"We need to get back on track" is not a corrective action

When delivery drifts, teams default to heroics—extra hours, vague scope promises, silent descoping. Leadership slides say "corrective plan in place" without naming what changes: date, scope, headcount, or parallel work frozen.

AI corrective action suggestions should mean structured option sets, not motivational prose. WKFGo's simulate_scenario compares futures like delay_release, cut_scope, add_people, and freeze_project across dimensions returned by the tool—time, scope, cost, risk—with assumptions explicit. Humans resolve_decision_item and log_decision; AI does not execute recovery alone.

Problem: corrective theater

Failed recoveries share patterns:

Actions without owners or dates
Scope cuts that never update tasks
Adding people without fixing review bottleneck
No forecast to calibrate in next month's brief

Corrective plans in email threads die when the next fire starts.

Framework: diagnose → simulate → choose → implement → calibrate

Diagnose — executive_brief, get_project_report, flow_aging, finance variance; label data gaps
Simulate — Run simulate_scenario for each viable fork; compare side-by-side
Choose — Executive meeting or decision_inbox resolution with named owner
Implement — Update tasks, wiki, assignments—MCP tools like update_status, cut scope tasks, or staffing changes on board
Calibrate — log_decision forecast; review via decision_log

Suggestions are inputs; implementation discipline determines outcomes.

Scenario modes (WKFGo MCP)

Mode	When to consider
`delay_release`	Contract or quality allows date move
`cut_scope`	Must ship date; defer low-value work
`add_people`	Bottleneck is capacity, not dependencies
`freeze_project`	Portfolio overload; stop starting

AI should explain trade-offs in plain language tied to returned metrics—not invent ROI percentages.

Solution: corrective workflow

Trigger from deviation alert

When decision_inbox fires on slip risk:

Pre-read brief for affected project
Run two scenarios minimum (e.g., cut_scope vs delay_release)
Attach comparison to steering wiki with update_wiki_page
Live meeting debates only deltas; resolve in inbox

Boardroom for contested recovery

If CFO and CPO disagree on cut vs delay, use executive_brief CEO lens board.tensions—stage debate, then simulate_scenario both paths. CEO records decision with confidence and unknowns.

Implementation checklist

After choice:

Tasks descoped or dates moved on board
Stakeholder comms drafted (human-owned)
Dependencies re-validated with add_dependency updates
Forecast logged for calibration

Skip checklist steps and corrective actions become slide fiction.

Corrective action templates by failure mode

Failure mode	First simulate	Implementation focus
Review bottleneck	add_people vs cut_scope	WIP limits, reviewer roster
Dependency chain break	delay_release vs descope downstream	`add_dependency` refresh
Budget overrun	cut_scope vs freeze_project	Finance entries, stakeholder comms
Quality escape	delay_release vs scope to testing	Approval tasks, test wiki update

AI should map observed signals to this table in pre-reads—not invent new scenario types outside MCP modes. Facilitators attach the chosen row to log_decision so calibration compares like-with-like next quarter.

Recovery communications stay human-owned: AI drafts timeline bullets from task data; PM or PMM sends after legal review when customer-facing.

Weekly recovery standup (15 minutes)

When a corrective path is chosen, run a short recovery standup separate from team status theater:

Review only tasks tied to the logged scenario (descopes, date moves, new hires)
Confirm finance rows reflect scope changes—CFO brief should not contradict the board
Check whether decision_inbox can close the triggering item or needs a follow-up fork
Update wiki recovery page status so smart_search finds current plan

Recovery fails when executives choose cut_scope in a meeting but the board still shows full scope Friday afternoon.

Anti-patterns

Simulating without changing tasks
Choosing add_people when review queue is the constraint
Multiple simultaneous recoveries across portfolio—WIP limit executive fixes
Letting AI mark_done on tasks to "show progress"

Corrective actions without simulation debate repeat last quarter's mistakes. Before committing to descope or delay, run simulate_scenario for the fork leadership is actually considering. Log the chosen path with resolve_decision_item so decision_log calibrates the next cycle.

Run a monthly audit: do AI answers cite task IDs from MCP? If not, tighten prompts to require citations before any stakeholder-facing draft leaves the team.

FAQ

Does AI implement corrective actions?
It suggests and narrates scenarios; humans update tasks and assignments.

How many scenarios should we run?
At least two for material slips; more if boardroom debate requires.

Can we automate scenario runs on alert?
Yes as pre-read drafts; human confirms before external comms.

Relation to decision_inbox?
Inbox raises the fork; simulation informs resolve_decision_item.

Pick a recovery path you can defend

Compare scenarios, choose explicitly, implement on the board, log the forecast.

Try it now

Put these patterns on live project data—not slide decks.