HarbourBridge is a fictional medium-sized organisation with twelve initiatives competing for limited review, engineering and funding capacity. The scenario asks a practical question: when AI helps people prepare more proposals, how can a portfolio keep assessing ideas without starting more work than people can responsibly support?
This is a demonstration of decision support. It is not evidence of an actual organisation’s investment returns, a completed enterprise rollout or successful delivery of the twelve initiatives.
The decision problem
A queue can look healthy while the people needed to check the work are already fully committed. A proposal may also depend on another initiative, lack important source information or carry an ongoing cost that a pilot budget does not cover.
The example keeps these questions separate. Capturing a proposal does not authorise assessment. Assessment does not automatically justify a start. A promising option is not a funding decision, and a funding decision does not necessarily release work.
How AIPF shaped the demonstration
The work framed a bounded portfolio decision, made responsibilities and evidence records explicit, and tested a connected path through comparison, human choice and recovery. The portfolio asks for evidence from Product work, or from Product Family work when shared decisions are needed. The family route is optional.
The later presentation follows a meeting: review existing work, examine assessment findings and data readiness, triage new proposals, compare choices, then record the agreed set at the end. A service-report drafting proposal provides a concrete thread through the discussion.
The account distinguishes recorded technical work from retrospective interpretation using AIPF. It does not claim that every original implementation action passed a formally documented AIPF gate.
A small example of a consequential constraint
In the retained September technical exercise, a synthetic reduction in review capacity from 24 to 16 days reduced the feasible combinations from 51 to 15. Those are results for that fixed historical fixture, not current operating figures or a universal rule.
The lesson is not to select the option with the largest score. Protected obligations, dependencies and evidence quality constrain the choice before someone considers its value. Missing information stays visible; it is not silently treated as zero cost, zero risk or zero value.
What was demonstrated
| Evidence | What it supports |
|---|---|
| Deterministic option calculations and checks | Feasible combinations can be compared under the modelled constraints. |
| Missing, stale and changed-source cases | A recommendation can be withheld when its basis is not suitable. |
| A bounded test handoff with interruption and replay | One historical test exercised reconciliation without duplicating the handoff. |
| Meeting and evidence-return rehearsal | People can inspect how a proposed decision should carry owners, limits and a return question. |
The historical engineering harness recorded 32 passing checks. The two calculation approaches were authored by the same AI, so their agreement is a cross-check, not independent assurance. Live organisational permission enforcement and production resilience were not established.
What changed when the checks failed
A recovery routine initially confused a delimiter in documented code with the actual receipt boundary. An in-memory check exposed the resulting content loss before the live test used that version. The boundary was corrected and preservation checks were added.
A separate fixture failed because its required test-only marker was missing. The fixture was corrected; the safeguard was retained. These examples matter because recovery and safety rules themselves need testing.
What readers can reuse
Begin with one real decision and its limiting resource. Keep source quality, recommendation, decision and release distinct. Show the consequence of removing a prerequisite or changing capacity. Require a human rationale, an accountable owner and a next evidence question.
Use the Portfolio Guide to frame the decision and the Portfolio Implementation Guide to connect its records. A new organisation would need its own sources, authority checks and observed user task before relying on this approach.
Evidence boundary
The public article summarises the documented demonstration and its limits. It does not connect to private workbenches or collaboration systems. No real investment approval, business benefit, independent user acceptance or production readiness is claimed. Figures belong to the explicitly dated synthetic exercise.
Help improve this page
Found a problem or a better way to explain it? Suggest a correction or improvement.