Executive Docket and Attention Compression

1. The real problem: executives need fewer decisions, not more messages

When email, tickets, alerts, chat and tasks accumulate, naive summarization merely shortens text. It can still show the same obligation four times or elevate a noisy low-impact item. An executive docket compresses many signals into a small number of decision items while preserving why each item matters and how to drill back to evidence.

2. Lesson boundary and owned artifact

The owned output is Executive Attention Docket & Compression Ledger v1. Each row records underlying_object, source_refs, decision_needed, actionability, urgency, impact, confidence, owner, next_action, due_or_recheck, duplicate/suppression data, priority_band, risk_if_ignored, and drilldown_ref. Notification Routing belongs to PLN-006; inbox privacy belongs to PLN-008.

3. Do not confuse a signal with a decision item

Four messages about one obligation are not four executive decisions. Normalize signals first, then group them by the underlying decision, obligation, or opportunity. Message count and channel count are observations, not units of executive attention.

4. Compression must be reversible

If an executive sees a short summary but cannot recover the evidence that produced it, the compression is not auditable. Every item needs source_refs and a drilldown_ref. Provenance is therefore part of the decision contract, not decorative metadata.

5. Actionability comes before urgency

Google SRE guidance says alerts should be actionable and that alerts with no expected human action create noise. Here that is used only as an operational boundary: if decision_needed or next_action is unclear, high urgency alone should not push the item to the top of the docket.

6. Separate immediate work from work that can wait

Google SRE distinguishes alerts that need immediate human action from tickets that need human action later. This lesson does not copy that taxonomy; it borrows the separation between need for action and response window. A consequential item can still belong in a later review window.

7. Impact, urgency, and confidence are different dimensions

impact asks what happens if the issue is real. urgency asks how short the response window is. confidence asks how strong the current evidence is. A dangerous shortcut is treating high confidence as high impact, or treating low confidence as a reason to hide a potentially severe risk.

8. Build priority bands from explicit dimensions

Priority should not emerge from an opaque single score. Examine actionability, urgency, impact, confidence, risk_if_ignored and due/recheck together. If a decisive dimension is unresolved, REVIEW_REQUIRED is more truthful than a false precise ranking.

9. Duplicate suppression must preserve evidence

If one obligation appears in email, chat, CRM and a notification, create one docket item, set duplicate_count=4, and record suppression_reason. Suppression removes repeated attention demand, not provenance; all relevant source references remain available for drill-down.

10. Owner and next action must be executable

Phrases such as “the team should follow up” are too vague. The owner must be a specific role in the fixture and the next action must be performable. If ownership is unresolved, an item may remain REVIEW_REQUIRED even when impact is high.

11. Due/recheck conditions prevent endless review states

Every item needs either a due time or a finite recheck condition. REVIEW_REQUIRED without an exit trigger becomes an unbounded queue. “Recheck privacy authority at 14:00; escalate if still unresolved” is auditable. “Review later” is not.

12. Worked example: 26 messages, only two decisions

Suppose 26 new messages arrive. Four messages and two notifications repeat one contractual obligation; seven are low-impact status updates; one payment issue needs a decision today. The ledger compresses duplicates into one item, defers or excludes routine status noise, and creates an explicit decision and next action for payment. Raw message count never determines priority by itself.

13. Keep four decision states explicit

INCLUDE_NOW means enough evidence supports the item in the current attention window. DEFER means action is needed but not now. REVIEW_REQUIRED means the item may matter but decisive evidence, ownership, or provenance is incomplete. EXCLUDE means there is no bounded executive action, or the signal is informational/duplicate-only.

14. Deterministic failure probes

Four shortcuts must fail: (1) the same obligation repeated across four channels becomes four docket rows; (2) a high-confidence informational item outranks low-impact/actionability limits; (3) a high-impact privacy concern is hidden only because confidence is incomplete; (4) a summary loses drill-down provenance. Each must change decision or rationale rather than relying on raw counts.

15. Transfer: low-confidence high-impact risk beside routine high-confidence work

In transfer, several routine tasks have high confidence while one privacy concern has incomplete evidence but potentially high impact. Do not copy the baseline order. Recompute impact and risk independently from confidence, preserve missing evidence, and create a finite recheck/escalation path. The goal is to keep uncertainty visible without hiding risk.