Inventory Count Policy and Count Settings

This page lists every setting that decides when a location is counted, how it is counted, and what happens to the difference.

Paths in the "Affects" lines are relative to crates/libs/ unless they start with another crate name.

1. How counting works in one page

• A count has one task per location. Each task is one location, one attempt.
• The count never changes stock. It records what is there. An accepted difference becomes an inventory adjustment, and the adjustment changes the stock.
• A location in a count is frozen (the location flag CycleCountActive) when the freeze is Hard. Movement, allocation, putaway and pick dispatch respect the freeze.
• Every task is also a Task Manager task of type Count. An operator needs a Count right in a work area (22-work-areas-equipment-and-permissions.md, section 6) to take it.
• Count types: Full, Cycle, Selective, Event. By hand you can start only Selective. A Cycle count comes from a policy. An Event count comes from an event (section 3). Any other try: a Cycle count is made by its policy or event, not started by hand.
• Status of a task: Open (being counted), Variance (counted, a difference waits for a decision), Closed (matched, or the difference was adjusted), Recounted (replaced by a new attempt), Cancelled.
• Status of a count: Open, Closed, Cancelled.

The RF path: Inventory menu, Count Location. Screens: Count — Work Area (scan the work area, or press Enter for the whole warehouse), Count — Next Location, Count — Go To Location ("Go to the location and scan it"), Count — Finish (1 = Done, 2 = Empty, 3 = Leave), Counted. A supervisor decides differences on the web: Sidebar > Inventory Balances > Inventory Counts (buttons Accept Variance, Recount, Cancel Count). Accept Variance asks why the count differs (section 6b).

2. Count triggers (what starts a count)

#	Trigger	Switch	Where you set it	Source value on the count	Count type
1	A cycle policy finds locations due	the policy	Sidebar > Policy > Cycle Count Policies (section 4)	Schedule + the policy id	Cycle
2	A supervisor starts a count	none	Sidebar > Inventory Balances > Inventory Counts > New (section 5)	User	Selective
3	A short pick, when the reason has the action CreateInventoryCount	the reason	Sidebar > Policy > Pick Short Reasons (23-pick-short-reasons.md)	ShortPick:<reason code>	Event
4	A full pick empties its source location	On an Emptied Location = Count	Sidebar > Policy > Picking Policies (21-picking-policy-settings.md, section 6)	PickToZero	Event
4a	The picker answers "is it empty?" against the stock (said not empty where nothing is, or empty where stock is)	On an Emptied Location = Confirm	the same	ZeroConfirmation	Event
5	A posted adjustment whose reason says "count after posting"	Count After Posting	Sidebar > Inventory Balances > Adjustment Reasons (section 6)	Adjustment	Event
6	A directed putaway landed through an acknowledged deviation	COUNT_ON_PUTAWAY_DEVIATION	Sidebar > System Settings (section 7)	PutawayDeviation	Event
7	An order is cancelled while one of its picks had its source confirmed (location, item, LPN or lot scanned)	COUNT_ON_PICK_CANCELLED	Sidebar > System Settings (section 7a)	PickCancelled	Event
8	An operator reports stock the system does not show (RF Report Unexpected Stock, or F6 on an empty Stock Inquiry)	none — always	—	UnexpectedInventory	Event

How the event triggers work (3 to 8): the transaction that causes them writes the event process.inventory.count_requested into the outbox. It does not start the count. A background subscriber starts the count after the first transaction has committed (wms-inventory/src/service/inventory_count/request_location.rs). So a rolled-back pick, adjustment or putaway asks for nothing.

Rules of an event count (request_location.rs:96-135):

• It counts one location.
• If a live task (Open or Variance) already counts the location, the request joins it. Nothing new is made.
• Mode Blind, freeze Hard, recount by another counter, 2 recounts per location. A pick short asks for a Soft count instead, but only when the short blocked the location (the reason has BlockSourceLocation); without that block its count stays Hard (23-pick-short-reasons.md, section 8).
• Task Manager priority 10. Planned counts have the default 50. Smaller is sooner.
• A location the count rules refuse (see sections 3 and 5) is an answer, not a failure. The event is done and is not retried. Examples: COUNT_LOCATION_NEVER_COUNTED, COUNT_LOCATION_NOT_COUNTABLE.

