TL;DR An IBC Classic timeout is a verified source-chain transition supported by destination-chain evidence. A dashboard clock passing the requested deadline does not return funds. For a standard fungible token transfer over an unordered channel, reconcile the packet identity, destination evidence of non-receipt at the applicable timeout, the accepted source timeout, and the token application's refund. A receive or acknowledgement follows a different path through the same record.
The incident begins with two balances
Mira supports a fictional member cooperative that buys cloud credits through a test transfer workflow. The team sends 50 units of a token along one IBC Classic channel. The source wallet shows the transfer amount leaving the available balance. On the destination, the intended account still shows zero. A status dashboard labels the transfer as late. Several people interpret that report as a verdict, while Mira is still trying to find which chain has made which decision.
The question for the incident desk is specific. Can it establish, with enough evidence to update the customer record, that one packet was never received and that the amount was returned by the source application? The question combines a communication result and a token accounting result. Both matter to the customer. The conditions under which each becomes observable differ. Reading the wallet again can show a difference in balance without showing which packet caused it.
The simulation uses an unordered IBC Classic channel and an ordinary one-hop ICS-20 fungible token transfer. IBC defines shared channel and packet semantics while the token module specifies its actions on send, receive, acknowledgement, and timeout. Actual deployments can add middleware, forwarding, fees, upgrades, or other application behavior. A team investigating those deployments must match its evidence to the applicable channel and version. The fictional desk uses the simple route so that none of those features hides the basic decisions. ICS-04 channel and packet semantics ICS-20 fungible token transfers
Sending-chain record. The first witness is the block in which the source accepted the transaction that sent the packet. The block produced a packet commitment under the source port, channel, and sequence. The operations note includes those fields, the complete chain identities, the channel counterpart, and the timeout settings. A transaction identifier can point to where the packet was created. It remains possible for an application transaction or a relayer transaction to involve other work as well, so Mira identifies the packet independently.
Destination-chain record. The second witness is the corresponding chain's state, which can prove receipt under the expected destination port and channel. For an unordered channel the receipt marker occupies an identified location for the specific packet. The destination can also store an acknowledgement, with application meaning that must be interpreted separately. A relayer can carry proof of relevant state back to the sender, while neither a spoken assurance from the relayer nor an empty dashboard exercises the chain's verification logic. ICS-04 semantics and packet handlers
Customer service record. The third witness is the cooperative's order system. It can say that a given transfer was supposed to pay for order M116. It can see which amount was expected and whether the credits were issued. It does not decide whether an IBC packet was received. A complete case links the order to a protocol record under an explicit policy. If that link was lost, even a perfect on-chain receipt may be hard to match to the correct customer.
Exhibit one. A possible timeout is still an open case
In the first imagined outcome the destination timeout height is revision 7, height 500. At a state whose revision and height are 7 and 498, a relayer checks the receipt path and finds no receipt. The cooperative's clock is already well beyond the application's estimated arrival time. The state is consistent with a transfer that has yet to arrive. In this example the packet uses only the height condition. The absence at 498 predates the height needed to establish that the packet's receive window has closed.
There is a reason the source chain needs evidence tied to a known destination state. A display can declare a deadline passed while the destination network has stopped progressing. The source may need a destination consensus state matching the proof. An unverified workstation time and a source-chain height cannot replace that. Likewise a timestamp condition, when set, concerns the corresponding chain timestamp under the packet rules. The proof must be read against the timeout data actually committed for the packet. ICS-04 timeout processing
In the hypothetical log the next useful state is revision 7, height 502. Receipt evidence there is still absent from the corresponding path. Now the chain state can support the configured timeout condition and the proof of non-receipt for this unordered route, subject to the source's verification. This answers whether the right ingredients for an accepted timeout are available. It is not yet proof that the source transaction applying the timeout happened.
Mira assigns a relaying task rather than a customer refund. The task includes the exact route, the packet, and the desired result. The evidence is sufficient only if the source validates it under the configured client and channel rules. If the source client does not have usable consensus state for the proof, the service must resolve that dependency through the applicable operational procedure. A status like verification unavailable can mean the sender has not obtained evidence it can accept, even if the destination is progressing.
The question becomes easier to explain when the operator reports two dates. One is the first observed destination state satisfying the committed timeout condition. The other is the source-chain acceptance of the timeout handler. The dates can be separated by a relay delay. A service that publishes only a timer reaches the first apparent event and has no language for the second. Its customers will quite reasonably ask why their wallets have not changed.
Exhibit two. A found receipt changes the remedy
Now replace the destination result at height 502. The verifier finds a receipt for the packet because the destination accepted it at height 499. An intermediary application has not yet shown the incoming funds. This is compatible with a delayed application acknowledgement, an error response still waiting to be relayed, or an account display that lags the chain. A successful receive and a successful token result are different assertions. A timeout case based on lack of receipt no longer fits this branch.
