- Input trigger
- Market signal, executive intent or quarterly planning cadence.
- Actors
- CEO office, strategy lead, finance, product/tech leads.
- Accountable owner
- Chief of Staff / VP Strategy.
- Decision rights
- Exec committee approves portfolio shifts >10% budget; strategy lead ranks within envelope.
- Applicable policies
- Investment policy, risk appetite statement, data classification for market inputs.
- Prerequisites
- Sized opportunities, constraint list, prior-quarter outcome review.
- Evidence sources
- Market research, win/loss, metric owner sign-offs.
- Steps
- Signal → opportunity sizing → constraints → portfolio construction → funding gates → OKR cascade → experiment gates → outcome & reversal review.
- Data assets
- Opportunity model, budget ledger, OKR tree.
- Tools
- Planning sheets; optional: OpenMetadata for metric definitions.
- Metric definitions
- Portfolio value at stake, % initiatives past gates, reversal rate.
- Expected outcomes
- Funded portfolio with named owners and kill criteria.
- Human gates
- Executive authority at funding; review at each experiment gate.
- Failure modes
- Pet projects bypass gates; metrics without owners; no reversal path.
- Rollback
- Stage-gated funding releases; kill criteria pre-agreed.
- Observability
- Decision log with version, owner and evidence links.
- Lessons
- Versioned decisions and reversibility beat annual lock-in.
THE ORGANIZATION MIND
The Workflow Atlas
Twelve generic organizational flows and the mindsets that govern them — sanitized templates, never a disclosure of any specific employer.
Foundational mindsets
Business outcome before model choice
Systems over AI wrappers: start from the value stream and the decision to be improved, then choose the tool.
Map → Redesign → Build → Govern → Measure → Learn → Expand
The transformation sequence; skipping redesign is why pilots never scale.
INDEXED ≠ RETRIEVABLE ≠ RETURNED ≠ VISIBLE
A generic search/system design principle about access boundaries — not a claim about any internal data.
Search output ≠ truth
Query results need semantic contracts and access rights explained before they are trusted.
Intelligence ≠ authority
Capabilities get bounded, revocable permissions; final authority stays human where consequences matter.
Every metric has owner, grain, time window and definition
Without certified definitions, two dashboards will disagree and both will be right.
Uncertainty, counterevidence and abstention are strengths
A system that can say 'I don't know' or 'both sources conflict' is more trustworthy than one that always answers.
Separate evidence from inference and narrative
Source evidence, hypotheses, narrative and retrospective revision are different record types.
Versioned decisions, reversibility, progressive rollout
Experiment before large-scale claims; keep a reversal path.
Robustness, evaluation, observability, economics beat token-heavy demos
Production quality is measured under drift, load and cost caps.
Design around value streams and human agency
Automate repetitive work, never accountability.