Landing Page Experiment Planner

Turn supplied landing page audit findings into a reviewable experiment backlog. The Actor connects a source observation to one proposed change, a primary conversion metric, a guardrail and explicit prerequisites. Agencies can use the output after an audit to prepare a client experiment brief.

What it does

The planner accepts up to 50 page audits and supports four finding IDs from the factory's landing page auditor. message_match produces a hero headline proposal tied to the supplied approved promise. cta produces a primary action proposal. form_length produces a conversion form proposal whose relevance the owner must review. trust_signals proposes placement of authentic evidence supplied and cleared by the customer.

Each experiment retains the original finding ID, status and message. Priority reflects source severity. Access, HTTPS, privacy, consent and analytics findings produce prerequisite fixes. PASS and INFO findings produce skipped records. Unknown IDs are preserved in the skipped summary.

Input

Provide conversionGoal as lead_submission, purchase or signup. Set measurementReady only after confirming the actual conversion event, eligible visitor denominator, assignment and guardrail measurement. This declaration remains customer supplied evidence.

Each page needs a unique pageId and a findings array. Findings use id, status and message. Status is exactly PASS, WARN, FAIL or INFO. Supply adPromise to prepare a hero headline variant. Set verifiedTrustEvidence only after authentic evidence and permissions have been reviewed.

The limits are 50 pages, 40 findings per page, 2000 Unicode code points per message and 1,500,000 UTF-8 bytes for the complete input. The Actor validates nested objects and duplicate IDs. URL identifiers are retained with query strings and fragments removed. Embedded URL credentials are rejected.

Output

One dataset row contains the complete JSON report, including pages, experiments, prerequisites and skipped. EXPERIMENT_PLAN stores the same report. EXPERIMENT_CSV contains one row per proposed experiment. REPORT provides a readable summary.

A card includes the source finding, control, proposed change, hypothesis mechanism, primary metric, guardrail, prerequisite IDs and an applicability review. readyForReview describes supplied prerequisites. Every card still requires review, and effectExecuted remains false.

CSV cells that resemble formulas and numeric page IDs receive a leading apostrophe for spreadsheet import. The JSON report retains original page IDs. Import the ID column as text when IDs resemble dates or scientific notation.

Pricing and cost control

One experiment-plan-report covers up to 50 supplied landing page audits. The Store displays the current price for the complete report before a run. Send your audit JSON to prepare cards with source evidence, metrics and prerequisites.

PRICING.json preserves the original USD 0.03 report proposal and USD 0.002 engineering planning allowance. Release receipts record price activation, cloud observations and public readback. The adjacent historical benchmark uses a different unit. Operating cost and margin require measurements for the intended workload and storage retention.

The runtime caps delivery at one dataset row, zero HTTP requests and 45 seconds of processing. The proposed cloud configuration is 256 MB and 60 seconds. Dependency versions are locked. Local acceptance uses the installed Apify SDK.

Limitations and fair use

Results are deterministic rules over customer supplied findings. Source observations and readiness declarations require validation by the owner. The planner produces draft experiment cards. Sample size, experiment duration, stopping rules and causal conclusions require an appropriate experiment design.

The audit's detected form may differ from the conversion form. Confirm applicability before approving a field change. Claims, testimonials and guarantees need authentic evidence and permission. Data should be limited to authorized page observations.

Resume and error handling

PLAN_DELIVERY records PREPARED before delivery and COMPLETE after one confirmed dataset write and the expected named event. A completed repeat validates JSON, CSV, the readable report, runtime, pricing and the physical dataset. It returns a reused result with zero additional emission.

RECONCILIATION_REQUIRED preserves the existing storage after uncertain delivery or changed input. Review the prior run and its artifacts before another operation. PRICING_UNVERIFIED stops before delivery when the charging model is unavailable. INVALID_INPUT identifies a bounded input contract violation without echoing supplied content.

Example

Use the exact input in examples/input.json, also provided as the input form prefill. It includes a lead page with four supported findings and a second page whose access failure requires repair. The resulting report contains four experiment cards and four prerequisites.

For a fresh checkout, run npm ci --ignore-scripts --no-audit --no-fund, then npm test. Start with npm start and a new local storage directory containing INPUT.json. Factory acceptance used the reviewed offline dependency cache. The SDK entry point retains storage so that a verified completed replay can be inspected.