ICS-04 separates receiving a packet from writing and relaying an acknowledgement. Some handling can be asynchronous. A receipt confirms a stage in packet delivery under the protocol. An acknowledgement conveys a response whose status the application must parse. A standard ICS-20 transfer handles both successful acknowledgement and acknowledgement error, with refund behavior for the error case. The service needs to inspect the actual result rather than assume that an acknowledgement always reports a successful receiver credit. ICS-04 acknowledgement handling ICS-20 acknowledgement and timeout behavior
Mira asks for the destination's acknowledgement, if present, and any matching source acceptance of that acknowledgement. If the response shows success, the team reconciles the destination amount and account before crediting the cooperative order. If it shows error, they track the source module's corresponding refund behavior. A late relayer can leave a success on the destination with the source still holding the packet commitment. It can also leave an error waiting for the source's application action. The support display can communicate both without treating a waiting acknowledgement as a transfer that can no longer arrive.
This distinction matters when chain observers use inconsistent identifiers. The cooperative's destination memo says transfer 41. Its source file has another sequence 41 under a different channel. A lookup that ignores the chain, port, and channel may associate evidence with the wrong packet. The incident desk asks the operator to navigate using the complete route key and actual committed fields. Imagining a single sequence number shared by every transfer would make unrelated successes and failures look interchangeable.
The baseline route is unordered. Its receipt check cannot simply be copied to a different channel order because delivery state may be represented differently. A traditional ordered channel can track the next sequence expected and later operations can have consequences for the channel after a timeout. Scope the incident to the deployed channel's actual ordering and version, and verify its completion rules before deciding what inspecting one receipt path establishes. The IBC specification includes different case branches for channel ordering and newer variants. ICS-04 ordering and packet semantics
Exhibit three. A reimbursed wallet has a real accounting entry
Return to the branch with a valid timeout. An accepted source transaction performs the packet timeout, and the applicable ICS-20 module handles the transfer amount. Depending on which side of the token path is sending, a token can have been escrowed or burned on send. The refund logic corresponds to that action. In one case the source application reverses the escrow by transferring the funds back. In the other it can restore the burned voucher according to the token protocol's rules. The operational record needs the denomination and path alongside the amount. ICS-20 send and refund logic
The example wallet begins with 100 units. It sends 50, leaving 50 available in the fictional bookkeeping. Assume any transaction fees are paid in a different asset and there are no other changes. The accepted timeout refunds 50, returning the test balance to 100. The source timeout state and the refund amount describe the same disposition, but they are established by different parts of the execution record. Consumers can validate the timeout handler at the protocol level and reconcile the token movement at the denomination level.
A second funding event makes the error in a balance-only approach obvious. While waiting, the cooperative receives 50 identical units from an unrelated payment. The available balance again reads 100 before the timeout transaction. Treating that balance as proof of refund can mark the order resolved early. Once the actual refund occurs, the balance rises to 150 and a different investigator might call it a duplicate reimbursement. The correct reconciliation identifies the unrelated incoming movement and the packet-specific refund separately.
A fee in the same token would change the expected arithmetic. If the wallet paid a one-unit fee on send and every other assumption remained unchanged, the post-send test balance becomes 49. After a 50-unit refund the balance reaches 99 rather than 100. The amount refunded can be correct even while the overall wallet balance remains below its starting amount. A source-chain fee and an application refund of packet principal should appear as separate entries. Fee policies depend on the actual chain and application setup.
The token path has another practical effect. A user may describe two vouchers from different routes with the same familiar name. The ledger should check the exact denomination evidence rather than assume that any fifty displayed units from anywhere balance the test. An amount and a ticker can support a chat explanation. Accounting identity may require the chain, denomination trace, and applicable application event. A search that confuses those identifiers could mark a packet reimbursed using an unrelated asset.
The report gives each team one next fact to obtain
Before any transfer attempt, Mira stores the order reference and an intended route. After a genuine send on the source the work record can include the committed packet identity. The desk's status changes when it has evidence about that packet, while still allowing a user to see that the service is attending to a delayed order. This works even if a relayer is operated by someone outside the cooperative and has its own operational rules. Each state retains the provider and chain at which the observation was made.
For the no-receipt branch a useful handoff names the deadline condition, identifies a compatible destination state, and records whether the source accepted the timeout. For the received branch it names the destination receipt, acknowledgement or pending response, and the source result if available. The business view reports whether the original order was fulfilled, returned to a payable state, or left for reconciliation. One actor can investigate relaying while another verifies the order ledger without confusing their findings.
The cooperation plan also avoids a premature second payment. Closing a browser tab does not cancel a sent packet. If a customer wants to retry, the desk needs the original disposition and whatever protections the application has for repeated orders. Once a timeout and refund are verified, the application can create a new attempt under its own commercial rules. A route already showing a successful receipt needs a different investigation. The process avoids paying twice just because the network's acknowledgement crossed the user's patience limit.
