Non-GAAP clarity protects internal decisions too

A useful metric tells its user what decision it can support. It should state its universe, numerator, denominator, timing, exclusions, source records and method. A label such as adjusted profit or customer value is not sufficient. The user should be able to reconstruct the calculation and identify the obligations that remain outside it. This paper proposes a metric user contract: a concise record connecting a management measure to its permitted interpretation and its reopening events.

The user contract is an original decision tool, not a legal classification of every internal metric. Reporting requirements vary with jurisdiction, entity, communication and measure. The SEC sources retain their US reporting scope, while the presentation-standard overview retains its applicable scope and effective-date context. The paper adapts a transparency principle for internal decisions without asserting that every private dashboard is regulated identically. That boundary should be explicit because borrowing regulatory authority can make an ordinary management choice seem compulsory.

The first contract field identifies the measure’s purpose. Meridian’s finite contribution value helps explain a supplied customer path; its operating result describes the declared business envelope; its cash balance describes a simplified working-capital path. These purposes are related but not interchangeable. A measure accepted for internal screening cannot automatically support external financial reporting, a loan covenant or a spending authorization. Changing the purpose is a reopening event even when the numerical calculation stays the same.

The second field states the calculation and its reconciliation. For a contribution measure, the record identifies the revenue and cost perimeter. For a ratio, it identifies both units and the treatment of unavailable denominators. For cash, it identifies the opening position and omitted obligations. The reconciliation is particularly important when a management measure is compared with accounting records. A difference may be intentional, but it must be explained rather than hidden behind an attractive adjective.

The third field identifies data and assumption status. A supplied illustrative parameter is different from a recorded amount, an estimate and a forecast. The canonical fixtures are explicitly illustrative. The user contract should preserve that status whenever their results are quoted or exported. A model output can be reproduced exactly while its input remains hypothetical. Exact computation establishes arithmetic under assumptions, not truth about the customer base or efficacy of an intervention.

The fourth field records method changes. A revised activation event, cost perimeter or appraisal horizon can change a reported ratio without improving the business. The metric record should explain the change, its reason and whether earlier comparisons remain meaningful. The user should not be expected to discover a denominator revision from an unexplained jump in a chart. A continuity bridge can be useful, but it must not be fabricated where the old records cannot support the new definition.

The final field identifies an owner and a stop condition. Meridian’s controller can reject a publication that omits the contribution perimeter or describes the cash proxy as unrestricted funds. The data steward can reject a trend whose customer event has changed without reconciliation. Treasury can reject use of a modelled funding amount as proof of available financing. These are concrete interpretation boundaries, not a generic instruction to add more caveats. They make the metric usable within a defined purpose.

The contract catches a particularly subtle failure: comparing a short-window customer value with a long-window value and calling the difference growth. The canonical shorter-horizon fixture changes the window, so the missing later contributions explain part of the difference by construction. A real metric change can do something similar when management extends its appraisal period. The appropriate report identifies the changed window and its purpose. It should not interpret a longer summation as evidence that customers became more valuable.

A second failure is a change in exclusions. Removing an inconvenient recurring expense can improve a management subtotal without changing any payment obligation. The SEC guidance is cited narrowly for transparency and misleading-presentation concerns in its own scope. Meridian’s internal rule is simpler: a perimeter revision must be approved by the controller and reconciled to the decision. If the excluded obligation still controls affordability or service, it remains a separate gate even when the new subtotal is labelled accurately.

A third failure is an unavailable denominator disguised as a good result. The engine returns null for starting-price break-even when price does not exceed recurring variable cost. A dashboard should preserve that state as unavailable, not format it as zero or remove the adverse case from the comparison. Missing and null values have different causes, but neither justifies presenting a favourable invented quantity. The user contract should state the display rule and the decision consequence.

Precision can create another misleading impression. The engine produces exact deterministic outputs to numerical precision, while the supplied acquisition, churn and service costs may be coarse illustrative assumptions. Printed decimals help readers reconcile fixtures; they do not supply evidence that the parameters are known to that precision. A metric report should describe sensitivity and input status rather than relying on rounding to solve the epistemic problem. Less precise formatting cannot transform an unsupported assumption into a supported one.

The contract also governs exports. A CSV or JSON file should identify its illustrative status, schema and method so that a detached result retains its meaning. A frozen output is useful for reproduction, while a new decision should recompute from validated inputs rather than trust an imported output blindly. The research fixtures and canonical runner preserve that distinction. The paper does not ask readers to treat a stored favourable result as proof that its economic or service premises remain valid.

The most useful falsifier is a reader who can reproduce the number but cannot explain its action boundary. In that case the technical method is available while the metric contract is incomplete. The reverse failure is also possible: a polished description with no reconstructible calculation. Both must be corrected. A management measure is ready when its arithmetic, evidence status and permitted use are connected, including the condition that can prevent a commitment despite a favourable result.

Meridian completes a metric revision entry when a customer-value report changes its window. Prior purpose: screening a standard acquisition relationship inside the supplied age horizon. New proposal: compare it with an extended-window presentation. Tomas records that additional age exposure changes the quantity, while Ravi retains the customer and activation mapping. The bridge must show whether the apparent improvement comes from the window, a customer assumption or a cost perimeter. If the underlying history cannot be reconstructed, the entry discloses noncomparability rather than inventing a restated series. This instrument makes revision control a diagnosis of changed meaning, instead of another general instruction to reopen every metric when anything changes.

The controller should retain the prior method and the reason for change. If a historical comparison can be reconstructed under the new method, the report can show that bridge with its source records. If it cannot, the limitation should be stated. Inventing a restated history to maintain a smooth chart would undermine the very transparency the metric contract is intended to provide. The objective is a truthful decision record, not visual continuity at any cost.

The metric contract then compares null, missing and zero through their decision consequences. The unavailable break-even denominator in the adverse-margin case is a defined model result. An absent service acceptance sheet is missing operating evidence. Zero acquisitions is a supplied action state and still leaves the hypothetical unit curve available. Formatting all of these as a blank or zero would erase their different explanations. The fictional display disposition preserves the unavailable result, identifies the absent record and states that the unit curve does not create an acquired business. This completed classification test connects a seemingly small presentation choice to the specific inference it can corrupt, without changing the canonical calculations.

The final contract-use test asks a reader to reconstruct an exported quantity from its named fixture, meaning and allowed use. Reproducing the amount while treating the horizon value as enterprise valuation fails interpretation. Explaining its limitations while being unable to locate the input and output fails reconstruction. Correctly identifying both but relying on absent funding fails the commitment branch. These are distinct defects and require different repairs. Meridian's fictional controller therefore retains a compact method link, the perimeter that matters for the intended decision and the event that invalidates that use. The contribution is an operational metric acceptance test, not a universal governance claim or an author-issued assurance grade.