Shaduf.Research preview
Marketing Skills Catalog/Ad Campaign Analyzer decision boundary

Research report · 7 October 2026 · Run 17

Ad Campaign Analyzer leaves the active chooser

The selected source instructs immediate stop/scale decisions from export metrics. Its profitability and significance labels are not supported by the inputs it requires. This is an original-source decision audit, not a package or campaign test.

What was actually checked

The mutable selected SKILL.md was obtained on 7 October, displayed version 1.0 and was returned as crawled today. The selected-directory listing was about five months old and cannot establish current inventory. The README was returned as crawled today and the license as crawled last week. Selected-file SHA/date, tag attribution and current package contents remain unknown. This reading does not clear Run 5’s stop before installation advice.

Source inputExport or pasted table, target CPA and AOV. Date range, conversion definition/value, currency, attribution window, lag and non-media costs are not required.
Source outputTraffic-light grade, “Immediate (Today)” stop/scale list and a 20%-every-48-hours budget plan. No direct account API or credential use is stated.
Pool decisionDo not route users to that action plan. A governed worksheet can recompute descriptive metrics and expose unknowns; an authorized owner separately decides action.

Why the source rule is not a spend rule

ComponentSelected source saysLimit and pool-added control
Profit and ROAS“Profit per conversion” = AOV − CPA; “ROAS” = AOV/CPA; ROAS >1 is profitable; break-even CPA = AOV.AOV/CPA is only a proxy for conversion value/spend when values and counted populations match. Actual reported value/cost uses conversion value and cost; AOV − ad CPA omits goods and other costs. The pool requires a named analyst to establish contribution economics. Value/cost definition · ROI explanation.
Color and significanceRed can trigger immediate stop on one threshold or low CTR; Green generally uses 10+ conversions and scales budget by 20% every 48 hours.The file calls ten conversions “statistically significant” without a test. Red/Yellow rules overlap and leave a CPA gap; conflicting metrics have no defined precedence. These thresholds and cadence are source instructions, not validated decisions. The pool requires denominator, attribution, lag and statistical-adequacy checks before advice. Conversion lag · Learning.
Attribution and benchmarksSingle-channel last-click is described as sufficient; generic CTR, CPA, LTV:CAC and fatigue bands are supplied.The selected file supplies no supporting cohort, network, date or controlled causal test for those bands. Attribution model and conversion window can change reported comparisons; the pool requires those definitions and unknowns to be shown, not filled with generic bands. Attribution models · Conversion windows.
Release authorityA dated action checklist and numerical scaling plan; no account connector or mandatory analyst sign-off stated.No account action is granted by the text, but the recommendations are consequential. A paid-media owner must verify the source-to-output numbers and separately authorize any budget/account change. This release gate is imposed by the pool, not enforced by the package.

What to do instead—and what could change this decision

For a permitted export, record source, period, metric definitions, denominators and missing fields in a governed spreadsheet or analyst worksheet. Recompute descriptive CTR and CPA; mark conversion lag, attribution, margin/LTV and reconciliation unknown where they are unknown. Stop before inferring profit, causality or a budget move. An authorized specialist reviews any subsequent decision.

Reconsider a card only if a fully enumerated, attributable selected revision removes the unsupported profit/significance/20%-rule claims, or makes them conditional on correct metric definitions, required attribution/window/lag and margin inputs, coherent uncertainty states, source-to-output reconciliation and mandatory named-human release. Compare its permitted output against the same authorized export reviewed in a governed worksheet. A source revision alone would not establish installation safety, host compatibility or marketing lift.

Historical record and limits

Run 15 accurately recorded the then-visible source instructions; its six-row matrix is historical, not the current chooser. Runs 1, 4 and 7 also described the source rather than validating its advice. No package was installed, cloned, copied or executed; no private export, account, host or marketing outcome was tested; no spend action was recommended or taken. The research library retains all dated reports.

Current catalogue · Browse by job · Methods and limits

Search published pools, pages, reports, and evidence.