Airside Labs Aviation Tools

What happened to a suggestion or unmet-need report filed here? No API key is needed. Pass the `fingerprint` from a submit_suggestion or report_unmet_need receipt (12 hex characters). Returns `status`: `shipped` (a person acted on it and the change is live, with what shipped and when), `declined` (read and not taken up, with the reason), `acknowledged` (read, still open) or `open` (recorded, not yet reviewed), or `unknown_fingerprint` if nothing was ever filed under it. `times_reported` says how many times that exact text has been filed, and `first_reported` / `last_reported` when. Resolutions are written by a person when something ships, so `open` means exactly that: it has not been read yet, or has been read and not written up. Filing again does not move it. Do NOT infer from `shipped` that the whole class of problem is fixed: the note says what was changed and the dataset build it landed in.

feedback_status

Which airframes have been heard operating this airline flight, how often and when -- from Airside Labs' own ADS-B receiver? Accepts a flight in any common form: BA49, BAW49, "BA 49", VS105. Returns each airframe heard broadcasting that Flight ID within the receiver's footprint, with its Mode-S address, registration and ICAO type (joined from the registrations dataset), how many receiver poll cycles it was in view for, over how many distinct days, and first and last heard. `most_seen` and `most_recent` name the likeliest next tail; `types_seen` lists every type that has flown it, and where there is exactly one the resolved type is embedded with its capacity block. Why this matters: the cabin, the seat map and the sub-fleet vary by REGISTRATION, not by type. Two A330-900s at one airline can be different aircraft inside. A seat-finding or fleet-watching agent needs the tail; this answer comes from what the receiver observed, not from a schedule. Coverage is the station's footprint: about 180 nm, which takes in Heathrow and Gatwick traffic and the westbound Heathrow departure flow. `not_heard` means exactly that -- the flight did not cross the footprint in the window, or did not fly -- and says nothing about whether the flight exists. This is what was observed, not a prediction: an airline swaps tails without notice, and the answer's confidence describes the strength of the observation (many sightings over many days), never the odds for a future date. Do NOT use it to assert what will operate a flight; say what has operated it.

flight_airframes

One real episode end to end: what happened, where and when (UTC), the actors, what it shows and what it must not be read as; the timeline, every event with its actor, source and the reference of the raw record it came from; and the lenses, each a use case's cut of the day with a small evidence table, an implication, and the use cases it illustrates with a reason for each. Use it when a brief needs the whole story rather than one finding: a worked example with a paper trail, an edge-case write-up for a data science team, a workshop narrative. For one use case's view, take the lens from get_use_case instead; it is smaller and carries the same caveats. `events=False` or `lenses=False` trims the reply. Every figure in a lens was measured by a probe over the named source and window and checked against its evidence before publication; quote it with the scenario id, the source and the window, and copy `not_to_be_read_as`. Do NOT extend a scenario past its window, treat its counts as rates, or read one airline's behaviour on one day as its policy. Do NOT call this for live status: nothing here is live, and the server never queries the feeds a scenario was cut from. An unknown id returns the known ids.

