Frame the retention question
Completed·Archivist
Unlocks 1 task
Research ProjectResearch Project template
Retention model study
Waiting approvalNormalExecution is paused at a gate until somebody decides.
82progress
progress: 82 out of 100Establish whether agent-handled accounts retain differently from human-handled ones, and say so with a confidence anybody can act on.
Next
Paused at Synthesis — is waiting on Nadia Brecht. Approve publication of a null result — agent-handled accounts show no retention difference.
Synthesis
Current stage
4
Stages
11
Tasks
9 closed
1
Open gates
1
Ready now
1 waiting
11 weeks
Planned
Ready to start1
1 waiting- State the recommendation and its confidenceAnselm Reyes
Held behind a dependency
- File the finding to enterprise memory×1
Assigned teams1
Departments, in routing order
Accountable
Executive Nadia BrechtManagers Anselm Reyes
Stage timeline
- STAGE 1
Question
Complete16h plannedState the question and the decision it changes.
2 of 2 tasks100% - STAGE 2
Method
Complete24h plannedAgree the method before running anything.
3 of 3 tasks100%Manager approvalApproved03 Jun 2026Approve the method before any data is touched.
Held by Anselm Reyes
Approved with the null result declared in advance.
- STAGE 3
Execution
Complete120h plannedRun the cohorts and record what happened.
3 of 3 tasks100% - STAGE 4
Synthesis
Active40h plannedTurn the result into a recommendation with its confidence attached.
1 of 3 tasks33%Executive approvalPending05 Aug 2026Approve publication of a null result — agent-handled accounts show no retention difference.
Held by Nadia Brecht
Waiting. The result is null and the CKO wants the confidence interval restated before it lands in enterprise memory.
Dependencies
Question
2State the decision it informs
Completed·Anselm Reyes
Unlocks 1 task
Method
3joins from Question
Design the cohort method
Completed·Delta
Unlocks 1 task
Define success criteria and the null result
Completed·Archivist
Unlocks 1 task
Assemble the account dataset
Completed·Delta
Unlocks 1 task
Execution
3joins from Method
Run the paired cohorts
Completed·Delta
Unlocks 1 task
Record observations at each interval
Completed·Delta
Unlocks 1 task
Check for account-size confounds
Completed·Archivist
Unlocks 1 task
Synthesis
3joins from Execution
Write the findings
Completed·Archivist
Unlocks 1 task
State the recommendation and its confidence
Review·Anselm Reyes
Ready — every dependency closed
Unlocks 1 task
File the finding to enterprise memory
To do·Archivist
Waiting on 1: State the recommendation and its confidence
One column per stage, in routing order. Tasks inside a stage run in parallel; the first task of each stage joins from every task of the one before it, which is the hand-off between departments made explicit.
Board11
Backlog
0Empty
To do
1File the finding to enterprise memory
synthesiswaiting ×1
Archivist08 Aug
In progress
0Empty
Review
1State the recommendation and its confidence
Highsynthesisready
Anselm Reyes05 Aug
Blocked
0Empty
Completed
9Frame the retention question
framing
Archivist22 May
State the decision it informs
framing
Anselm Reyes23 May
Design the cohort method
Highmethod
Delta29 May
Define success criteria and the null result
method
Archivist30 May
Assemble the account dataset
method
Delta02 Jun
Run the paired cohorts
Highexecution
Delta04 Jul
Record observations at each interval
execution
Delta11 Jul
Check for account-size confounds
Highexecution
Archivist18 Jul
Write the findings
Highsynthesis
Archivist29 Jul
Approval gates2
1 openManager approvalApprovedNot reached02 Jun 2026
Retention model studyApprove the method before any data is touched.
Approved with the null result declared in advance.
Executive approvalPendingHolding work05 Aug 2026
Retention model studyApprove publication of a null result — agent-handled accounts show no retention difference.
Waiting. The result is null and the CKO wants the confidence interval restated before it lands in enterprise memory.
Deciding a gate writes the decision and nothing else. The status change, the release of downstream stages and the “ready to close” state are all derived from it on the next render — which is why an approval here is one field rather than a transaction.
Messages2
- Anselm Reyes3d ago
The result is null and that is the finding. It is worth filing precisely because nobody expected it.
Open in the Hub → - Delta1w ago
Account size was the confound. Controlled for it, the gap disappears entirely.
Open in the Hub →
Related projects1
Performance
200h planned · 24h remaining82%
Progress
9 of 11 tasks
3/4
Stages closed
1/2
Gates cleared
0
Blocked tasks
1
Departments
routed through
2
Agents
named on the work
24h
Remaining
from the breakdown
—
Estimate variance
on plan
Task completion82%
Task-weighted, not stage-weighted. A two-hour approval stage and a two-hundred-hour build stage are not half the work each, and treating them as such is the usual way a project dashboard reports 50% before anything has been built.
Completion rule · Deliverable produced
Closed when the findings exist as a memory record — whatever the result turned out to be.
The named deliverable exists as a memory record linked to this workflow.
Automation7
1 would fire · 0 willNext routing decisionHeld
Held at Nadia Brecht's executive gate.
Computed by engine.peekAdvance(), which returns the next stage rather than moving to it. The Task Orchestrator owns the write; keeping this side pure is what lets a card show “next: Engineering” with no risk that rendering a page advances a workflow.
- Advance stage on completion
- WHEN
- Every task in the active stage reaches a terminal column and the stage carries no pending gate.
- THEN
- Mark the stage complete and route work to the next stage in the chain.
- NOW
- 2 tasks still open in Synthesis.
- Notify approver on gate openWould fire
- WHEN
- A stage with a gate becomes active and the gate is still pending.
- THEN
- Raise a workflow notification addressed to the named approver in the Communication Hub.
- NOW
- Nadia Brecht has an open decision.
- Escalate a stale gate
- WHEN
- A gate has been pending longer than its stage's estimate.
- THEN
- Escalate one authority level up and notify the accountable executive.
- NOW
- Needs a real clock. Placeholder timestamps are relative strings.
- Release dependent tasks
- WHEN
- A task reaches a terminal column and other tasks depend on it.
- THEN
- Move every newly-ready dependent from backlog into to-do.
- NOW
- No backlog task is waiting only on a dependency.
- Write outcome to memory
- WHEN
- The completion rule is satisfied.
- THEN
- Write a project-scope memory record summarising the outcome and link it back.
- NOW
- Completion rule is not satisfied yet.
- Auto-approve low-risk gatesHeld for a human
- WHEN
- A QA or manager gate carries no rejection history and sits under the risk threshold.
- THEN
- Nothing. Recorded so the policy is visible and reviewable — a gate an agent can clear is not a gate.
- NOW
- Held for a human by design.
- Stand down a stalled workflowHeld for a human
- WHEN
- No activity for longer than the estimated duration and no open gate explains it.
- THEN
- Nothing automatic. Surfaces the workflow for a human decision.
- NOW
- Held for a human by design.
Nothing here executes. There is no scheduler, no queue and no writer — these are evaluations of the current state, refreshed on every render. Adding a trigger is one row in AUTOMATION_TRIGGERS; it appears on every workflow immediately so a rule can be checked against real data before anything is built to act on it.
Activity log4
- 3d ago
Workflow Engine Opened the executive gate on Synthesis — publication of a null result.
Synthesis - 1w ago
Delta Closed Execution — confound identified and controlled.
Execution - 8w ago
Anselm Reyes Approved the method gate with the null result declared in advance.
Method - 10w ago
CEO Opened from the Research Project template — 4 stages, 2 approval gates.