3. Rules that are not settings

These are fixed in code. There is nothing to switch.

1. What a difference is. Any difference larger than 0.000000001 per line is a difference (classify at the end of reconcile_moving). A count with no difference closes by itself and unfreezes the location. A small difference can be accepted on its own when the tolerances allow it (section 6a). Without tolerances every difference waits for a supervisor.
2. A line is one identity: owner, SKU, pack, unit of measure, lot and container. Moving stock between two pallets at one location shows as two differences.
3. Result types per identity: Match, QtyVariance (both sides have stock, quantities differ), ExpectedButMissing (the system had stock, the counter saw none), UnexpectedFound (the counter saw stock the system did not have).
4. Zero is not counted. The counter cannot enter a quantity of 0. "Nothing here" is the button 2 = Empty (confirm empty). A location with no entries and no empty confirmation cannot be finished: nothing was counted in A-01-01: record what is there, or confirm that it is empty.
5. Never counted types: InTransit (open pick pallets) and Lost (lost stock). A policy or a count that names them is refused: location type InTransit is never counted, never counted (lost stock or open pick pallets): INTRANSIT.
6. Who decides a difference: Accept Variance, Recount and Cancel Count are for the roles Admin, Manager, Supervisor (app-server/src/web/rpcs/inventory_count_rpc.rs, the role list at the top). Counting itself (take, observe, complete, empty, leave) is also open to Operator. Reading counts and policies: Admin, Manager, Supervisor, Viewer.
7. Approval. Accepting a difference makes an adjustment with the reason COUNT_VARIANCE and posts it in the same step. That reason needs no approval and no comment (section 6). If someone changed it to need approval, the accept is refused: the COUNT_VARIANCE reason asks for approval, which a count cannot give; adjust this variance by hand.
8. The accept can fail on reservations. If the counted quantity is below what is reserved or picked at that balance: balance 4412: the target is below what is reserved (allocated 5, picked 0). Nothing is written. Release the reservation or recount.
9. The accept can fail on moved stock (Hard freeze): the stock changed since it was counted (counted against 40, holds 38 now): count it again. The task stays in Variance. Press Recount.
10. Fixed Task Manager priority of an event count: 10.

4. Cycle count policy

Menu path: Sidebar > Policy > Cycle Count Policies. Table inventory_count_policy (wms-app 058, 073). A policy says which locations, or which SKUs, are counted how often. A daily run turns what is due into one ordinary count. The field Basis decides which: Locations (the policy narrows by zone, type, ABC and aisle of the locations) or Items (it narrows by the ABC class and group of the SKUs, and counts each SKU wherever it lies).

• Roles that change: Admin, Manager, Supervisor. Read: Admin, Manager, Supervisor, Viewer.
• Buttons: list New, Refresh; card Save, Run Now, Delete, Back.
• Seed: none. No policy exists at the start and no dev seed makes one.
• A save that is stale: version conflict on inventory_count_policy 7 (code 40, the number is the policy id). Reload.

Code (code)

• UI label: Code. Read-only: the system gives a number with the prefix CC and 4 digits when you press New (counter key inventory_count_policy, wms-meta-ui 098).
• Column: varchar(30), unique, not null. The service refuses an empty code (a policy needs a code) and one over 30 characters (the code is at most 30 characters).
• Duplicate: another policy is already called CC0007.
• To verify the exact form of the generated code: press New on Cycle Count Policies and read the Code field. The code reads CC plus a counter (counter key inventory_count_policy, size 4).
• What it does: the name. It is written in the comment of every count the policy makes.

Description (description)

• UI label: Description. varchar(200), null allowed. Message: the description is at most 200 characters. Stored only (shown in lists and comments).

