Skip to main content

​
Daily Brief

The Daily Brief is Alleato’s source-backed executive read of the previous business day. It turns recent meetings, email, Teams messages, and documents into one portfolio-level assessment, the decisions that need executive attention, and a current intelligence record for every project with activity. The Daily Brief is designed to answer four questions:
  1. What changed across the portfolio?
  2. What needs an executive decision now?
  3. What is each active project doing, waiting on, or exposed to?
  4. What should become durable project intelligence, a task, or a follow-up?
The Daily Brief reads every source record included in its configured business-day window. It does not sample from that included set. This is different from claiming that every real-world message has already been ingested: connector health and source synchronization remain upstream requirements.

​
Where to find it

  • Daily Brief history: /daily-briefs
  • A specific brief: /daily-briefs/[briefId]
  • Daily Deep Read review: /executive/daily-deep-read-review
  • Project intelligence: /{projectId}/intelligence
Daily Brief history and source-level detail require the appropriate executive access.

​
How the workflow operates

​
1. Select the business-day window

The scheduled compiler runs at 6:00 AM Eastern on weekdays. Unless a date or custom range is supplied, it compiles the previous completed business day in the America/New_York time zone. Each potential source is evaluated against the covered window:
  • Meetings use their source or transcript timestamps.
  • Email uses message or source timestamps.
  • Teams conversations use the timestamps embedded on individual messages when available.
  • Documents use their source or synchronization timestamps.
Rows outside the window are excluded and recorded as skipped evidence. If records exist in a source lane but none can be included, the compiler fails instead of quietly presenting that lane as empty.

​
2. Materialize and inspect the source set

The compiler builds a source manifest and a full source corpus before asking the model to write the brief. Each included record carries:
  • a stable source identifier and short citation alias;
  • source type;
  • title;
  • source timestamp;
  • project ID and project name when known;
  • original source link when one is available;
  • the source text used for analysis.
Duplicate records are removed before writing. Project attribution is checked against source titles, and detectable mismatches are corrected before synthesis. The full source text remains in the evidence corpus. Individual model inputs are capped at 12,000 characters so one unusually large thread or transcript cannot consume the entire context window; a truncation marker makes that limitation explicit.

​
3. Deep-read every included source

Sources are grouped into meetings, email, Teams, and documents. Every item in each non-empty group is passed through the source-reading stage.
  • Email is processed in batches of 20.
  • Other source lanes are processed in batches of 8.
  • Up to four batch workers run concurrently.
The source-reading prompt extracts decision-grade information only: decisions, risks, money, schedule movement, commitments, blockers, and owner-relevant context. Every extracted fact must retain its exact citation alias. The next synthesis step uses those cited lane notes to produce a detailed executive report. If the report is missing required sections or sufficient inline citations, the compiler retries with the exact quality failure. It does not silently accept an uncited report.

​
4. Generate the structured brief

The detailed report is converted into a structured Daily Brief with the following sections.

​
Executive signal

The opening assessment contains:
  • Headline: the most important portfolio truth today;
  • Why it matters today: why the signal deserves attention now;
  • Your focus: the highest-leverage executive move;
  • Watchouts: up to three concrete risks, conflicts, or evidence gaps.
This is the only portfolio-level synthesis. It should not repeat entire project summaries.

​
Your calls today

This is a short decision index containing only decisions the executive must make. Context and team work stay in their project blocks so the same story is not repeated in multiple sections.

​
Project blocks

Every project represented by source activity receives exactly one project block. Projects are ordered by urgency. Each block contains:
  1. concrete, named action items;
  2. source-provided due dates when they exist;
  3. honest urgency language when no date exists;
  4. the cited context explaining money, schedule, commitments, risk, and recent movement.
The compiler is prohibited from inventing a deadline. If a source does not state a date, the action remains undated and explains the urgency instead. Projects that do not need an executive decision are still included. They appear in the lower-priority Also moving — nothing needed from you group rather than disappearing from the brief.

