agentleFS
MadalaVijayMadalaVijay/marketing-skills-deep/skills/data-trust/SKILL.md

data-trust

Skill0 starsChanged 2 months agoMITView on GitHub

Decide whether a marketing number is good enough to make a decision on, before making it. Audits tracking coverage, source taxonomy, deduplication and platform-versus-record gaps. Use before any budget reallocation or channel comparison.

The file

We split this skill into the rules it gives an agent

1 rule. About 0 are common, 1 is distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    When a series drops to zero without a business explanation, check the field names before checking the market

    line 77

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Data trust

The most useful sentence this skill can produce is "these numbers are not decision-grade." Say it when it is true.

The four failures, in order of frequency

1. Taxonomy gaps. The source field has no value for organic, direct or referral, so those conversions fall into whatever the form defaults to — usually the paid bucket. This flatters paid and starves organic, and it is by far the most common failure.

Audit: list every distinct value in the source field with its share. If organic, direct and referral are absent or implausibly small, every channel number downstream is wrong.

2. Missing tags. Measure the share of sessions or leads arriving with no UTM at all. Above roughly 10-15% and channel splits are indicative, not decisive.

3. Double counting. Two platforms claiming the same conversion, or a pixel firing twice. Symptom: platform-reported conversions summed across channels exceed what the business actually recorded. Deduplicate on an order or transaction id, never on a session.

4. Value corruption. Counts reconcile, revenue does not — usually a pixel sending a static value on every event. Check both, separately.

5. Counting rows instead of events. Operational logs are usually append-only: a record is written every time something is touched, so one underlying event appears as many rows. Row count then overstates reality by whatever the average touch count happens to be, and the multiple is not stable over time.

Never report a row count from an operational log as an event count. Count distinct keys — the order, the customer, the date, whichever identifies the thing once — and state which key you used. If no such key exists, that is the finding.

The same trap in calendar form: counting filled cells rather than distinct periods. A schedule with one row per participant per occurrence counts an occurrence once per participant, not once.

Reconcile against something the platform does not control

Ad platforms are graded by their own homework. Pick a source of truth deliberately — the payment processor, the CRM, or the operational record the business actually runs on — then compare, for the same window and definition:

Under 10% gapNormal. Attribution windows and restatement explain it
10-25%Investigate before reallocating
Over 25%One of the two is wrong. Find out which before anyone moves budget

Traps

  • Referrer is not channel. Traffic arriving from a social domain is often a paid placement on that platform. Only the UTM medium separates organic social from paid social. Diagnosing "organic is working" from referrer alone is a common and expensive error
  • Compare like windows. 7-day-click and 1-day-click answer different questions. Pick one and hold it
  • Yesterday is provisional. Attribution restates for days
  • Two events with similar names are not the same event. A generic Lead and a product-qualified Product Lead measure different populations, and merging them to tidy a report destroys the only distinction that made either useful. Before combining any two events, confirm they fire from the same trigger on the same population — not merely that the names look alike
  • A field name that changes breaks every downstream parser silently. An integration that labels a value Phone: and later Phone Number: will keep delivering data while anything matching the old label quietly reads empty. When a series drops to zero without a business explanation, check the field names before checking the market
  • Absence of a source value is not absence of the channel. It usually means nobody added it to the form

Output

The taxonomy audit as a table of values and shares. The untagged share. The platform-versus-record gap per channel with a verdict. Then one line: are these numbers safe to decide on, and if not, the single fix that would make them so.

See also form-friction, which depends on this skill's verdict before any completion rate can be believed.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.