get_scenario

The full record for one use case, by id from a search or landscape result. Returns the use-case text; the role that owns it with its description; the organisation type (canonical and as the corpus names it); every data requirement -- title, description, nominal update rate with a normalised cadence class (real_time … annual), and the kind of system it typically comes from; the EASA screen (AI level, hazard class, confidence, rationale) with the framework rows it points at, page-cited; keywords; up to three `data_stories`; and the five nearest use cases by text similarity. A data story is what this use case's data does when you actually touch it: a measured finding from Airside Labs' own receiver, the FAA flight-plan feed, BTS traffic data or the served entity registry -- a fill rate, a reused identifier, an ambiguous timestamp, the reach of one sensor. Each carries `claim` (the finding, with its scope in the sentence), `implication` (what to do about it), `identifiers` (exact ids to test with), `evidence` (a small table), `sources` with editions and windows, `caveats` and `not_to_be_read_as`. Put the claim in the requirements section, the identifiers in the test plan, and cite the source and window. A use case with no data stories is undecorated, not unillustratable. Use this for the few worked examples an argument actually needs, chosen with search_use_cases, use_case_landscape or similar_use_cases first. Calls are counted against your key's daily allowance, so walking the catalogue record by record is the wrong tool: use the aggregates. The data requirements are what this use case *needs*, stated generically ("airport operational database", "ADS-B feed"), not what any particular airport has. A requirement with cadence_class `unknown` had an update rate the normaliser could not read; the raw `update_rate` is still there. Do NOT treat the EASA screen as a determination: it is a title-level assessment by Airside Labs with a confidence, made to sort a portfolio, and a real classification needs a ConOps. Cite the framework rows by their cp_ref if you quote them. Do NOT generalise a data story past its `applies_when`: one receiver, three US airports and one data edition are a witness, not a survey. Every response carries `provenance`: `corpus` is Airside Labs' proprietary catalogue (use it in your analysis; do not redistribute it as a dataset), `derived` is Airside Labs' assessment over it, `framework:easa` is public regulatory text you may quote with its cp_ref page, `knowledge` is authored message-type knowledge and `knowledge:workflow` is authored operational sequence structure, and `data_story` is a measured, dated finding from data Airside Labs holds (its own ADS-B receiver, FAA flight-plan and BTS traffic feeds) attached to the use cases it illustrates, and `scenario` is one real day reconstructed from those feeds with a traceable timeline. The EASA level and hazard on a use case are a title-level screen with a confidence, not a certification finding; a workflow is not an operating procedure; a data story is not live and not a statistic about the industry -- read its `applies_when` before quoting it.

get_use_case

Which real operational episodes does the catalogue carry, to illustrate use cases with? A scenario is one real day reconstructed from data Airside Labs holds and may publish: FAA SWIM flight plans, FAA NOTAMs, its own ADS-B receiver. Each is a UTC window at a place, a timeline whose every event names the raw record it came from, and a set of lenses -- one use case's cut of the day, with the actual rows, an implication, and the use cases it is attached to. This lists them with their window, place, event and lens counts and how many use cases they touch. Fetch one with get_scenario; the lenses also arrive inside get_use_case as `data_stories` of kind `lens`. Use it when a brief needs a worked example with real rows behind it: a PRD's assumptions section, a workshop narrative, an edge-case note for a data team. Do NOT read a scenario as a rate or as how an operator usually behaves: it is one day, one or two airports, dated. The list is short by design; an empty result means the dataset carries no scenarios, not that nothing happened.

scenarios

Split a flight designator into carrier and number, and identify the carrier. Accepts the forms that arrive in real message traffic: TK1979, BAW276, "U2 8341", BA02490, BA2490A. Returns the parsed carrier with its resolved airline, the flight number, any operational suffix, which scheme was used (IATA or ICAO), and both normalised forms. Two things this tool will not do. It will not tell you who operated the flight: a designator names the marketing carrier, and a codeshare is invisible in the string. And it will not invent a carrier for a bare flight number -- pass `context` and it will look for an identifier in that text, but the result is capped at 0.6 confidence and returned as a candidate, because inference is not identification. Pass `date` when parsing historical data: the carrier depends on it, since SN2103 was a Sabena flight in 1995 and a Brussels Airlines flight in 2020. A string that parses correctly but names no known airline returns `unresolved` with a note saying exactly that -- the string was read and nothing was identified, which are different achievements. Where a two-character designator matches nothing but its reverse does, the reverse is offered as an alternate and never as the answer. Reading `confidence`: 1.0 means an exact unique match on an unambiguous identifier for the as_of date. 0.7-0.99 means a unique match reached through normalisation, an alias or a historical record. 0.4-0.69 means the best of several plausible candidates and `alternates` is populated -- prefer asking the user over picking one. Below 0.4 is speculative: do not act on it. `status` is resolved, ambiguous or unresolved. An unresolved answer is a real result, not an error: it means this dataset cannot identify the thing, and inventing one would be worse. Every field in `best` has a citation in `provenance`. When the thing exists and this dataset could not identify it, report_unmet_need with gap_kind `unresolved` is how that gap gets prioritised -- report it, then tell the user you could not find it.

