Proof Packs: Linking Every Number to Its Evidence

Every advertising report makes a request of its reader: trust this number. Trust that the impressions were real, that the conversions were caused, that the precision was achieved, that the spend went where it says. Usually the only thing backing that request is the authority of the dashboard itself — a clean figure in a clean box, computed somewhere you cannot see, by a party with an interest in the answer.

A proof pack inverts the request. Instead of asking you to trust a number, it lets you check one. It ties every reported claim to the underlying, event-level evidence behind it, and attaches an honest grade of how confident the system is in that claim. This is reporting as an evidentiary discipline rather than an act of faith — and it is the natural endpoint of owning your data.

The trust deficit a dashboard can't fix

The advertising industry has a measurement credibility problem, and it is structural. When the same party sells the media, computes the metric, and presents the result, the buyer is asked to accept a self-graded exam. Even with the best intentions, a dashboard:

• shows a derived number, not the events it was derived from;
• uses the vendor's definitions of "viewable," "converted," or "in-geo";
• offers no way to reproduce the figure independently; and
• presents every number with the same flat confidence, whether it rests on a million events or a handful.

None of these are fixed by a nicer chart. They are fixed by changing what a report is — from an assertion into a package of evidence.

What a proof pack actually is

A proof pack is a reporting artifact in which each claim carries three things: the number, the evidence that produced it, and a confidence grade. The pattern is concrete in Ad360's hyperlocal reporting, where proof-led reporting is built in rather than bolted on. The verification surface includes:

• Fence-level proof packs — for each geo-fence, the evidence behind what was delivered: impressions, unique users, and domain/app transparency.
• Precision-grade distribution — not a single "precision targeting" claim, but the spread of achieved precision across the campaign.
• A Geo Precision Grade with a confidence level — surfaced in reporting at fence and line-item level, so each claim says how sure it is.
• QA exposure logs and a test mode — the audit trail and the means to validate before trusting.

The principle generalizes beyond geo. Any reported outcome — a delivery figure, a click, a conversion — can be presented this way: here is the claim, here is the event-level evidence, here is how confident we are.

Confidence grading: not all numbers are equal

The quietly radical part is the confidence grade. Conventional reporting presents every figure with identical authority. A proof pack refuses to. A precision claim backed by dense, high-accuracy signal is graded High; the same claim built on sparse or low-quality signal is graded Medium or Low — and the grade travels with the number.

This does two things at once. It protects the buyer from over-trusting a thin result, and it protects the vendor from over-promising one. A graded number is a more honest number, because it admits the difference between "we measured this well" and "this is our best estimate." (It mirrors how the underlying precision system itself grades achieved accuracy in tiers with High/Medium/Low confidence — the report inherits the discipline of the engine.)

Proof packs need owned data underneath

A proof pack is only as real as the evidence it can point to — which is why it cannot stand on aggregates alone. To link a number to its evidence, that evidence has to exist and be accessible: the event-level records of auctions, impressions, clicks, and conversions. This is the bridge to data ownership.

When a client holds raw, event-level data in their own storage, joinable on a shared key, the proof pack stops being something they take on faith from the vendor and becomes something they can reconstruct themselves. The two ideas reinforce each other: transparency-by-architecture supplies the evidence; the proof pack is the disciplined way of presenting and grading it. Aggregates can be illustrated by a dashboard; only event-level evidence can be audited by a proof pack.

What proof packs change about the conversation

• Disputes become checkable. A disagreement over a number becomes a question about evidence, not a standoff between dashboards.
• Confidence is explicit. Decisions can weight a High-confidence result differently from a Low-confidence one, instead of treating all figures as equally solid.
• Overclaiming becomes visible. A vendor that grades honestly cannot quietly present a weak result as a strong one; the grade exposes it.
• Measurement maturity rises. It pushes the whole conversation from "what does the dashboard say" toward "what does the evidence support" — which is where serious measurement (attribution vs. incrementality, for instance) has to live.

Common misconceptions

• "A detailed dashboard is a proof pack." Detail is not evidence; a proof pack links each claim to the underlying events, not just more aggregates.
• "Confidence grading makes us look weak." It makes the strong numbers more credible by distinguishing them from the weak ones.
• "Proof packs are only for hyperlocal." Geo is where the pattern is most developed, but any reported outcome can be made evidentiary.
• "This is just transparency rebranded." Transparency is access to data; a proof pack is the disciplined presentation of that data as graded, claim-linked evidence.
• "If the number is right, the proof doesn't matter." In a low-trust market, an unverifiable right answer and a wrong one look identical to the buyer.

What good operation looks like

• Present each material claim with its number, its evidence, and a confidence grade.
• Grade honestly — let weak signal produce a Low grade rather than a confident-looking figure.
• Build proof packs on owned, event-level data, so claims can be independently reconstructed.
• Include the audit trail — exposure logs, test mode, distributions — not just headline figures.
• Use grades to drive decisions, weighting high-confidence evidence more heavily.

Open questions

• Can the industry agree on a standard confidence-grading scale, so grades are comparable across vendors?
• How should confidence be computed and validated against ground truth rather than asserted?
• What is the minimum evidence a proof pack must contain to be genuinely auditable rather than merely detailed?

A dashboard is a closing argument with the evidence withheld. A proof pack is the evidence, organized, graded, and handed over for inspection. In a market where measurement credibility is the scarce resource, the platforms that win trust will not be the ones with the most confident dashboards — they will be the ones willing to show their work, number by number, and to say plainly how sure they are. That is reporting that can be checked. Everything else is reporting that must be believed.