Active (is_active)

• UI label: Active. Checkbox. boolean, not null, default true.
• What it does: the scheduler runs only active policies. Run Now runs only an active, saved policy: Only an active, saved policy runs (button reason) and policy CC0007 is not active: activate it to run it (service).
• Open work: counts already made go on. Making a policy inactive does not cancel them.

Basis (basis)

• UI label: Basis. Dropdown. Enum inventory_count_policy_basis: Locations (default), Items (Items — SKUs, wherever they lie). Helper text: Locations: the locations matching the zone, type, ABC and aisle below. Items: the SKUs of the class and group beside, counted wherever they lie.
• What it does: decides what the run looks for (InventoryCountPolicyBasis, wms-lib-core/src/model/inventory/inventory_count_policy.rs; service/inventory_count_policy/run.rs). A policy names the properties of its own basis only. Mixing them is refused: a policy counts locations or items, not both: name the properties of its basis only (code COUNT_POLICY_MIXED_SCOPE). The database check chk_inventory_count_policy_scope repeats the rule (wms-app 092).
• Open work: none. New policies start with Locations.

SKU ABC (sku_abc) and SKU group (sku_group)

• UI labels: SKU ABC (dropdown A, B, C) and SKU group (text). Columns: sku_abc (enum abc_speed_type), sku_group varchar(10). Used only when Basis is Items.
• Helper texts: Items: the SKUs of this ABC class. A SKU with no class set is B. and Items: the SKUs of this group. A SKU group longer than 10 characters is refused (COUNT_INVALID_INPUT).
• At least one must be set for an Items policy. Refusal: a policy over items narrows by the SKUs' ABC class or group: one over every SKU is a full inventory (code COUNT_POLICY_WITHOUT_ITEM_SCOPE). When both are set, a SKU must match both.
• Example: Basis = Items, SKU ABC = A, Every (days) = 30, Locations per Run = 20. Every run counts up to 20 SKUs of class A that have not been counted for 30 days.
• A SKU is due when it is active, matches the class and group, holds stock (quantity above 0) in a location that a count may count, is named by no open count, and has not been counted within Every (days) (InventoryCountPolicyBmc::due_skus). "Counted" means that a closed count by item named the SKU code and did not narrow by lot or LPN. A count of a whole location does not count the SKU. The due SKUs are sorted: never counted first, then the longest uncounted, then by code; the first Locations per Run are taken. One code across owners is one item.
• The list of policies shows Basis and SKU ABC as columns.

Zone (putawayzone), Location Type (loc_type), ABC (abc), Aisle (aisle)

• Used only when Basis is Locations. UI labels: Zone (popup picker, free entry allowed), Location Type (dropdown), ABC (dropdown), Aisle (text).
• Columns: putawayzone varchar(20), loc_type (enum loc_type), abc (enum abc_speed_type: A, B, C), aisle varchar(10). All null allowed. Messages for length: the putaway zone is at most 20 characters, the aisle is at most 10 characters.
• The dropdown for Location Type offers only Case, Pick, SpeedPick, Other, Pnd, Pickto, Sort. The API accepts any type except InTransit and Lost.
• What they do: the scope. A location is in scope when it matches every property that is set (AND). Helper text: Count only locations matching every property set here — at least one.
• At least one must be set on a Locations policy. A policy over every location would be a full inventory. Refusal: a policy narrows by at least one location property (zone, type, ABC, aisle): one over every location is a full inventory (code COUNT_POLICY_WITHOUT_SCOPE; database check chk_inventory_count_policy_scope).
• Affects: wms-lib-core/src/model/inventory/inventory_count_policy.rs (due_locations).
• Example: Zone = RACK, ABC = A. Only locations of zone RACK with the ABC class A are considered.
• Open work: none. The scope is read at each run.

Every (days) (frequency_days)

• UI label: Every (days) (required). integer, not null. Check 1 to 3650 (chk_inventory_count_policy_frequency).
• Message: a location is counted every 1 to 3650 days, not 0.
• What it does: a location counted within this many days is not due. The last count is the newest task of that location with a count time that is not Cancelled (a task waiting in Variance counts as counted).
• Helper text: A location counted within this many days is not due.
• Example: Every (days) = 14. Location A-01-01 was counted 10 days ago: not due. 15 days ago: due.
• Never counted locations are due first.