parse_flight_identifier

Tell us what you came looking for and could not get from these tools. Call this when you needed something aviation-related that this toolset did not give you. It is not an error channel and it is not a retry: it is how the next version of these tools learns what is missing. A need reported here is appended to this service's own feedback store, on the same server that answered you, and reviewed later by Airside Labs; nothing is sent to any other system. Use it when: - a tool returned `unresolved` and you believe the thing exists - a tool returned `ambiguous` and nothing available could break the tie - the entity resolved but a field you needed was null or absent - no tool here covers the question at all - a tool answered confidently and the answer looked wrong `gap_kind` must be one of: `unresolved`, `ambiguous`, `missing_field`, `no_tool`, `wrong_answer`, `no_use_case` (the use-case catalogue had nothing on the operational need you searched for). `sought` is the important field: say what you were trying to find, in your own words, as specifically as you can. "The wake turbulence category for A21N" is useful. "Aircraft data" is not. Do NOT use this instead of answering the user. Report the gap, then tell the user plainly that you could not find it — filing this does not get you an answer, now or in this conversation. Do NOT paste conversation history, personal data or customer identifiers into it. Send the shape of the need, not the payload: what field, about what kind of entity, for what purpose. Text is length-capped and the log is treated as sensitive, but the cheapest way to keep data out of it is not to send it. Returns a receipt: whether it was recorded, a fingerprint, and how many times this same need has been reported before — which tells you whether you have found something known or something new.

report_unmet_need

Which aircraft type does this designator or marketing name refer to? Accepts an ICAO type designator (A21N) or a marketing name people actually say ("A321neo", "Dash 8-400", "777-300ER", "Q400"). Returns the ICAO designator, manufacturer, model, engine count and type, aircraft class, and the ICAO wake turbulence category. A name that identifies a family rather than a variant -- "Dreamliner", "777X", "A330neo" -- returns `ambiguous` with the variants as alternates and a confidence in the 0.4-0.69 band. That is the correct answer to an imprecise question; do not collapse it to the first alternate. `wtc` (wake turbulence category) is stated for almost every type, from FAA Order JO 7360.1K, and is cited like any other field. Where it is null the document leaves it blank -- do NOT fill it in from your own knowledge if the caller needs it for separation or charging, say it is unavailable. IATA aircraft type codes are NOT held -- the three-character form used in schedules and booking systems, "77W" or "32N". IATA's list is licensed and is not reproduced here, so those return `unresolved` with a note saying so. If you are working from schedule data, convert to the ICAO designator before calling (B77W, A20N); if you cannot, report the mapping as unavailable rather than guessing it. `capacity`, where present, carries what a network or fleet planner needs from the TYPE: `seats_max_certified` (the type certificate's maximum), `seats_typical` (the band the manufacturer publishes, with the cabin configuration it assumes), `range_nm` (maximum range at typical payload, a band where variants share a designator) and `mtow_kg`. Every value was read from a type certificate data sheet or the manufacturer's own page and confirmed by a second check; each is cited. The exact seat count of any airframe is the operator's configuration and is NOT held -- do not present the typical band as a particular aircraft's layout. For US markets, frequency, seats flown and load factor by route and type ARE held -- use route_capacity and airport_fleet_mix (BTS T-100, about three months in arrears). Yield is not held anywhere: supply it to build a revenue model. `capacity` is null where no confirmed record exists yet, and a null field inside it means the value was researched but not confirmed, and the notes say so. Do NOT use this to determine what type operated a particular flight; use resolve_registration for a specific airframe. Reading `confidence`: 1.0 means an exact unique match on an unambiguous identifier for the as_of date. 0.7-0.99 means a unique match reached through normalisation, an alias or a historical record. 0.4-0.69 means the best of several plausible candidates and `alternates` is populated -- prefer asking the user over picking one. Below 0.4 is speculative: do not act on it. `status` is resolved, ambiguous or unresolved. An unresolved answer is a real result, not an error: it means this dataset cannot identify the thing, and inventing one would be worse. Every field in `best` has a citation in `provenance`. When the thing exists and this dataset could not identify it, report_unmet_need with gap_kind `unresolved` is how that gap gets prioritised -- report it, then tell the user you could not find it.

