A Computer Cannot Be Held Accountable

An old maxim in computing holds that a computer cannot be held accountable, and that therefore a computer must never make a management decision. It reads like a truism. It is in fact a design instruction. Accountability is a human property. It requires a name, a signature, and the willingness to answer for a choice when the choice is questioned. A machine has none of these. It cannot stand before a disciplinary board. It cannot be sanctioned by a bar. When software is allowed to decide, responsibility does not disappear. It goes missing, which is worse.

Software earns its place by doing three things a person should not do by hand. It prepares: it drafts the invoice, assembles the conflict search, lays out the docket. It records: it writes down who did what and when, in a form no one can quietly alter. It enforces: it holds a rule steady against pressure and fatigue. None of these is a decision. Each is the scaffolding a decision rests on. In Mersal the audit ledger is append-only, enforced by database triggers that reject any modification for every role including administrators, so the record of a choice cannot be rewritten after the fact.

Draw the line at the moment of consequence. A conflict search may run automatically, but the waiver of a match is a partner's act, recorded against a partner's name. An invoice may be prepared to the last serial, but issuing it is a person's decision, and once issued it freezes. A matter may be assembled by an associate, but activating it belongs to someone the rules trust with that weight. Mersal enforces the partner approval gate at the database, not in the interface, so the boundary cannot be argued away by a clever screen. The software should make the right decision easy and the wrong one visible. It should never make the decision itself.