Locations per Run (daily_limit)

• UI label: Locations per Run (required). integer, not null. Check 1 to 10000.
• Message: a run counts 1 to 10000 locations, not 0.
• What it does: the most locations in one run (one count). On an Items policy it is the most SKUs in one run. The due locations are sorted: never counted first, then the longest uncounted, then by code. The first N are taken.
• Example: 120 locations are due and the limit is 50. Today's count has 50. Tomorrow's run takes the next ones, because the first 50 are now counted.

Mode (count_mode)

• UI label: Mode (required). Dropdown. Enum inventory_count_mode:
  • Blind (Blind — the counter does not see the system quantity). Default.
  • Informed (Informed — the counter sees it).
• What it does: with Blind the system hides the expected quantities from the counter while the task is Open (service/inventory_count/mod.rs). With Informed the task answer carries system_stock on the first attempt while the task is counted: the location's stock now, line by line (SKU, lot, LPN, quantity), a soft count's changes since the snapshot included, and for a count by item only the item. A recount is blind whatever the mode. The RF screen CNT02 shows it in the line System (inventory count spec section 8; rf-screens/114_uc_inv_count_informed.sql).
• Interaction: the counts the policy makes copy it. A manual count has its own Mode.
• Open work: only counts made after the change.

Freeze (freeze_mode)

• UI label: Freeze. Dropdown. Enum inventory_count_freeze: Hard (Hard — the locations are frozen while they are counted), Soft (Soft — stock keeps moving; what moved after the snapshot is expected). Default Hard. Helper text: For the counts this policy makes.
• What Hard does: the location is frozen from the start until the task closes. Nobody can move, allocate or pick there. The expected quantity is the snapshot taken at the start.
• What Soft does: nothing is frozen. The expected quantity is the snapshot plus the stock changes committed after the snapshot and before the count was completed (the location journal location_stock_effect). A pick made after the snapshot is not a difference. On accept, the lines are target = holds now + (counted − expected).
• A Soft count starts only at a location a frozen one could freeze: the location must be Ok and have no other flag. COUNT_LOCATION_NOT_COUNTABLE otherwise.
• Refusal for soft: more stock left A-01-01 after it was counted than the count found there: recount it.
• The journal rows are kept 30 days, except for locations a live count task still covers.
• Example: snapshot 100. Pick 30 later. Counter finds 70. Hard would never allow the pick. Soft expects 100 − 30 = 70: match.
• Open work: only counts made after the change.

Trust Empty Confirmations (trust_empty_confirmation)

• UI label: Trust Empty Confirmations. Checkbox. boolean, not null, default false (wms-app 112). Count spec Appendix L.6.
• What it does: a location a picker confirmed empty after a pick (On an Emptied Location = Confirm on the picking policy, answer 1 = Yes while the stock held nothing — location_evidence, kind ConfirmedEmpty) counts as checked at that moment. The location is due Every (days) after the later of its last count and its last confirmation. A count after the confirmation is newer, so it wins.
• Only a Locations policy reads it. On an Items policy the server stores false.
• Affects: InventoryCountPolicyBmc::due_locations.
• Example: Every (days) = 30. A-01-01 was counted 40 days ago, and a picker confirmed it empty 3 days ago. Off: due today. On: due in 27 days.
• Not trusted: an answer that contradicted the stock (DeniedEmpty, EmptyAgainstStock) — that one asked for a count instead.

Recounts per Location (max_recounts)

• UI label: Recounts per Location. smallint, not null, default 2. Check 0 to 9 (chk_inventory_count_policy_max_recounts).
• Message: recounts per location are 0 to 9, not 10.
• What it does: how many times a location with a difference may be counted again. Attempt 1 is the count; each recount is one more attempt. With 2: attempt 1 blind, attempt 2 by another counter, attempt 3 the last. 0 means no recount. After the limit: A-01-01 has been recounted 2 time(s), the most this count allows: accept the variance or cancel the count.
• Affects: domain/inventory_count/aggregate.rs (ensure_recount_allowed).
• Open work: counts already made keep the limit they were made with.

