and that corroboration would have carried it: within-reach; the threshold and the decision value are not exported.
Limit: A hold is not a finding that the underlying claim is false. Diagnostic decision arithmetic is excluded.
holdClass: far-below
Reported, and far enough below the level we require that further corroboration would not have changed the answer.
Assignment: Assigned by the existing public hold-reason classifier; unrecognised reasons remain other.
Illustration: A stored hold reason records that the decision fell short of the threshold for its kind by a margin further corroboration would not close: far-below; the threshold and the decision value are not exported.
Limit: A hold is not a finding that the underlying claim is false. Diagnostic decision arithmetic is excluded.
holdClass: not-new
The reports were about something already on the map, or did not clear the bar we set for its kind.
Assignment: Assigned by the existing public hold-reason classifier; unrecognised reasons remain other.
Illustration: A later report describes an occurrence already represented on the map: not-new.
Limit: A hold is not a finding that the underlying claim is false. Diagnostic decision arithmetic is excluded.
holdClass: other
Recorded, and not published. The engine gave a reason we do not summarise here.
Assignment: Assigned by the existing public hold-reason classifier; unrecognised reasons remain other.
Illustration: A stored hold reason matches no public reason class: other; its diagnostic text is not exported.
Limit: A hold is not a finding that the underlying claim is false. Diagnostic decision arithmetic is excluded.
modality: occurred
Reported as an occurrence.
Assignment: Copied from the event classification at capture, not inferred from publication status.
Illustration: Reporting describes a completed occurrence, so its recorded modality is occurred.
Limit: Modality records how reporting was classified; it does not settle every detail. Null means unrecorded.
modality: planned
Announced, scheduled or being prepared. We record what has occurred.
Assignment: Copied from the event classification at capture, not inferred from publication status.
Illustration: An announcement schedules an action for a future date: planned.
Limit: Modality records how reporting was classified; it does not settle every detail. Null means unrecorded.
modality: speculated
May, might, could, is expected to, analysts believe.
Assignment: Copied from the event classification at capture, not inferred from publication status.
Illustration: An analyst says an action might happen, without an announcement or occurrence: speculated.
Limit: Modality records how reporting was classified; it does not settle every detail. Null means unrecorded.
modality: claimed
A claim we are not willing to carry as a record of what occurred.
Assignment: Copied from the event classification at capture, not inferred from publication status.
Illustration: A party says an incident occurred and independent confirmation is absent: claimed.
Limit: Modality records how reporting was classified; it does not settle every detail. Null means unrecorded.
modality: denied
A denial is not a record of the thing being denied.
Assignment: Copied from the event classification at capture, not inferred from publication status.
Illustration: A statement disputes an alleged incident; the statement is classified denied, not proof of non-occurrence.
Limit: Modality records how reporting was classified; it does not settle every detail. Null means unrecorded.
modality: retrospective
An anniversary, a history or a feature. Real, and not news.
Assignment: Copied from the event classification at capture, not inferred from publication status.
Illustration: An anniversary feature recounts an old incident: retrospective.
Limit: Modality records how reporting was classified; it does not settle every detail. Null means unrecorded.
modality: analysis
An explainer, an interview, an argument or a piece asking what something means. It concerns a real event without being one.
Assignment: Copied from the event classification at capture, not inferred from publication status.
Illustration: An article explains the consequences of an earlier event without reporting a new occurrence: analysis.
Limit: Modality records how reporting was classified; it does not settle every detail. Null means unrecorded.
modality: roundup
A digest, a daily briefing or a live blog. Not one occurrence.
Assignment: Copied from the event classification at capture, not inferred from publication status.
Illustration: A daily digest combines unrelated incidents rather than describing one occurrence: roundup.
Limit: Modality records how reporting was classified; it does not settle every detail. Null means unrecorded.
modality: aftermath
Recovery, cleanup, a revised casualty count or a damage assessment. The report is recent; the event it follows is not.
Assignment: Copied from the event classification at capture, not inferred from publication status.
Illustration: A cleanup report follows an already recorded incident: aftermath.
Limit: Modality records how reporting was classified; it does not settle every detail. Null means unrecorded.
availability: not-polled
No observation attempt was recorded.
Assignment: Derived from archived polling outcomes, exposure and supported receiver geometry.
Illustration: An expected feed has no poll record.
Limit: Availability is distinct from traffic and from reporting activity. Home readers cannot support absence claims.
availability: unreachable
Attempts failed without a clean response.
Assignment: Derived from archived polling outcomes, exposure and supported receiver geometry.
Illustration: A receiver returned an error throughout the window.
Limit: Availability is distinct from traffic and from reporting activity. Home readers cannot support absence claims.
availability: partial
The observation window is incomplete.
Assignment: Derived from archived polling outcomes, exposure and supported receiver geometry.
Illustration: A receiver answered during only part of the window.
Limit: Availability is distinct from traffic and from reporting activity. Home readers cannot support absence claims.
availability: answered-empty
The source answered and no arrival was recorded.
Assignment: Derived from archived polling outcomes, exposure and supported receiver geometry.
Illustration: A fully observed receiver window contains no contacts.
