Understand why one action appears before another
Keep the Action Center goal and queue context visible while reviewing one proposed action.
Several dimensions
Deterministic selection
Inspectable evidence
Read each action as a decision record
- Action family, category, subject, and exact destination
- Priority tier and guide band: Do now, Do next, Confirm, or Monitor
- Expected impact, goal fit, feasibility, urgency, evidence confidence, time to value, effort, and irreversibility
- Business-fit and relative-rankability state where the action requires them
- Source facts with value, unit, observation time, freshness, and limitations
- Prerequisites, missing-data needs, implementation steps, and verification code
- Evidence fingerprint and profile revision that make the decision reproducible
Not every detected signal is eligible for the guide
Hard blockers suppress a candidate. A contradicted business fit or market mismatch suppresses work that requires fit. Unknown fit, unknown rankability, or incomplete minimum evidence can keep a candidate visible in Confirm with explicit limitations rather than pretending certainty.
Research-only keywords with unknown business fit are not promoted into a business question. A calibration prompt is considered only when the target is already saved in Keyword Database or measured in Position Tracking, and the report admits at most one such question.
The dimensions answer different questions
| Dimension | Current meaning |
|---|---|
| Expected impact | A relative demand or scope signal, not forecast traffic or revenue. |
| Goal fit | How closely the action family matches the selected operational lens. |
| Feasibility | Relative rankability when available, otherwise a bounded family prior. |
| Urgency | How quickly delay can matter for this action family and evidence. |
| Evidence confidence | Coverage and confidence of the facts used, not confidence in a guaranteed outcome. |
| Time to value | A policy prior for how soon useful evidence can appear after implementation. |
| Effort | A rule-based implementation estimate, not a team-specific hour quote. |
| Irreversibility | Extra caution for actions such as content consolidation that can be costly to undo. |
Utility is transparent, but it is still a heuristic
Within a comparable tier, utility adds 28% expected impact, 22% goal fit, 18% feasibility, 12% urgency, 10% evidence confidence, and 10% time to value. It subtracts 10% effort and 10% irreversibility risk. Critical loss or blocker tiers use an additional immediate-priority view built from current-loss risk, severity, enabling value, and time sensitivity.
The number displayed internally for list order is not a calibrated probability and is not a promise of uplift. The public card deliberately converts dimensions to plain bands and keeps the underlying evidence available for review.
Keep priority, tier, band, and order separate
Protect now and Unblock lead the tier order, followed by Quick progress, Strategic growth, Confirm or measure, and Monitor. A prerequisite or unresolved limitation moves an otherwise strong candidate into Confirm. Monitor stays separate from active implementation. No more than two actions keep the Do now label in one guide.
Impact is Very high at 85, High from 68, Medium from 48, and Low below that. Urgency is Now from 75, Soon from 50, then Later. Confidence is Strong from 80, Good from 60, then Limited. Effort is Easy through 35, Medium through 65, then Hard. These cutoffs make labels consistent; they do not remove the need to inspect the raw fact.
Two kinds of priority can coexist
The goal lens affects server admission and ordering. Growth, Protect, Technical, and Quick wins each admit a different subset, while critical blockers can cross that boundary. The client then gives saved High-priority work items precedence and applies a smaller goal-affinity adjustment before retaining the guide's deterministic order.
Changing a Board item from Normal to High changes personal work order. It does not rewrite source evidence, engine dimensions, or another teammate's board. Likewise, switching the goal does not delete the saved item or its captured snapshot.
The guide removes repetition before it removes nuance
Candidates sharing one practical action are deduplicated. Dependency cycles, hard conflicts, and actions dominated by a clearly better candidate in the same family, lane, tier, and evidence neighborhood are suppressed. The policy also caps repeated context questions and selected data-refresh families.
The remaining order receives penalties when one action family, lane, or source has already appeared repeatedly. The five-item focus portfolio adds another diversity pass. This is why the first five are not always the five largest raw impact values.
A source badge is only the start of the evidence trail
Each evidence observation has a stable ID, source, subject, predicate, value, unit, observation time, projection time, freshness state, provenance, adapter version, and limitation codes. The public recommendation can carry up to 12 observations; the focus disclosure shows a smaller configured subset for readability.
Facts are projected from stored product evidence. Several projections are intentionally lossy or bounded. A missing field can therefore mean not collected, not authorized, too old, outside the projection cap, or unsupported by the adapter. It does not automatically mean zero.
For page-level technical evidence, open the URL dossier and check the stored facts behind the recommendation.
