Skip to content

Systems and records

Hotel group booking changes dataset: one block change, many rooms

By SourceX Editorial · Reviewed by Noah Loul ·

10 sections

Short answer

Treat a group booking change as one event with every dependent reservation attached: freeze the block grid and each linked reservation's state, rate and dates, record the contract terms, shares and approvals that limit the change, capture the decision, and grade the result against an expected end state for every record. A hotel group booking changes dataset pairs each case with a counterexample where the change should not cascade.

License terms

Hotel group booking changes

This page covers hotel group booking changes dataset: one block change, many rooms.Illustrative

Key takeaways

  • The unit of record is the change event with the block and every linked reservation, not one row per reservation.
  • Block state, cutoff or release position, inherited terms, shares and billing routes decide which reservations may follow a change.
  • Grade the end state of every linked record against an expected state; a Confirmed status or a polite reply proves little.
  • Pair each cascading case with a counterexample where the same request should touch fewer records.
  • Guest identifiers, contract figures and brand terms are separate clearances from access to the PMS.

What should a hotel group booking changes dataset capture when one change touches many reservations?

It should capture the change as one event with every dependent reservation attached: the block header and room-night grid before and after, each linked reservation's state, rate and dates before and after, the contract terms and approvals that limited the change, and a check that proves every record ended where it should. A row per reservation with a final status loses the thing that makes group work hard, which is that one edit can quietly break twenty others.

Group business sits in two layers that most property systems keep apart. The block holds an allocation of rooms by date and room type for a company, planner or wedding party; the individual reservations draw down that allocation as guests are named on the rooming list or book themselves. When the planner shortens the program by a night, the block grid changes, but whether each reservation follows depends on how it was created, what it was sold at and who is sharing it.

That gap between the block-level request and the reservation-level result is the training signal. A model that reads the planner's email and edits the block has done half the job; the dataset has to show the other half and grade it.

Why a room block behaves differently from a large reservation

A block is an inventory container with its own lifecycle, not a booking with a big room count. The Mews Connector API documentation describes availability blocks with distinct states: a Confirmed block deducts availability and can have reservations assigned, an Optional block deducts availability but cannot take reservations, and Inquired and Canceled blocks deduct nothing. A training case has to record which state the block was in when the change arrived, because the same request means different work in each.

Release rules add a clock. Mews documents three release strategies for a block (a fixed release time, a rolling release at a set offset from each day, or no automatic release), and Oracle's OPERA Cloud help describes cutoff as returning rooms not picked up to house inventory on or before the cutoff date. Oracle notes that a cutoff cannot be undone; rooms come back only by increasing the block allocation again. A change requested the day after cutoff therefore starts from a different inventory position than the same request a week earlier.

Capacity is enforced at the block, not the hotel. The Mews events guide says reservations added with the block identifier consume the block's allocation instead of general availability, and that a block's capacity cannot currently be exceeded; to fit more reservations, the allocation has to be increased first. It also notes that reservations can have arrival or departure times extending beyond the block's interval, which is exactly how an extended stay ends up half inside and half outside the group.

Which block setting, sold rate, share or approval stops a reservation from following the group?

Four kinds of limit decide what may change: the block's own state and release position, the terms each reservation inherited or was sold under, links between reservations such as shares and billing routes, and the authority of the person approving. Record each as a named constraint with the field it lives in, so a grader can say which limit an agent ignored rather than only that the end state was wrong.

Inheritance matters more than most teams expect. Oracle's OPERA Cloud help describes the block's posting master reservation as a template: reservations entered on the rooming list take rate code, market and source codes, routing and payment instructions, packages and other attributes from it. A reservation created that way behaves differently under a block change from one a guest booked separately and later attached to the group.

Which block setting, sold rate, share or approval stops a reservation from following the group?
ConstraintWhere it usually livesWhat it limitsHow to record it in the case
Block state and allocationBlock header and room-night gridWhether rooms can be added, moved or must come from house inventorySnapshot of state and grid by date and room type at the request time
Cutoff or release positionBlock release strategy, cutoff date or scheduleWhether unpicked rooms still sit in the block or were returnedRelease setting plus the date the request arrived relative to it
Inherited reservation termsTemplate or posting master reservation, block rateWhich rate, packages and routing a moved reservation keepsPer-reservation flag: created from rooming list or attached later
Individually sold termsReservation rate plan and deposit or cancellation ruleWhether a guest's own booking may be shortened or repricedRate plan identifier and the policy text in force at purchase
Shares and companionsShare links, companion or roommate recordsWhether one reservation can move without the otherList of linked reservation identifiers that must move together
Billing routesRouting instructions, master or group billWhich charges follow the group and which stay with the guestRouting rules before and after, by charge type
Contract termsSigned group agreement and amendmentsAttrition, cutoff and room-night commitmentsClause reference and version, never the full contract text
Approval authorityUser role, sales or revenue manager sign-offWho may release rooms or override ratesApprover role and timestamp from the change log

What the group and individual booking consistency task set contains

Each task in the set is one group change with its frozen starting state, the request as it arrived, the constraints above, the actions an agent may take, the decision a person actually made and the expected end state for every linked record. The fields below are written so a hotel team can copy them as a packet schema and rename them to match its own property system.