​
Loose ends and prevention findings

Loose ends hold material items that cannot be assigned confidently to a project. Prevention findings describe a repeated operational failure only when the sources support the observed condition, missing control, accountable role, recommended system, leading indicator, and confidence level.

​
5. Validate before publishing

The compiler validates the complete result before writing the canonical packet. It checks that:
  • the executive signal is present and complete;
  • calls, projects, sources, and prevention findings use the expected structure;
  • citations resolve to included sources;
  • every project with source activity has a brief project block;
  • every project with source activity has a project-intelligence record.
Structured generation can make up to five corrective attempts. If it still fails, generation stops and reports the validation error.
A polished but incomplete brief is treated as a failure. If 17 projects have activity, the system must produce 17 project blocks and 17 project-intelligence records before it can publish the packet.

​
6. Publish one canonical packet

Successful generation writes one current packet to intelligence_packets under the target daily-executive-brief. The previous current packet becomes a snapshot, preserving history without creating a second Daily Brief source of truth. The packet stores:
  • the covered business date and exact time window;
  • the source manifest and source counts;
  • the detailed report;
  • the structured brief;
  • project intelligence records;
  • compiler version, confidence, and freshness metadata.
The /daily-briefs pages, AI Dashboard, API routes, exports, and delivery helpers read this canonical packet. A normal page request does not regenerate a competing brief.

​
7. Update project intelligence

After packet creation, the Daily Deep Read consumer processes the same packet. It does not independently reinterpret the raw sources. For every project with activity, it writes the fresh packet-derived current-state projection through the controlled project-intelligence boundary. A second coverage gate runs before these writes. If even one active project lacks a current-state record, all consumer writes stop with the missing project IDs in the error. The consumer also derives:
  • source-signal candidates for review;
  • trackable tasks from structured actions;
  • project progress-report sections;
  • evidence previews and a run summary.
Project current-state coverage is automatic from the canonical packet. Candidate promotion is a separate governed step.

​
8. Review and promote candidates

The Daily Deep Read Review page shows extracted claims that need human judgment.
  • Accept means the claim is accurate and useful enough to become ready for promotion.
  • Reject means the claim should not become durable downstream intelligence.
  • Unassigned candidates should be assigned to the correct project before promotion.
The promotion process handles accepted candidates and writes the appropriate durable project-intelligence or task outcome. This human-review gate prevents an uncertain extracted claim from being treated as an approved operating fact merely because it appeared in a model response.

​
How to read the Daily Brief

Start at the top and stop when you have enough information:
  1. Read the Executive signal for the portfolio truth and recommended focus.
  2. Review Your calls today for decisions that require executive action.
  3. Open the urgent project blocks for cited context and assigned work.
  4. Scan Also moving to confirm that quieter projects were still covered.
  5. Use source links when a claim needs verification.
  6. Open Daily Deep Read Review when extracted candidates need acceptance, rejection, or project assignment.

​
Trust and failure behavior

The Daily Brief is built to fail loudly at the first incomplete boundary.

​
What “reads everything” means

The Daily Brief reads every source record that has been successfully ingested, materialized, and included in the configured window. It does not mean:
  • every message in an employee’s historical mailbox;
  • sources that failed connector synchronization;
  • records outside the covered time window;
  • content unavailable because of permissions;
  • more than 12,000 model-input characters from one unusually long source.
Source synchronization health, searchable content, attribution quality, and link availability are therefore part of Daily Brief readiness. The source manifest and coverage counts make these boundaries inspectable.

​
Operational ownership

​
Summary

The Daily Brief is not merely a generated email or meeting summary. It is a governed daily intelligence compilation:
  1. establish the completed business-day source set;
  2. deep-read every included source;
  3. synthesize a cited executive report;
  4. require complete project coverage;
  5. publish one canonical packet;
  6. refresh every active project’s current intelligence;
  7. route uncertain claims through human review before promotion.
That sequence keeps the executive brief, project intelligence, tasks, and follow-up workflow grounded in the same daily evidence.