Know what each source can and cannot prove
| Evidence family | Review boundary |
|---|---|
| Keyword demand and intent | Keyword Database or Research can support volume, difficulty, intent, provenance, and candidacy. Monthly trends are not fully projected into the guide. |
| Relative competition | Domain and competitor evidence can support domain strength, gaps, and aggregate SERP referring-domain context. It is not a fresh SERP for every keyword. |
| Ranking movement | Position Tracking supplies comparable current and previous measurements. A single position cannot establish cause. |
| First-party performance | Search Console facts appear only from an authorized stored snapshot and are bounded to approved keyword or page facts. Connection status alone has no clicks or impressions. |
| Technical findings | Site Audit supplies run-bound severity and affected scope. Open the issue and affected URLs for the complete finding context. |
| Internal links | Several stored edges to the same target become one action. The displayed count is the number of bundled recommendations, not an edge score disguised as a count. |
| Content execution | Plan and page status can support a handoff action. They do not prove publication, indexation, or ranking until those states are measured. |
Rankability compares available evidence, not certainty
The relative model weights domain-strength match and SERP accessibility at 25% each, topical proof at 20%, target-page readiness at 15%, ranking momentum at 10%, and intent compatibility at 5%. Domain strength versus difficulty and aggregate SERP referring domains are mandatory for a numeric score. Missing either leaves rankability Unknown.
Available evidence is reweighted, but low coverage or confidence caps the score below the Competitive band. Current bands are Strong from 80, Competitive from 60, Stretch from 35, and Blocked below 35. Page readiness may be only a ranking proxy, so read the limitation codes before treating the score as an implementation decision.
Acquire more data only when it can change the decision
Data needs are grouped by missing fact, linked to affected candidates, and ranked by possible decision change, interaction cost, latency, staleness risk, and known price. Needs below the current decision-change threshold or with non-positive utility are left out of the user-facing list.
A billable SERP comparison has no invented estimate inside the decision policy. It requires a live quote in the enrichment flow. A free refresh is still explicit because it changes source evidence. Waiting for ranking history costs no acquisition balance, but it also cannot be rushed into a factual trend.
Distinguish the built-in plan from optional AI interpretation
Every action has a deterministic three-step implementation outline selected by action family plus a verification statement. It is intentionally generic enough to remain grounded in the known destination and evidence.
Optional paid action intelligence uses the exact active fingerprint and a bounded, sanitized evidence packet. It can add a tailored assessment, steps, verification method, review timing, and missing-data notes. It cannot change the queue order, rewrite the underlying facts, or turn a weak source into strong evidence.
Follow prerequisites before opening the implementation surface
A recommendation can point to a keyword detail, research result, domain, competitor, tracked keyword, audit finding, internal-link opportunity, content execution item, Search Console setup or refresh, business-context review, or a manual external instruction. The link carries the known entity IDs so the source product can reopen the right scope.
If a prerequisite is unfinished, the action belongs in Confirm even when its impact is high. Finish or verify the prerequisite first, then return to a current guide. A destination link is a handoff to investigation and execution, not proof that the proposed change is safe for every template or market.
Saving freezes the decision context for later work
A saved item captures candidate identity, source fingerprint, destination, source, engine priority, guide band, verification code, and a bounded baseline from the matching decision snapshot. If the candidate disappears from a later guide, the Board labels it Historical instead of silently replacing the old claim.
Start can promote a saved item to In progress. Completed records a user state and a future review time. Archived removes it from the active Board. Reopening an archived item captures the current recommendation again, while saving an already-open item does not silently overwrite its original snapshot.
Example: a missing or conflicting canonical pattern
Why the next action can change
- A new crawl, rank measurement, Search Performance snapshot, research result, or domain snapshot changes the evidence fingerprint.
- A confirmed offering, market, exclusion, or target override changes business-fit admission.
- A different goal lens changes eligible families, tier order, and local affinity.
- A completed prerequisite can move dependent work from Confirm to an actionable band.
- Deduplication, conflicts, diversity, and guide caps can change when a stronger candidate appears.
- A personal High priority can move a saved action forward without changing the engine's evidence.
Review the recommendation before creating a commitment
Example: A conflicting canonical pattern
An action flags a conflicting canonical pattern. The strategist opens the Site Audit run behind it, confirms the issue, its severity and the affected scope, and samples representative URLs.
Some differences turn out to be an intentional template rule. The team fixes the rest at template level and schedules a comparable crawl as the verification the action already names.
Frequently asked questions
How is the utility of an action weighted?
Within a comparable tier, utility adds expected impact (28 percent), goal fit (22), feasibility (18), urgency (12), evidence confidence (10) and time to value (10), and subtracts effort and irreversibility risk (10 each). It is a transparent heuristic, not a forecast.
What are the priority tiers?
Protect now and Unblock come first, followed by Quick progress, Strategic growth, Confirm or measure and Monitor. Missing prerequisites or open limitations move a strong candidate into Confirm.
How is rankability estimated?
It weights domain strength against difficulty and SERP accessibility at 25 percent each, topical proof at 20, target page readiness at 15, ranking momentum at 10 and intent compatibility at 5. It compares available evidence and is never a ranking guarantee.