resolve_aircraft_type

Which airline does this designator, name or callsign refer to on a date? Accepts an IATA two-character designator, an ICAO three-letter designator, an airline name or a callsign. Returns the operator with both designators, country, operational status and, where an airline ceased, its successor. `as_of` matters more here than for any other tool. IATA two-character designators are heavily reused: SN was Sabena until 2001 and has been Brussels Airlines since 2007, so "SN" without a date is a question with two answers. Where a date falls between two holders the tool returns `unresolved` with both as alternates rather than guessing. ICAO three-letter designators are not recycled the same way and are the safer identifier to carry through a pipeline. `status` is reported as at the date asked about, not as at today: asked about 2005, Northwest Airlines is `active`, with a note that it merged into Delta in 2010. A status of `unknown` means no reliable source states it -- the bulk open data flags long-dead airlines as active, so it is not trusted. Unknown is not evidence of ceasing. Do NOT use this for fleet lists, routes, schedules, alliance membership or financial status. Reading `confidence`: 1.0 means an exact unique match on an unambiguous identifier for the as_of date. 0.7-0.99 means a unique match reached through normalisation, an alias or a historical record. 0.4-0.69 means the best of several plausible candidates and `alternates` is populated -- prefer asking the user over picking one. Below 0.4 is speculative: do not act on it. `status` is resolved, ambiguous or unresolved. An unresolved answer is a real result, not an error: it means this dataset cannot identify the thing, and inventing one would be worse. Every field in `best` has a citation in `provenance`. When the thing exists and this dataset could not identify it, report_unmet_need with gap_kind `unresolved` is how that gap gets prioritised -- report it, then tell the user you could not find it.

resolve_airline

Which aerodrome does this airport code or name refer to on a given date? Accepts an IATA three-letter code, an ICAO four-letter code, an airport name fragment or a city name. Returns the aerodrome with its codes, location, elevation, IANA timezone and type (large, medium, small, heliport or closed). Pass `as_of` whenever the question concerns a past date. Airport codes move between aerodromes: ATH meant Ellinikon until 2001 and Athens International after it; HKG meant Kai Tak until 1998. Without a date these questions get today's answer, which is silently wrong for historical data. `timezone` is always an IANA identifier such as Europe/London, never a UTC offset, because an offset cannot express daylight saving. Do NOT use this for live operational status, runway or stand data, slots, or whether an airport is currently accepting traffic. It answers identity questions only. A name fragment matching several aerodromes returns `ambiguous` with candidates rather than picking one; use `country_hint` (ISO 3166 alpha-2) to narrow it. Reading `confidence`: 1.0 means an exact unique match on an unambiguous identifier for the as_of date. 0.7-0.99 means a unique match reached through normalisation, an alias or a historical record. 0.4-0.69 means the best of several plausible candidates and `alternates` is populated -- prefer asking the user over picking one. Below 0.4 is speculative: do not act on it. `status` is resolved, ambiguous or unresolved. An unresolved answer is a real result, not an error: it means this dataset cannot identify the thing, and inventing one would be worse. Every field in `best` has a citation in `provenance`. When the thing exists and this dataset could not identify it, report_unmet_need with gap_kind `unresolved` is how that gap gets prioritised -- report it, then tell the user you could not find it.

