Make the snippets again instead of reusing stored ones.
Competitor page limit
Page Mode only. Each competitor page is cut to this length before it goes to the models. The median page is about 17,000 characters. A higher limit keeps very long pages complete but costs more per round. The target page is never cut.
Single runs
Loading…
New Labs Batch
i
Pick what to sweep — each item becomes a child run with target URL auto-detected via preflight. Entity sweeps use the entity name as the SERP query.
Lines without a `|` are skipped. Duplicates collapsed.
Options
Advanced Options
Page copies
Force refresh
Make the snippets again instead of reusing stored ones.
Client's own pages in results
If another of the client's pages also ranks for a query, batches never stop to ask — this sets what they do. The other owned page is left out of the competitor set either way.
Competitor page limit
Page Mode only. Each competitor page is cut to this length before it goes to the models. The median page is about 17,000 characters. A higher limit keeps very long pages complete but costs more per round. The target page is never cut.
Batches
Loading…
Run history
Every run for this property, newest first. The Trend column shows how each run's rank moved across rounds. The Batch column shows membership; click a batch link to open the batch detail page.
Loading…
Batches
Every batch for this property, newest first. Click a batch to open its full detail view.
Loading…
Pages
Every page with optimizer runs, most runs first. Deleted runs are not counted. Click a page to see its Search Console timeline, its runs and the curator.
Loading…
Open in Pages to log page events and see their before and after.
Search Console
Runs
Tick the runs the curator should blend. Deleted runs and runs with no rounds start unticked.
Loading…
Curator
The curator reads every edit the ticked runs tested, with what each one did to the rank, and blends them into one merged brief and one blended page.
Ablation runs
Solo runs
Loading…
Draft runs
Loading…
Image runs
Loading…
Recent checks
Loading…
Saved optimizations
Loading…
Recent runs
Loading…
status —baseline —tokens –cost –page
Keep going
Run more steps from where this left off. Add a tip to steer the next attempt.
Test a caption
Score any caption against this image without a full run.
#
Match
Δ
Caption
Step
New Image Optimization
i
Options
Page images
i
Page scans
Loading…
Bulk Image Optimizer
i
Bulk jobs
Loading…
Image runs
Loading…
Page scans
Loading…
Loading…
Narrative
Plain-English summary of what this ranker has learned for this property, written from the same cross-run evidence shown in the Positive, Negative and Attribution tabs. Every summary you generate is kept — each one is stamped with its date and the tags it covers, so the same tags written up months apart can be read side by side.
Pick a ranker above and the narrative will load.
Positive Rank Factors
Cross-run evidence on which claims and rank factors moved the needle in the right direction for this property + ranker. By rank = SERP-position movement vs each run's baseline. By posterior = Bayesian belief shift on the claim (final μ vs initial prior).
Pick a ranker above and the table will load.
Negative Rank Factors
Cross-run evidence on which claims and rank factors hurt the target's rank or whose posterior fell below their prior. Same axes as Positive: by rank vs by posterior. A claim with mixed rounds appears in both Positive and Negative tabs in different sub-tables.
Pick a ranker above and the table will load.
Attribution
Descriptive rollup of claims by Rank Factor Class and Rank Factor for the selected ranker. Shows what's been tested + how often it's winning. Not a Bayesian posterior over the tag — claims within a class aren't exchangeable per locked design.
Per Rank Factor Class
Pick a ranker above and the table will load.
Per Rank Factor
…
Active taxonomy
Currently-active classes + factors (the choices the classifier sees). Class-only rows describe the class itself; factor rows are the assignable choices.
Loading…
Factors
Each model's belief that a factor helps rank (win-rate posterior, with trial count), the gap between models (Δ), the confidence they genuinely differ, and the pooled provider belief. A win = an edit that reached a new best rank.
Sort
Pick a provider above and the table will load.
Narrative
Plain-English write-up of what this provider rewards across its models and — the point — where the models diverge, built from the Factors table for the current scope. Generated on demand and cached.
Pick a provider above and the narrative will load.
Classify a title or a full article as genuine substance, template-driven fluff, or ambiguous. Every check is saved to this property.
This tool is designed to evaluate article content. It can separate flat uninspired pieces you see on blog networks from true blogger and journalist pieces written with purpose. It may misclassify dry, factual pages such as product and service pages.
New ablation
i
Advanced options
Ablation runs
Loading…
Edit Run Label
network_intelligence
Auxy
Add credits to continue
This runs on pay-as-you-go credits. Add credits to unlock it.
Billing for this project is handled by someone else