Portfolio PDLC — Improve

Outcome

The operating system gets treated as a product: probes run against real workspace data, each fired probe becomes either an improvement card (benefit hypothesis, leading indicators, success/kill criteria, discovery approach) or an explicit "checked, quiet" line — and nothing gets changed on the spot. The improvement backlog is where leverage accumulates; impulse edits to pdlc.md are how operating models rot.

Focus argument

• focus:process — Family A probes (the PDLC itself)
• focus:topology — Family B probes (portfolio construction: dependencies, platforms, team shapes)
• focus:auto (default) — read the board and recent briefs, run the 3–5 probes the data is already pointing at across both families

Workflow

1. Load the probe library: skills/portfolio-pdlc/references/probe-library.md (Family A: flow boundaries, altitude drift, busy board, right-to-left habit, explicit discovery-vs-skip, rubber-stamp commits, measure-and-learn tail, SLE, recurring stale statuses, watermelon retros, risk-balance drift. Family B: dependency concentration, platform extraction, oversized initiatives, factions, outcome-team candidates, T-shaping, option-comparison discipline).
2. Gather the data first: current board.md, flow-log.csv, recent reviews/, learnings/, topology.md, and the cards' dependencies: fields (count name frequency — the 80/20 read). Probes read data; they don't free-associate.
3. Run the selected probes. For each: signal present? cite the data. No signal? one "checked, quiet" line. Recurring learnings in learnings/ count as signal (three similar card-quality learnings = a template/process probe firing).
4. Capture bets, not fixes. Each fired probe worth acting on becomes improvements/<slug>/improvement.md (template): benefit hypothesis, leading indicators, success/kill criteria, and the cheapest discovery — a data probe, a one-review pilot, a single-initiative tracer, or a portfolio-pdlc-simulate scenario (name the scenario in the card). Topology bets MUST include the option-comparison table (B7): each option vs current state on % work localized / change size / risk / future-proofness.
5. Right-size the ceremony. A bet big enough to reorganize teams or change funding deserves full spec treatment — compose your spec-writing skills on its card if you run a larger library; otherwise the improvement template's sections are the lightweight spec. A policy tweak (right-to-left review order) needs no ceremony at all.
6. Report. Append findings to the current review brief (or write reviews/YYYY-MM-DD-improve-probe.md): probes run, fired/quiet, cards created, and which existing improvement bets this evidence strengthens or kills.

Rules

• Capture, never implement. Changes to pdlc.md, topology.md, tiers, or WIP limits happen only when their improvement card passes plan-commit with a human decision — walk the talk; the operating model rides its own lifecycle.
• Every bet gets kill criteria at birth. An improvement we can't revert or falsify is a mandate, not a bet — say which it is and who mandated it.
• Probes cite data or stay silent. "The org should consider…" without a signal is consulting prose; delete it.
• Respect improvement WIP: if improvements/ already holds more open bets than the org digests, strengthen/kill existing ones before minting new cards.
• Simulation results are discovery evidence for a bet — never a substitute for the human decision to adopt it.

Quality Gates

• Every probe run is accounted for: fired-with-citation or checked-quiet.
• Every new card has hypothesis, indicators, success/kill criteria, and a named, timeboxed discovery approach.
• Topology cards include the option-comparison table with current state as an option.
• Zero direct edits to pdlc.md / topology.md / tiers in this invocation.

References

• skills/portfolio-pdlc/references/probe-library.md — the probe families (source: the Portfolio Agility Trailmap minibook + enterprise portfolio practice).
• portfolio-pdlc-simulate — the discovery engine for flow/topology what-ifs.