Monitoring can count how long a transfer spends awaiting receive, awaiting acknowledgement, eligible for proof submission, and settled on source. These spans measure different dependencies. A large count of packets waiting after the deadline may indicate a proof or client problem, while a build-up of successful destination responses can point to acknowledgement relay. A dashboard that includes only total time hides which team should act. Each timing measurement should identify the chain observation it uses and tolerate that business time differs from chain time.
The cooperative practices a communication example after the incident. An operator says the destination has advanced beyond a specific deadline and unreceived packet evidence is available. The next sentence gives the source timeout transaction status and expected application confirmation rather than promising instant customer restitution. Once the refund is reconciled, the support note reports it from a source-chain record with its denomination and amount. This is precise without asking the customer to learn how to verify the entire channel.
The operations schedule has more than one bottleneck
Mira rehearses a morning where three fictional packets wait for different reasons. The first still has a viable destination receive window, and the service's usual relayer has not submitted it. The second already has an acknowledgement on the destination waiting to be carried back. The third has an expired receive window and no recorded receipt at a suitable destination state. Those cases may look identical from a desktop that shows incoming funds as missing. A relay queue that treats them as the same job would try the wrong action on two of them.
The cooperative opens three investigation records with different completion conditions. For the first it asks whether its party can complete ordinary delivery while the window remains. The review records the freshness of its evidence before asking for action, since a later destination result can make yesterday's plan obsolete. For the second it asks whether the appropriate acknowledgement can be verified on the source chain. For the third it checks whether the needed proof can be accepted and who can submit the timeout. After each submission the desk waits for competent chain evidence of the outcome.
A fictional staffing calculation makes the distinction meaningful. Mira has two people for ninety minutes. Checking destination identities and account displays takes about ten minutes per case in a practice drill. Coordinating proof submission with another team can take substantially longer and does not always require someone to watch the request continuously. Assigning the full ninety minutes of one employee to every old ticket would stall useful triage. She allocates a short evidence pass across a set of cases and assigns follow-up to the party able to advance that stage. This estimate is an internal exercise, not a protocol performance claim.
The team discusses its external commitments without claiming to control a chain's block production. A promise to reply in one hour can be fulfilled with a clear result about which evidence remains unavailable. A promise that the funds will return within an hour depends on actions and observations the desk may not command. The operational plan separates a customer-contact deadline from chain settlement. A delayed but accurate update keeps the customer from repeatedly changing wallets or sending again while the transfer state is unsettled.
A route with a faulty observer can waste everyone's time. If the explorer omits the counterparty channel identifier, every team member could be starting from the same incorrect path. The review determines which deployment owns the configured channels and records a current configuration reference safe to share internally. That record holds identifiers and supported data interfaces. It does not hold wallet secrets or permission to operate another team's relayer. The moment different parties look at the same packet successfully, many apparent disagreements vanish.
One person stays responsible for the customer's overall case. More than one person can complete evidence tasks, but the final record needs one signed-off decision. Mira's handoff protocol records what was observed before the last support reply, which statuses may change asynchronously, and who checks the next result. If a support agent wakes up to a source acknowledgement after a similarly timed timeout alert, the saved evidence tells them which protocol transition won rather than letting the order of email arrival decide.
The last schedule item is a short replay after closure. The staff use the saved chain references to reproduce the arithmetic, then document where the elapsed time was spent. If most delay came before an operator could locate the route key, the next improvement belongs to instrumentation rather than a shorter expiry setting. If an accepted timeout was visible promptly but customer accounting was slow, the remedy belongs to the reconciliation task. This review converts the incident into a specific operational change without changing the protocol's stated deadline.
Preserve the result when tools disagree later
Two weeks later, a destination explorer redesign hides the packet receipt view. The cooperative still has its initial investigation notes and contextual sources. An evidence record should contain enough references for an authorized reviewer to derive the same result using supported node data, retained records, or another appropriate source. A screenshot showing a colored icon can supplement the evidence. It cannot supply a proof height or packet identity it never displayed.
A data import can also replay identical relay results during an outage. The cooperative keeps observations as records related to one packet and only applies a business disposition when an appropriate state transition is proved. This lets the same accepted timeout be ingested twice without issuing two refunds from its internal system. Where accounting corrections are needed, the replacement note identifies which evidence was missing or mismatched at original close. It should never replace a previously unresolved state with an unsupported success simply because a different data provider returns fewer fields.
The same careful language applies when a service cannot fully settle a case. If a destination chain is unavailable and the source lacks an accepted proof, the desk can say that no refund has been confirmed. It can identify the dependency and state who is checking it. A token application, intermediary order system, and human support operation can each have different obligations depending on their rules. The cooperative's proof record means someone can see where the chain stopped providing evidence and where a separate decision was made.
The original incident closes with the packet identity, accepted timeout, and a reconciled refund for the correct amount and denomination. The desk marks the original order available to retry under its ordinary rules. Another customer case can use the same evidence vocabulary without assuming the same outcome. What matters is whether each status points to the specific observed fact that justifies it.