Last Run (last_run_on)

• UI label: Last Run. Read-only. date, null.
• What it does: the scheduler runs a policy at most once a day. A run (also Run Now) writes the day. A run does not change version.
• Affects: domain/inventory_count_policy/aggregate.rs (runs_on). The day is the warehouse day: the date in the warehouse's time zone, the system setting WAREHOUSE_TIMEZONE of that warehouse (an IANA name such as Europe/Sofia, seed UTC; SystemSettingsBmc::warehouse_today). The IANA zone supplies its daylight-saving rules, so the boundary stays local midnight when the UTC offset changes. A blank value uses Default Value; a missing row uses UTC; an unknown name is refused (see 42-system-settings.md).
• Example: the policy ran today and found nothing due. It will not run again before the warehouse's midnight. Run Now still works.

Version (version)

• UI label: hidden on the card. integer, default 1. Moves on save. Protects from two editors.

Scheduler and Run Now

• The scheduler loop runs when the service starts and then every hour (COUNT_POLICY_INTERVAL in cleanup_tasks.rs). A new or re-activated policy runs within one hour. Several servers do not run one policy twice a day: a run locks the policy and skips it if it already ran on the warehouse day.
• Each policy runs in its own transaction. If one fails (for example somebody froze a location between the read and the freeze), its day is not written and it tries again next hour.
• Run Now runs one policy. It does not ask whether the policy already ran today. The message is Count CNT-000012 started for 50 location(s) due under policy CC0007 or Nothing is due under policy CC0007. A failure is The policy did not run: {error}.
• A run makes one count: type Cycle, source Schedule, source id the policy id, the policy's mode, freeze and recount limit, recount by another counter on, comment Cycle count by policy CC0007 (every 14 days). For an Items policy the count is a count by item (Scope = Only the item): its SKUs are the due SKUs, and its locations are the countable locations that hold them. Locations that another count already holds are left out. The message is Count CNT-000012 started for 5 item(s) due under policy CC0007, in 8 location(s). If nothing can be taken, no count and no number are made: Nothing is due under policy CC0007.
• A location is due when it is Ok, has no operation flag (loc_flag = None), is not of a never-counted type, has no live count task, is in scope, and was not counted within Every (days).
• Delete: button Delete, message Policy CC0007 deleted. The counts it made stay.
• Existing counts and tasks are never changed by editing or deleting a policy.

5. Settings of one count (manual or from API)

On Sidebar > Inventory Balances > Inventory Counts. Table inventory_count. The card shows Number, Status, Mode, Freeze, Locations, Comment, Recounts per Location, and read-only Type, Source, counters and dates. After the count starts, you do not edit what you typed.

Field	Column	Values and default	What it does
Locations	tasks	codes separated by commas, semicolons or spaces	The locations. Helper: Location codes, separated by commas or spaces. Saving freezes them for the count.
Scope	scope_mode	Whole location = LocationFull (default), Only the item = IdentitySubset	What a counter counts in a location. Only the item needs an item.
SKUs, Lots, LPNs	item_skus, item_lots, item_containers	lists	The item of a count by item. An identity is in the item when it matches every list that is named. Every location that holds the item is added when the count is saved; busy ones are left out. Allowed on a Selective or a Cycle count only: only a selective or a cycle count is counted by item. If nothing countable holds the item: COUNT_SCOPE_EMPTY. A counter's entry for anything else is refused (COUNT_OUTSIDE_ITEMS). See the reference page reference/inventory/27-count-header.md.
Mode	count_mode	Blind (default), Informed	As in section 4.
Freeze	freeze_mode	Hard (default), Soft	As in section 4.
Recounts per Location	max_recounts smallint	0 to 9, default 2	Message recounts per location are 0 to 9, not 12.
Comment	comment varchar(500)	text	a comment is at most 500 characters.
(API only) recount_by_other_counter	boolean	default true	A recount is counted by somebody other than the counter of the attempt before it. Refusal: a recount of A-01-01 is counted by somebody other than the first counter. The web card always uses true; the policy always uses true.
Number	count_number varchar(20)	system number; a count number is at most 20 characters