At a glance

What the group and individual booking consistency task set contains

Provenance labels keep the set honest. Mark every field as recorded (read from a system), reconstructed (rebuilt from history or a log after the fact) or illustrative (invented for teaching). A reviewer should be able to see at a glance which parts of a task came straight from the PMS and which were inferred.

What the group and individual booking consistency task set contains
FieldWhat it holdsTypical sourceProvenance label
Task and block identifiersTask id, block id, property id (pseudonymous)Block headerRecorded
Starting snapshotBlock state, grid by date and room type, pickup count, release settingBlock grid and history at request timeRecorded or reconstructed
Linked reservationsEach reservation's state, dates, room type, rate plan, creation path, sharesReservation records filtered by block idRecorded
Change requestWhat the planner or guest asked for, with timestamp and channelEmail, sales note or event system changeRecorded, de-identified
ConstraintsNamed limits from the constraint table with field referencesBlock, reservation, contract and role dataRecorded or reconstructed
Permitted actionsEdits the agent may make and edits that need approvalProperty policy and user rolesRecorded
Human decisionWhat was changed, what was left alone, approver roleChange log and block historyRecorded
Expected end stateTarget values for the block and every linked reservationPost-change records, reviewedReconstructed
Consistency checksAssertions the end state must passWritten by the reviewerEditorial
Counterexample linkA paired task where the same request should not cascadeAnother block or a modified copyRecorded or illustrative

How do you check that a group change actually landed on every reservation?

Compare the end state of the block and each linked reservation with the expected state, field by field, rather than reading the agent's reply. Benchmarks for tool-using agents already work this way: the tau-bench code computes its reward by comparing a hash of the environment's data after the agent acts with the data produced by replaying the ground-truth actions, and checks that required outputs appear in the agent's responses.

The airline policy in the same benchmark shows why written rules belong in the task: it states that the API does not check certain change rules for the agent, so the agent must make sure they apply before calling it, and that the agent must list the action details and get explicit confirmation before any booking update. Hotel group work has the same shape, because a PMS will often accept an edit that the contract or policy forbids.

  • Pickup never exceeds allocation on any date and room type after the change, unless the case records an approved increase.
  • Every reservation that was inside the removed night is either shortened, moved to a house rate with an approval, or listed as an approved exception.
  • Shared reservations moved together: no sharer left on a night the primary no longer occupies.
  • Reservations created from the rooming list kept the block rate; reservations sold separately kept their own rate plan unless an approver changed it.
  • Billing routes still send group charges to the group bill and incidentals to the guest after the dates move.
  • Rooms released from the block are back in house inventory, and the release is visible in block history rather than only in a note.
  • Any cancellation carries a reason, and no reservation lost its block link without a recorded decision.

Worked case (illustrative): a conference program loses its first night

Everything in this case is illustrative: the counts and timings are invented to show the task set working and describe no real property or group. A hotel holds a confirmed block for an association meeting over three nights with 20 reservations picked up. Fourteen came from the planner's rooming list and inherited the block rate and a group billing route; 4 of those are two pairs of shared rooms. Six guests booked through the group link themselves, and 2 of the 6 added a night after the program on their own rate plan.

Nine days before arrival, the planner emails that the opening reception has moved and the first night is no longer needed. The block uses a fixed cutoff that falls in 5 days, so unpicked rooms are still inside the allocation. The property's policy, as written into the case, lets a reservations agent shorten group-rate stays but requires a sales manager to approve any change to a guest's separately sold nights.

The human decision recorded in the change log was: reduce the block grid for night one to zero, shorten the 14 rooming-list reservations by one night, move both shared pairs together, leave the 2 post-program extensions untouched, and email the 6 self-booked guests asking whether to drop the first night, because their bookings were sold to them individually. Two of the six declined to change, so those two reservations keep night one at a house rate after the sales manager approved it.

The expected end state therefore has 18 reservations starting on night two, 2 starting on night one outside the block, a block grid of zero for night one, and unchanged extension nights. An agent that shortens all 20 reservations passes a naive status check (every reservation still Confirmed) and fails three consistency checks: two guests lost nights they did not give up, a sold rate was changed without approval, and the billing route on those two nights no longer matches the guest's own terms.

Counterexamples: when the same request must not cascade

Pair every cascading case with one where the right answer is to touch fewer records. Without counterexamples, a model learns that a group change always means editing every reservation that carries the block identifier, which is the most common way automated group handling goes wrong.

  • The block is Optional or Inquired, so there are no assigned reservations to move and the change is to the grid alone.
  • Cutoff has already released the unpicked rooms; shrinking the grid changes nothing, and the real question is whether the remaining named reservations agreed to the change.
  • A guest attached to the block later, on a non-refundable rate bought before the group existed; the group request does not authorize changing it.
  • One member of a shared pair has already checked in on a split stay; moving the pair together is no longer possible, so the case expects an escalation.
  • The planner's request conflicts with the signed agreement's room-night commitment; the expected action is to route the request to sales, not to edit inventory.

