Recommendations. The Diagnostic Engine
8 deterministic rules that diagnose your workspace. Each one names what's broken and how to fix it.
By the end of this module
What Recommendations Actually Are
Recommendations are not AI suggestions. They're not generic advice. They are 8 deterministic rules that watch the state of your workspace and fire when the rule's trigger condition is true.
Each rule has:
- A trigger (a query against your data, entirely predictable)
- A priority (how urgent the system thinks it is)
- A specific fix (the action that resolves the trigger)
This is important: when a recommendation fires, there is a known way to clear it. The system isn't being vague, it's pointing at a specific structural condition and naming the action that resolves it. Dismissing without fixing leaves the condition in place; it'll fire again the moment the dismissal expires.
The 8 Rules
| Rule | Trigger | Priority | Fix |
|---|---|---|---|
| ------ | --------- | ---------- | ----- |
| OVERLOAD_ALERT | Active projects > capacity limit | High | Park or archive until you're under the limit |
| CAPACITY_WARNING | 2+ DRAINING and 0 COMPOUNDING projects | High | Start a compounding project OR park a draining one |
| OVERDUE_REVIEW | Decision revisitAt date passed, no verdict recorded | High | Record VALIDATED / PROVED_WRONG / SUPERSEDED / STILL_ACTIVE |
| STALE_CAPTURES | 3+ RAW ideas older than 30 days (High at 10+) | Med/High | Triage or bulk park/archive |
| STALLED_TRIAGE | An IN_TRIAGE idea sitting 7+ days | Medium | Finish scoring it or move it back to RAW |
| REVISIT_PARKED | PARKED ideas 60+ days old with score ≥ 60 | Medium | Re-evaluate: archive, restart, or accept park status |
| CONVERT_TO_ASSET | 2+ CLASSIFIED ideas with asset-eligible capture types, 14+ days old | Medium | Convert to vault asset OR archive if no longer valued |
| RECURRING_THEME | A tag used across 2+ object types, 5+ total uses | Low | Decide if the theme deserves its own project (see Relationships module) |
Dismissal vs. Resolution
Every recommendation has two clearance paths: dismiss (snooze for 30 days) or resolve (fix the underlying condition).
Dismiss when: you're already actively working on the fix and don't need the badge in your face. Example: you just parked two projects to resolve OVERLOAD_ALERT, but the index won't recompute until tonight, dismiss until then.
Resolve when: anything else. The vast majority of "I dismissed it" instances are actually procrastination wearing a productivity costume. If you keep dismissing the same alert, the system is telling you something structural that you keep refusing to address.
A useful rule: dismiss at most once per type. If the same alert fires again after dismissal expires, treat that as a forcing function, resolve it now, even if it's inconvenient.
An Empty Recommendations Page
If the page is empty, one of two things is true:
Healthy operation. You're under capacity, your portfolio has balanced energy, you're closing decisions on time, your capture pipeline is current. Congratulations, go execute. Don't manufacture work to fill the screen.
Hidden dysfunction. You may be operating below the system's measurable surface: capturing nothing (no STALE_CAPTURES possible), making no decisions (no OVERDUE_REVIEW possible), running few projects (no OVERLOAD possible). The page is empty because there's nothing for it to read.
To distinguish: glance at /overview. If you see active workflow (recent captures, active projects, recent journal entries), you're healthy. If you see months-old timestamps everywhere, you're hidden, the system has nothing to recommend because you're not using the surfaces it watches.
Common Mistakes
Treating recommendations as a to-do list. They aren't tasks. They're diagnostic outputs. Once resolved, they disappear automatically, you don't 'complete' them, you fix the upstream condition.
Expecting alerts for everything important. The 8 rules catch the most common structural failures. Many important things (the right strategic call, the missing customer conversation) won't fire because they aren't deterministically detectable. That's what weekly review and the AI Founder Briefing are for.
Dismissing RECURRING_THEME alerts. They're the rarest and often the most strategically loaded. When one fires, take it seriously, the system noticed a pattern in your own work that you may not have named.
Practice Checkpoints
Sign in to track your progress and mark checkpoints complete.
Next in Foundations Path
Weekly Review. The Hypothesis Loop
15 minutes that close last week's loop and place this week's bets.