Other refusals when you start or edit a count:

Message	Cause
a count needs at least one location	The list is empty.
cannot be frozen for a count (unknown, blocked or under another operation): A-01-01	The location does not exist, is not Ok, or has another flag.
already being counted: A-01-01	A live task counts it (COUNT_LOCATION_BUSY).
nothing can be counted in ZONE-X: every location is blocked or already being counted	Add by zone found nothing.
already in this count: A-01-01	Added twice.
A-01-01 has been counted: it stays in the count — accept, recount or cancel	Remove of a counted location.
count CNT-000012 has been counted: it is cancelled, not deleted	Delete of a count with counted tasks.
the count is Closed, not open	A command on a closed count.
the count of A-01-01 is Closed: it is not being counted	Observation after the task closed.
A-01-01 is not given to you: take your next location	The counter does not hold the task.
A-01-01 is being counted by anna	Another counter holds it.
nothing to count / nothing to count in area PICK-A	The pool is empty for the counter.
no work area ZZ	Unknown area scanned.
quantity 0 is not a positive number	A zero observation.
SKU X1 does not exist for this owner, SKU X1 belongs to several owners: name the owner, no owner O1, no container C1, lot L9 is not a lot of X1: F2 on Lot finds a lot by its attributes, SKU X1 has no default unit: name the unit	The counter's entry does not resolve.

6. Adjustment reasons (what the difference becomes)

Menu path: Sidebar > Inventory Balances > Adjustment Reasons. This is the list of reasons of inventory adjustments (code list inv_adjust_reason, id 1031). The list columns are Code, Reason, Active, Adds Stock, Removes Stock, Comment Required, Approval Required, Count After Posting.

An accepted count difference uses the reason COUNT_VARIANCE (service/inventory_count/accept.rs). The adjustment has source InventoryCount. Each line: target quantity = what was counted, expected quantity = the snapshot. The adjustment is validated and posted in the same step. A count never asks the adjustment for approval.

Seeded reasons (wms-app 053, 054):

Code	Reason	Adds stock	Removes stock	Comment required	Approval required
DAMAGE	Damage	no	yes	no	no
FOUND	Found stock	yes	no	no	no
LOST	Lost stock	no	yes	no	no
COUNT_VARIANCE	Count variance	yes	yes	no	no
ERP_RECONCILIATION	ERP reconciliation	yes	yes	no	yes
DATA_FIX	Data fix	yes	yes	yes	yes
QUALITY	Quality	no	yes	no	no
EXPIRY	Expiry	no	yes	no	no
CUSTOMER_REQUEST	Customer request	yes	yes	yes	no
OTHER	Other	yes	yes	yes	yes
REVERSAL	Reversal of a posted adjustment	yes	yes	yes	no

These flags are values of the code list value metadata. The code list is not editable by hand (editable = false). A missing key reads as the strict answer: no direction allowed, comment and approval required.

Refusals of a line: reason X is not on the adjustment reason list, reason X is retired, reason LOST does not allow adding stock (+5), reason FOUND does not allow removing stock (-3), reason OTHER requires a comment.

Count After Posting (count_after_post)

• UI label: Count After Posting. Buttons Count After Posting, Do Not Count, Refresh. Tick one reason first (Tick one reason).
• Roles: Admin, Manager (inventory_adjust_reason_set_count, inventory_adjustment_rpc.rs:44).
• Store: key count_after_post in the metadata of the code list value. Absent means off. No reason has it on at the start.
• What it does: after an adjustment with this reason is posted, each location whose stock it changed is counted (trigger 5). One request per location, each once.
• An adjustment that a count made (source InventoryCount) never asks for a count. That is how a count ends.
• Messages: Reason OTHER: a posted adjustment counts the locations it changed, Reason OTHER: a posted adjustment counts nothing, Reason OTHER was not changed: {error}, The list has moved on — tick the reason again and retry.
• Example: DATA_FIX has it on. A manager posts an adjustment for A-01-01 and A-01-02. Two Event counts start, one per location, after the post commits.
• Open work: only adjustments posted after the change.