Which block and reservation history proves what was decided and what happened?

The proof is a chain, not a field: the request record, the block history showing the grid change and who made it, each reservation's change history, the billing route before and after, and the stay or cancellation that followed. A status reading Confirmed or Canceled proves very little on its own, since a reservation can be Confirmed on the wrong dates or Canceled with no reason.

Mews documents the reservation fields that make this chain joinable: each reservation carries a reservation group identifier and, when it draws on a block, an availability block identifier, plus an origin and, for cancellations, a reason from a fixed list. OPERA Cloud exposes comparable block and reservation history. Whatever the system, keep the identifiers that link reservations to the block and to each other, replace them with consistent pseudonyms, and never break the joins during de-identification.

Guest names, planner contracts and brand terms to settle before cases leave the property

Three separate questions need separate answers before any group case is shared: whether the property owner approves the use, whether guest and planner identifiers can be removed without breaking the links between records, and whether the group agreement, management agreement and any franchise or channel terms allow booking records to be used this way. Access to the PMS is not permission to license what is in it, and removing names does not by itself create a licensing right.

Rooming lists are dense with personal data: names, sometimes arrival times and companions, occasionally accessibility or dietary notes. Strip them to pseudonymous guest keys, drop free-text notes unless a reviewer has cleared them, and keep only the fields the consistency checks use. Group contracts usually carry commercial terms the planner expects to stay confidential; record clause references and outcomes, not the text or the rates.

The Events Industry Council's APEX contracts report describes an attrition clause as defining the group's liability for falling short of its minimum commitments, and says the contract should state the cutoff date and what happens to rooms not reserved within the block by then. Those concepts belong in the task as constraints. The counterparty's actual figures do not, unless the agreement allows sharing them; counsel reviews that question for the specific contracts.

Licensing group booking change histories through SourceX

Years of block changes, pickups and released rooms are a record of hard operational judgment, and few properties have ever looked at them as data. SourceX licenses records of this kind for hotel businesses as AI datasets, taking on the rights review, contracting, preparation, delivery and payment that surround a deal.

To learn whether it qualifies, a hotel or management company completes a brief company assessment and uploads nothing. If it proceeds, it approves the scope and terms and signs before records are delivered.

Frequently asked questions

What should the group and individual booking consistency task set include?

Each task needs the frozen block and reservation state at the time of the request, the request itself, the constraints that bind the change, the actions an agent may take, the decision a person made and the expected end state for every linked record. Add consistency checks that test the end state and a paired counterexample where the request should not cascade.

Is a hotel group booking changes dataset the same as a reservation history export?

No. A history export lists reservations and their latest status, while a group change dataset keeps the block, every linked reservation and their before and after states together as one event. Without that grouping a model cannot learn which reservations a block change was supposed to touch and which it was not.

Why record the block's cutoff or release setting in every case?

Because it decides where the unpicked rooms are when the change arrives. Oracle's OPERA Cloud help says a cutoff returns unpicked rooms to house inventory and cannot be undone, so the same request before and after cutoff calls for different actions.

Can an agent be graded on the confirmation email it sends to the planner?

Only partly. The message should contain the facts the planner must hear, but the grade should come from the block grid, each reservation's dates and rate, the shares and the billing routes after the change, compared with an expected state.

Which records should never go into a group change case?

Payment card data, free-text guest notes that have not been reviewed, accessibility or health details, and the commercial terms of the group contract unless the agreement allows sharing them. Keep pseudonymous identifiers so the links between reservations survive de-identification.

Sources

  • Mews Connector API documents availability block states (Confirmed, Optional, Inquired, Canceled), release strategies (fixed, rolling, none) and pickup distribution. Source
  • The Mews events guide says reservations added with an availability block identifier consume the block's allocation, block capacity cannot currently be exceeded, and reservations may extend beyond the block interval. Source
  • Mews reservations carry a reservation group identifier, an optional availability block identifier, an origin and a cancellation reason from a fixed list. Source
  • Oracle OPERA Cloud help: a block cutoff releases rooms not picked up on or before the cutoff date, and a cutoff cannot be undone; rooms are added back by updating the block allocation. Source
  • Oracle OPERA Cloud help: the block's posting master reservation serves as a template, and rooming-list reservations inherit attributes such as rate codes, market and source codes, routing and payment instructions and packages. Source
  • The Events Industry Council APEX contracts report describes an attrition clause as defining the group's liability for failing to meet minimum commitments, and says contracts should state the cutoff date and what happens to rooms not reserved by then. Source
  • The tau-bench environment computes reward by comparing a hash of the data after the agent acts with the data after replaying ground-truth actions, and checks required outputs in responses. Source
  • The tau-bench airline policy says the API does not check some change rules for the agent and requires listing action details and getting explicit user confirmation before booking updates. Source

See if your company qualifies

A short company assessment. No data uploads are needed.

See if your company qualifies