resolve_airport

Which airframe wore this registration on a given date? Accepts a tail number in any common form: VH-OQA, VHOQA, lowercase, padded with whitespace. Returns the aircraft with its country of registry, ICAO type designator, an embedded resolved aircraft type, operator, owner, serial number and Mode-S hex address. A registration is a slot rather than a permanent name: marks are surrendered and reissued. N803AL was one airframe from 1987 to 1993 and a different one, a 787-8, from 2015; N264CP moved to a different Mode-S address in 2015. Pass `as_of` for any historical question. Where the date falls between two holders the tool returns `unresolved`, reports the country from the nationality mark, and offers both airframes as alternates. When the tail is unknown but the nationality mark is recognised, the response carries a `partial` object with the country of registry at confidence around 0.5. That is a partial answer, not a resolution: the aircraft is still unidentified. Coverage is not global. An unresolved registration means this dataset does not hold it, NOT that the aircraft does not exist. Do NOT infer that an aircraft is fictitious from an unresolved result, and do NOT use this for airworthiness, lease status or current position. Reading `confidence`: 1.0 means an exact unique match on an unambiguous identifier for the as_of date. 0.7-0.99 means a unique match reached through normalisation, an alias or a historical record. 0.4-0.69 means the best of several plausible candidates and `alternates` is populated -- prefer asking the user over picking one. Below 0.4 is speculative: do not act on it. `status` is resolved, ambiguous or unresolved. An unresolved answer is a real result, not an error: it means this dataset cannot identify the thing, and inventing one would be worse. Every field in `best` has a citation in `provenance`. When the thing exists and this dataset could not identify it, report_unmet_need with gap_kind `unresolved` is how that gap gets prioritised -- report it, then tell the user you could not find it.

resolve_registration

How much scheduled capacity is flown nonstop from one airport to another, by whom and on what, over the last N months of BTS T-100 data? Returns monthly departures performed, seats and passengers with the load factor, and the same broken down by carrier and aircraft type with seats per departure and the sector distance (BTS statute miles, and nautical miles converted). The pair is DIRECTIONAL: SEA to LHR is not LHR to SEA; ask both if you need the round trip. `months` defaults to the last 12 of the window; pass null for the whole window. `scheduled_only` (default true) restricts to scheduled passenger service, BTS class F -- charter and all-cargo rows exist in the source and would distort seats and load factors. This is the frequency, capacity and load-factor side of a route model. Combine with resolve_aircraft_type's capacity block for the type's certified and typical seating, and supply yield yourself: fares are not held here. Airports may be given as an IATA code, an ICAO code or a name. T-100 is filed under three-character codes that follow IATA, so an ICAO code or a name is resolved to the airport's IATA code first and the answer says so in its first note; `query` echoes the code that was looked up. Reading the answer: `status` is resolved, no_service or unresolved. `unresolved` means nothing was looked up (no airport matches, or it has no IATA code) and is never a statement about service. `totals` and `monthly` are sums of BTS-reported departures performed, seats and passengers; `load_factor` is passengers over seats. `provenance.edition` is the latest month held and `provenance.window` the months summed -- BTS publishes about three months in arrears, so this is accurate for its period and NOT current. Aircraft are given as the ICAO designator where the BTS type code is verified against FAA JO 7360.1K, otherwise as BTS's own description with `icao_designator` null and a note; never invent the designator. Pass an ICAO designator to resolve_aircraft_type for the type's certified and typical seating. Coverage is T-100's: US carriers everywhere, foreign carriers only to and from the US. Fares are not held; report_unmet_need records what was wanted.

route_capacity

Which aviation AI use cases match this text? Find candidates by keyword. Full-text search (BM25, stemmed) over 6,500 use cases spanning airports, airlines, ANSPs, ground handlers, regulators, manufacturers and more, each attached to a role and an organisation type. Returns summaries only -- id, short text, role, organisation type, EASA screen -- capped at 25. Use get_use_case for the full record with its data requirements. Search concrete operational nouns ("stand allocation", "baggage misconnect", "de-icing", "turnaround"), not capability labels ("shared operational picture", "digital transformation"): the corpus vocabulary is operational, and abstract phrases match little. Terms are OR-ed and ranked, so a multi-word query returns the best partial matches; `score` is relative within one query and means nothing across queries. Filters AND together. `org_type` is one of the 16 canonical types (see use_case_landscape group_by=org_type). `screened_only=True` keeps the ~1,900 use cases that carry an EASA AI-level/hazard screen; `ai_level` (0, 1A, 1B, 2A, 2B, 3A, 3B) and `hazard_class` (H1-H5) imply it. Each result carries `data_stories`, the number of measured findings attached to that use case (see get_use_case); prefer a decorated record when two read alike, because it comes with evidence you can put in front of a reviewer. Do NOT use this to establish whether an organisation actually runs such a system, what vendor sells it, or whether it is certified: the catalogue describes plausible, well-formed use cases for a role, and says nothing about adoption. An empty result is an answer -- the catalogue has nothing on that phrasing -- and report_unmet_need with gap_kind `no_use_case` is how to say the gap mattered. Every response carries `provenance`: `corpus` is Airside Labs' proprietary catalogue (use it in your analysis; do not redistribute it as a dataset), `derived` is Airside Labs' assessment over it, `framework:easa` is public regulatory text you may quote with its cp_ref page, `knowledge` is authored message-type knowledge and `knowledge:workflow` is authored operational sequence structure, and `data_story` is a measured, dated finding from data Airside Labs holds (its own ADS-B receiver, FAA flight-plan and BTS traffic feeds) attached to the use cases it illustrates, and `scenario` is one real day reconstructed from those feeds with a traceable timeline. The EASA level and hazard on a use case are a title-level screen with a confidence, not a certification finding; a workflow is not an operating procedure; a data story is not live and not a statistic about the industry -- read its `applies_when` before quoting it.