6a. Count tolerances (what is accepted on its own)

Menu path: Sidebar > Policy > Count Tolerances (right after Cycle Count Policies). Table inventory_count_tolerance (wms-app 096). There is one row per SKU ABC class: A, B, C. You cannot add or delete a class. A SKU without a class counts as B.

• Roles: Admin, Manager change a tolerance (update_inventory_count_tolerance). Admin, Manager, Supervisor, Viewer read them.
• The card is saved with its version. If someone else saved first, the save is refused. Refresh and repeat.
• At the start all three rows are empty: nothing is accepted on its own, and nothing needs a manager.

Field	Meaning	Example
Accept up to (units) (auto_qty)	A line is within tolerance when its difference is at most this many units.	2
Accept up to (%) (auto_pct)	… and at most this percent of what was expected. Expected 0: the percent is never met.	5
Accept up to (value) (auto_value)	… and the difference × the SKU's cost attribute is at most this sum. A SKU without a cost skips this rule.	50
Manager above (value) (approval_value)	A difference worth more than this is accepted only by a user with the right inventory_count.accept_large_variance (by default ADMIN and MANAGER, not SUPERVISOR).	1000

When it applies:

• A task is within tolerance when every line with a difference meets every rule its class sets, and its class sets at least one of Accept up to (units) and Accept up to (%). A class with neither never accepts on its own.
• The task's total value must also stay within Manager above (value) of its classes. Accepting on its own never skips the manager.
• It happens when the counter finishes the location, on any attempt. The adjustment is the same one a supervisor would make. The RF screen Counted says Within tolerance - adjusted, released. The cause is Within tolerance (section 6b), and the decision is in the counter's name.
• If a tolerance changed between the count and the accept, or the adjustment is refused (moved stock, reservations), the task stays in Variance for a supervisor.
• A supervisor above their limit gets COUNT_VARIANCE_ABOVE_LIMIT with the sum and the limit. A manager accepts it.

6b. Variance causes (why the count differs)

Every accepted difference records why the count differed. The causes feed the board panel Variance causes and the report Count Variance Causes (08-reports-and-dashboards.md).

The cause list is the code list count_variance_cause (id 1042, wms-app 132). Its seeded causes:

Code	Cause	Who sets it
PICK_ERROR	Pick error	supervisor
PUTAWAY_ERROR	Putaway error	supervisor
MIX_UP	Mix-up with another item	supervisor
RECEIVING_ERROR	Receiving error	supervisor
DAMAGE	Damage	supervisor
THEFT_LOSS	Theft or loss	supervisor
COUNT_ERROR	Count error	supervisor
UNKNOWN	Not established	supervisor, when nobody knows
WITHIN_TOLERANCE	Within tolerance	only the system (section 6a). It is not offered in the window.

On Russian screens the causes have Russian names.

Add your own cause:

1. Open Sidebar > Code > Code List. Find count_variance_cause.
2. In Code List Values, click New Value.
3. Fill Code, Display Order, Description, Active. Click Save. The window offers the cause from the next accept, in Display Order.

• The seeded causes cannot be renamed or deleted. You can switch one off (Active off). A switched-off cause is no longer offered. Accepts already made keep it.
• Your own cause appears under its description everywhere: in the window, on the board, in the report.

Accept a difference with a cause:

1. Open Sidebar > Inventory Balances > Inventory Counts. Open a count.
2. Tick a location in Variance, or open its card.
3. Click Accept Variance. The window Accept Variance opens.
4. Pick Cause (required). Type Comment if it helps (up to 200 characters).
5. Click Accept. Cancel closes the window and changes nothing.

From the list, the list stays and refreshes. From the card, you go back to the list.

After the accept, the list and the card show Cause and Cause Comment. The adjustment comment ends with the cause code, for example — cause MIX_UP.

