Mars Climate Orbiter’s unit mismatch. The mission was already lost.

A campaign should stop before calling when the team cannot show which consent language each contact heard, accepted, or opted out of. A call log alone cannot answer that question.

In September 1999, NASA lost contact with the Mars Climate Orbiter as it approached Mars. The spacecraft had launched the previous year, and the investigation later found that a ground software interface used pound-seconds while another part of the system expected newton-seconds. NASA’s Mars Climate Orbiter Mishap Investigation Board documented the mismatch.

The problem was not that nobody had data. The problem was that the systems did not share a verified meaning for it. By the time the discrepancy mattered, the mission was already lost.

A call campaign can reach the same kind of stop point on a much smaller, but still serious, scale. A manager may have a list, a dialer status, recordings, and an export showing that calls happened. Then a customer challenges the call. The team needs to answer a narrower question: what exactly did this person hear before the conversation continued?

A consent record needs more than a completed-call status

“Completed” tells you that a call reached an end state. It does not establish the consent wording, the language used, whether the person opted out, or which version of an agent made the call.

That distinction matters when a campaign uses Twi and English. A disclosure translated casually by a team member, changed in an agent prompt, or replaced during a campaign can leave a real gap in the record. The contact may have heard one version while the campaign manager can only produce another.

Before a production run, define the consent record that must exist for every call:

• The consent language shown or spoken, including its version.
• The language selected for the disclosure.
• The time the call entered each lifecycle state.
• Any opt-out signal and the action taken afterward.
• The agent configuration used for that call.

This is operational work, not a compliance paragraph added after launch. If a support desk cannot retrieve these details for one disputed number, it will struggle to defend the campaign as a whole.

Treat language changes as controlled changes

The Mars Climate Orbiter failure is a useful warning because the data looked usable until the units had to agree. Consent language has a similar failure mode. A familiar phrase can hide a meaningful difference in what a customer was told.

Keep the wording, language, and agent version together. If the first message or disclosure changes, create a new version and decide which contact groups can receive it. Do not rely on a campaign name such as “September follow-up” to reconstruct the decision later.

This is especially important for teams evaluating native Twi voice experiences. Natural speech recognition and synthesis can make a conversation easier to follow, but the consent record still needs to identify the disclosure that was used. The same care applies when a caller switches languages mid-call.

Asenda Talk is built around native Twi speech recognition and synthesis, with agent persona, first-message, and voice configuration available in the platform. Its consent, opt-out, and audit-trail capabilities are intended to give operators call-level records to inspect. Teams should validate their own approved wording and review process during evaluation rather than treating an early-access configuration as a finished production control.

Stop conditions protect the campaign and the people receiving calls

A good stop condition is specific. “Pause if we cannot prove consent” is too broad to guide someone at 8:30 in the morning. Write down what triggers a hold.

For example, pause the run if the agent configuration cannot be tied to a consent-language version, if opt-out events do not appear in the review record, or if a call state conflicts with the telephony event trail. Make one person responsible for deciding when the evidence is sufficient to resume.

Asenda Talk’s telephony lifecycle webhook pipeline tracks call truth across events, rather than treating every attempted call as a completed conversation. That distinction supports the kind of review described in Voice campaign audit records: Why attempted calls are not completed conversations. It also helps a team separate a delivery issue from a consent-record issue.

Outbound calling remains gated in Asenda Talk while the telephony-provider decision is not live. That means current work should focus on evaluation, agent configuration, audit expectations, and review workflows, rather than assuming a production outbound campaign can be launched from the platform today.

Build the evidence path before the first call

The strongest time to test an audit trail is before there is a dispute. Run a small internal test set. Have reviewers check whether they can identify the exact consent version, language, call state, opt-out outcome, and agent configuration without asking the person who set up the campaign.

If they cannot, the campaign manager has found the real issue early. Fix the record design, then test again.

NASA’s investigation did not describe a shortage of technical effort. It showed what happens when critical information crosses a boundary without a shared, checkable interpretation. For voice campaigns, the boundary is often between the approved disclosure, the configured agent, the telephony event, and the record a team can produce afterward. Make those records agree before the call run starts.