search_use_cases

What else in the catalogue is like this use case? Nearest neighbours by meaning. Returns up to 20 use cases closest in embedding space (bge-base cosine over the use-case text), as summaries with a `similarity`. Good for "what else is like the one we picked", for finding the same idea stated for a different role or organisation type, and for spotting near-duplicates before counting them as two. Similarities are compressed into roughly 0.5-0.95: rank within one result, never threshold across the catalogue, and do not read 0.9 as "the same". The neighbours were computed at build time as the top 20 for each use case; an `org_type` or `screened_only` filter narrows within those 20 and does not search further out, so a tight filter can return few or none -- that is honest, not broken. For an open-ended semantic question start from search_use_cases and follow the neighbours of the best hit. Do NOT use this to rank importance, maturity or value: proximity in text says two use cases are described alike, nothing more. Every response carries `provenance`: `corpus` is Airside Labs' proprietary catalogue (use it in your analysis; do not redistribute it as a dataset), `derived` is Airside Labs' assessment over it, `framework:easa` is public regulatory text you may quote with its cp_ref page, `knowledge` is authored message-type knowledge and `knowledge:workflow` is authored operational sequence structure, and `data_story` is a measured, dated finding from data Airside Labs holds (its own ADS-B receiver, FAA flight-plan and BTS traffic feeds) attached to the use cases it illustrates, and `scenario` is one real day reconstructed from those feeds with a traceable timeline. The EASA level and hazard on a use case are a title-level screen with a confidence, not a certification finding; a workflow is not an operating procedure; a data story is not live and not a statistic about the industry -- read its `applies_when` before quoting it.

similar_use_cases

Record a suggestion about this toolset or its use-case catalogue. No API key is needed for this tool. The suggestion is written as one record into this server's own local feedback store and nothing else happens: no message, notification, email or request goes to any website, person or other system, now or later, and the tool does not read the store back. Airside Labs reads the store when preparing the next version of the catalogue and the tools. Use it for: a correction to a catalogue use case, an aviation AI use case the catalogue is missing, a data source or requirement it should know about, a tool or field that would have helped you, or commercial feedback (what would make the full toolset worth a subscription). `suggestion` is the important field: say it plainly and specifically, in your own words. `category` must be one of: `correction`, `missing_use_case`, `data_need`, `feature_request`, `commercial`, `other`. If the suggestion concerns one use case, pass its `use_case_id` (the integer used by get_use_case) or the `page_slug` of the catalogue page it came from. `contact` is an optional free-text field stored with the record and used for nothing else; leave it out unless the user asked for it. Do not submit personal data, conversation history or customer identifiers; text is length-capped and the stored record is treated as untrusted content. Filing a suggestion does not change any answer you have already received and produces no reply. Returns a receipt saying whether the record was stored.