Message	Meaning	Fix
Choose the cause: «Not established» if nobody knows	The window: no cause picked.	Pick a cause. If nobody knows, pick Not established.
The variance of {location} was not accepted: {error}	The server refused. {error} says why.	See the rows below and section 3, rules 7 to 9.
say why the count differs: choose a cause ("Not established" if nobody knows)	The server: no cause (COUNT_CAUSE_REQUIRED). Seen from the API.	Send cause_code.
X is not a cause you can choose	The cause is unknown, switched off, or the system's own (COUNT_CAUSE_NOT_OFFERED).	Refresh the page and pick again.
this variance is worth X, more than the Y you may accept: a manager accepts it	Above Manager above (value) (COUNT_VARIANCE_ABOVE_LIMIT, section 6a).	A manager accepts it.

A recount does not record a cause. The cause belongs to the attempt that is accepted, the last one. A difference accepted before causes existed shows on the board as no cause recorded.

7. System setting COUNT_ON_PUTAWAY_DEVIATION

• Where: Sidebar > System Settings. Table system_settings, key COUNT_ON_PUTAWAY_DEVIATION, editable.
• Value: 0 or 1. Seed 0 (and default 0). Description: Directed putaway with an acknowledged deviation (code 85): 0=nothing, 1=count the destination location.
• What it does: when 1, a directed putaway that the operator pushed through an acknowledged deviation (capacity, location type or mixing) asks for the destination location to be counted. Source PutawayDeviation. Anything other than 1 reads as off.
• Affects: wms-inventory/src/movement/validation.rs:752-766, movement/service.rs:856-880.
• Open work: only moves made after the change.

7a. System setting COUNT_ON_PICK_CANCELLED

• Where: Sidebar > System Settings. Table system_settings, key COUNT_ON_PICK_CANCELLED, editable (wms-app 119).
• Value: 0 or 1. Seed 0 (and default 0). Description: Order cancelled while its pick's source was already confirmed: 0=nothing, 1=count the source location.
• What it does: when 1, cancelling an order (shipment_order_cancel) asks for a count of the source location of every pick it removes whose source an operator had already confirmed — the operator may have been at the shelf, units in hand. Source PickCancelled, the pick as the record; one count per location. A pick nobody had started, the operator's own reject (F7), reassignments and un-allocation never count: they leave the shelf as it was. Anything other than 1 reads as off.
• Affects: wms-outbound/src/picking/cancelled_pick.rs, called from service/order/cancel.rs.
• Open work: only orders cancelled after the change.

8. Count permissions in Task Manager

A count task is a Count task. The operator needs:

1. A row in the grid Task Permissions with Task Type = Count and a Work Area.
2. The zone of the location in that area, and not excluded by the operator's equipment profile.

Without them: you have no right to take Count tasks in any work area — ask a supervisor. The pool rules, the lease and the cooldown are the same as for picking (22-work-areas-equipment-and-permissions.md). The count does not use Pick Execution.

9. What happens to open counts when you change something

Change	Effect
Edit or delete a policy	Counts already made are not touched. The next run uses the new values.
Make a policy inactive	Open counts go on. No new run.
Change Count After Posting, Count an Emptied Location, a short reason, COUNT_ON_PUTAWAY_DEVIATION, COUNT_ON_PICK_CANCELLED	Only later events. Events already written in the outbox still start their counts.
Remove a Count permission	The counter keeps the location they hold.
Cancel a count	The tasks are cancelled and the locations unfrozen.
Delete the task on the Task Board	The next hourly pass publishes work again for a location still being counted.
Change a count tolerance	Only counts finished later. A task already in Variance waits for a supervisor.
Switch off a variance cause, or add one	Only later accepts. Accepted differences keep their cause.

10. Inconsistencies found

1. (Fixed, P-20: the day is the warehouse day, and the first pass is at start.)
2. (Fixed: count tolerances, section 6a.)
3. The Location Type dropdown of the policy card offers 7 types, while the service accepts every type except InTransit and Lost.
4. The publish message of short reasons says counting "is not implemented" (see 23-pick-short-reasons.md).
5. The option recount_by_other_counter has no field on any web screen.