Limit: Availability is distinct from traffic and from reporting activity. Home readers cannot support absence claims.
availability: arrived
Material arrived during observed coverage.
Assignment: Derived from archived polling outcomes, exposure and supported receiver geometry.
Illustration: A receiver returned a position report.
Limit: Availability is distinct from traffic and from reporting activity. Home readers cannot support absence claims.
availability: unsupported-coverage
No usable geographic observation footprint is recorded.
Assignment: Derived from archived polling outcomes, exposure and supported receiver geometry.
Illustration: A vessel socket connection carries no geographic footprint.
Limit: Availability is distinct from traffic and from reporting activity. Home readers cannot support absence claims.
assessment: insufficient-history
Comparable observation history has not matured.
Assignment: Assigned independently for activity and reporting by the observation-history engine.
Illustration: A newly observed cell has too little comparable history; it remains insufficient-history even if the current poll is empty.
Limit: A comparison describes observations, not a cause, forecasted event or certainty. Internal thresholds are not exported.
assessment: coverage-changed
The coverage regime differs from the available history.
Assignment: Assigned independently for activity and reporting by the observation-history engine.
Illustration: A receiver footprint changes; observations from the previous footprint cannot establish the new baseline: coverage-changed.
Limit: A comparison describes observations, not a cause, forecasted event or certainty. Internal thresholds are not exported.
assessment: coverage-unrecorded
Required source or language coverage was not recorded.
Assignment: Assigned independently for activity and reporting by the observation-history engine.
Illustration: The source's reporting-language coverage was never recorded: coverage-unrecorded, even when it answered.
Limit: A comparison describes observations, not a cause, forecasted event or certainty. Internal thresholds are not exported.
assessment: calibration-failed
Recent held-out observations do not fit the earlier baseline.
Assignment: Assigned independently for activity and reporting by the observation-history engine.
Illustration: Recent observations reserved for checking no longer fit the older training period: calibration-failed.
Limit: A comparison describes observations, not a cause, forecasted event or certainty. Internal thresholds are not exported.
assessment: unsupported-coverage
Geographic observation is not supported.
Assignment: Assigned independently for activity and reporting by the observation-history engine.
Illustration: A vessel socket has no recorded geographic footprint: unsupported-coverage; arriving positions do not supply one.
Limit: A comparison describes observations, not a cause, forecasted event or certainty. Internal thresholds are not exported.
assessment: not-assessed
The observation cannot support this comparison.
Assignment: Assigned independently for activity and reporting by the observation-history engine.
Illustration: A feed is unreachable during the window: not-assessed, rather than an absence of activity.
Limit: A comparison describes observations, not a cause, forecasted event or certainty. Internal thresholds are not exported.
assessment: ordinary
The observed activity is within its historical baseline.
Assignment: Assigned independently for activity and reporting by the observation-history engine.
Illustration: A complete observation window falls within the calibrated range for comparable periods: ordinary.
Limit: A comparison describes observations, not a cause, forecasted event or certainty. Internal thresholds are not exported.
assessment: above-baseline
The observed activity is above its historical baseline.
Assignment: Assigned independently for activity and reporting by the observation-history engine.
Illustration: A fully observed window exceeds its calibrated comparable-period range: above-baseline; the code does not identify a cause.
Limit: A comparison describes observations, not a cause, forecasted event or certainty. Internal thresholds are not exported.
assessment: below-baseline
The observed activity is below its historical baseline.
Assignment: Assigned independently for activity and reporting by the observation-history engine.
Illustration: A fully observed window contains activity below its calibrated range: below-baseline; it is not empty.
Limit: A comparison describes observations, not a cause, forecasted event or certainty. Internal thresholds are not exported.
assessment: absence
An observed empty window is outside its historical baseline.
Assignment: Assigned independently for activity and reporting by the observation-history engine.
Illustration: A fully observed window is empty where mature, comparable and calibrated history supports activity: absence; no event is inferred.
Limit: A comparison describes observations, not a cause, forecasted event or certainty. Internal thresholds are not exported.
status: published
The record was published.
Assignment: Copied from the record at capture; a withdrawal keeps the permanent identifier.
Illustration: a published page remains citable at its permanent address.
Limit: A later capture may have a different status; an old snapshot is not today's record.
status: withdrawn
The published record was subsequently withdrawn.
Assignment: Copied from the record at capture; a withdrawal keeps the permanent identifier.
Illustration: a withdrawn page remains citable at its permanent address.
Limit: A later capture may have a different status; an old snapshot is not today's record.
outcome: corrected
A published correction changes the record.
Assignment: The shared correctionOutcome function maps the published update kind.
Illustration: the published update states corrected.
Limit: Follow the permanent record for the reason and subsequent corrections.
outcome: downgraded
A correction lowers the assessment while retaining publication.
Assignment: The shared correctionOutcome function maps the published update kind.
Illustration: the published update states downgraded.
Limit: Follow the permanent record for the reason and subsequent corrections.
outcome: withdrawn
The record is withdrawn.
Assignment: The shared correctionOutcome function maps the published update kind.
Illustration: the published update states withdrawn.
Limit: Follow the permanent record for the reason and subsequent corrections.