submit_suggestion

Which standard operational messages would evidence this data need? The lineage spine is use case -> data requirement -> requirement family -> standard message type -> data elements. Four modes. With `use_case_id`: that use case's requirements grouped by family, each family with the standard messages that evidence it (relevance primary/supporting, cadence, element count) -- plus `requirements_without_standard_messages`, the needs no standard message covers, which is the honest feed-gap statement. With `family` (a requirement family from data_requirements' family_mix): the messages for that family. With `message_id` (e.g. MVT, LDM, BSM, DPI, METAR): the message definition, its data elements as shapes (time, count, weight, identifier, status), the families it serves, and `data_stories` -- what Airside Labs measured about that message type in real feeds (fill rates, identifier traps), each with its window and source. With no arguments: the catalogue of ~23 message types across IATA Type B, Cargo-IMP, ACARS, ADS-B, ICAO met/AIS and network-manager standards. Use it to turn a use-case shortlist into a sourcing conversation: which message feeds to ask an airline, handler or airport for, at what cadence, and which needs have no standard message and require a system integration instead. Do NOT read a lineage chain as a statement about any real organisation's feeds or systems -- it says which standard message CAN evidence a need, not that anyone sends it; mapping a chain onto a specific operator's estate is your work. Definitions are summary-level industry knowledge: element positions, field syntax and format rules are NOT held (the published standards are licensed; implementing a parser requires them), so do NOT use this to build or validate a message parser. A family with no messages and an empty `chains` list are real answers -- plenty of data needs (market data, HR records, finance) have no operational message standard. Every response carries `provenance`: `corpus` is Airside Labs' proprietary catalogue (use it in your analysis; do not redistribute it as a dataset), `derived` is Airside Labs' assessment over it, `framework:easa` is public regulatory text you may quote with its cp_ref page, `knowledge` is authored message-type knowledge and `knowledge:workflow` is authored operational sequence structure, and `data_story` is a measured, dated finding from data Airside Labs holds (its own ADS-B receiver, FAA flight-plan and BTS traffic feeds) attached to the use cases it illustrates, and `scenario` is one real day reconstructed from those feeds with a traceable timeline. The EASA level and hazard on a use case are a title-level screen with a confidence, not a certification finding; a workflow is not an operating procedure; a data story is not live and not a statistic about the industry -- read its `applies_when` before quoting it.

trace_data_lineage

What happens before and after this use case, and who hands off to whom? The catalogue is otherwise flat -- role, use case, data requirement -- with no edges. This is the edges: named operational sequences with their steps in order, the role at each step, the `gate` that must hold before the next one may start, and what passes between roles at each handoff (headset, radio, hand signal, system, visual, document, verbal, formal correspondence). Three modes. With `use_case_id`: every workflow that places that use case, its position, and the steps immediately `preceded_by` and `followed_by` it with what transfers. That is the real operational dependency -- distinct from `similar_use_cases`, which is text similarity, and from `data_requirements(query=...)`, which infers a relationship from two use cases sharing a feed. A gate is a dependency; a shared feed is a correlation. With `workflow_id`: the whole sequence end to end. With `phase` (arrival, turnaround, departure, abnormal, oversight, certification, crew qualification, planning, mission) or nothing: the catalogue of workflows with their step and handoff counts. Use it to answer "if this prediction improved, what downstream step benefits and who would have to act on it", to see which roles a change touches, and to find the abnormal branches -- communication failure, coupling break-off, spillage, a ramp finding serious enough to stop the aircraft -- which are sequences in their own right and where the costly failures live. Some sequences cross an organisation: a finding travels from an authority to an operator and back, and those edges carry the channel `formal correspondence`. Coverage is ground handling, regulatory oversight and certification, crew qualification and rostering, and specialised missions -- not the whole catalogue. A use case with an empty `placed_in` is outside the covered sequences, which is a statement about coverage and not about the use case, and some are unplaced because they are standing monitoring states rather than sequences. A step whose `role_in_catalogue` is false is a real participant the catalogue holds no use cases for -- the flight deck on a turnaround, the officer who signs a certificate -- not a data error. The sequences are Airside Labs' derived structure: they say how this work is ordered in the industry, NOT that any particular operator runs it this way, and they are not an operating procedure. Do NOT use one as a checklist to work to; the authority for that is the operator's own manual. Every response carries `provenance`: `corpus` is Airside Labs' proprietary catalogue (use it in your analysis; do not redistribute it as a dataset), `derived` is Airside Labs' assessment over it, `framework:easa` is public regulatory text you may quote with its cp_ref page, `knowledge` is authored message-type knowledge and `knowledge:workflow` is authored operational sequence structure, and `data_story` is a measured, dated finding from data Airside Labs holds (its own ADS-B receiver, FAA flight-plan and BTS traffic feeds) attached to the use cases it illustrates, and `scenario` is one real day reconstructed from those feeds with a traceable timeline. The EASA level and hazard on a use case are a title-level screen with a confidence, not a certification finding; a workflow is not an operating procedure; a data story is not live and not a statistic about the industry -- read its `applies_when` before quoting it.

trace_workflow

How many use cases are there, sliced one way? Counts you can quote. `group_by` is one of org_type (16 canonical organisation types: Airport, Airline, Air Navigation Service Provider, Ground Handling, Regulator / Authority, Aerospace Manufacturing, MRO / Maintenance …), sector, role (500+ roles), ai_level, hazard_class, or cadence_class (counts data requirements rather than use cases). Filters AND together and apply before grouping, so group_by=role with org_type=Airport lists airport roles by how many use cases each carries. Also returns the total in scope and how many of those carry an EASA screen. This is the tool to call first: it tells you what the catalogue covers before you search it, and the exact spellings that the filters accept. Grouping the whole catalogue by ai_level or hazard_class shows a large null bucket -- only the ~1,900 airport-operations use cases were screened; pass screened_only=True to see the screened distribution alone. Counts describe the catalogue, not the industry: a type with many use cases is a type the corpus describes in detail, not one that adopts more AI. Do NOT present a count as market evidence. Every response carries `provenance`: `corpus` is Airside Labs' proprietary catalogue (use it in your analysis; do not redistribute it as a dataset), `derived` is Airside Labs' assessment over it, `framework:easa` is public regulatory text you may quote with its cp_ref page, `knowledge` is authored message-type knowledge and `knowledge:workflow` is authored operational sequence structure, and `data_story` is a measured, dated finding from data Airside Labs holds (its own ADS-B receiver, FAA flight-plan and BTS traffic feeds) attached to the use cases it illustrates, and `scenario` is one real day reconstructed from those feeds with a traceable timeline. The EASA level and hazard on a use case are a title-level screen with a confidence, not a certification finding; a workflow is not an operating procedure; a data story is not live and not a statistic about the industry -- read its `applies_when` before quoting it.

use_case_landscape

Do these aviation identifiers describe the same thing on this date? Give any combination of a registration (tail number), a Mode-S 24-bit address, an ICAO type designator, an operator name, an airline designator (IATA or ICAO), a callsign and a flight designator, plus `as_of`. Each pair that can be checked is checked, and every check comes back with a verdict -- `consistent`, `contradicted` or `unverifiable` -- the detail, and the rule that decided it. `contradictions` lists the failures on their own so an agent can act on them without reading everything. The top-level verdict is `consistent` ONLY when every check verified. When some checks passed but others could not be checked it is `consistent_where_checkable` -- a different answer, because an absent record silences exactly the checks that would catch a false claim about that airframe. `unverifiable_count` says how many checks were silent; treat anything above zero as partial coverage, not a pass. Use it before acting on identifiers assembled from more than one message: a movement message's registration against a surveillance track's address, a schedule's flight number against the operating airline, a type in a load message against the airframe's record. One contradiction is a contradiction; the top-level `verdict` never averages it away against the checks that passed. What the rules are: the registration record on the date (address, type, operator); a *previous* holder of a re-issued mark is named as such rather than reported as a mismatch; the FAA allocation arithmetic for N-numbers (an address decodes to exactly one N-number, so a wrong pairing is provable without any second source); a military or government airframe is said, not judged; the airline's callsign against its record and the FAA contractions order; the carrier inside the flight designator against the airline given; and whether the type designator exists at all. `unverifiable` means this dataset cannot say -- there is no record, or the record lacks the field -- and is a different answer from `consistent`. Do NOT read it as a pass. It does not check hex-to-country for non-US addresses, performance plausibility (range, block time) or anything live, and it does not resolve identifiers you did not give it: call the resolve_* tools for that. Pass `as_of` for any historical question; marks and designators are reused and an undated check is silently wrong for past data.

validate_identifiers