HK IPO Parameter Manager

Use this skill for stateful scoring-rule management.

Runtime

This publish bundle includes the required CLI runtime under runtime/hkipo-next.

From the skill folder:

cd <skill_dir>
uv run --directory runtime/hkipo-next hkipo-next ...

By default parameter versions are stored in ~/.hkipo-next/data/hkipo.db.

Parameter Model

• Weights: snapshot-quality-weight, affordability-weight, pricing-stability-weight, sponsor-support-weight, cost-efficiency-weight
• Thresholds: participate-min, cautious-min
• Costs: handling-fee-hkd, financing-rate-annual-pct, cash-opportunity-rate-annual-pct, lockup-days

Workflow

1. Inspect current versions with params list or params show.
2. Save new candidates instead of mutating old versions.
3. Activate a version only when the user wants it to become the new default.
4. Compare candidate versions on a representative symbol before broad rollout.

Commands

List versions:

cd <skill_dir>
uv run --directory runtime/hkipo-next hkipo-next params list --format json

Show the active version:

cd <skill_dir>
uv run --directory runtime/hkipo-next hkipo-next params show --format json

Save and activate a tuned version:

cd <skill_dir>
uv run --directory runtime/hkipo-next hkipo-next params save \
  --name conservative-margin \
  --sponsor-support-weight 0.20 \
  --cost-efficiency-weight 0.25 \
  --participate-min 78 \
  --cautious-min 62 \
  --lockup-days 6 \
  --notes "Bias toward sponsor quality and lower carrying cost" \
  --activate \
  --format json

Compare two versions:

cd <skill_dir>
uv run --directory runtime/hkipo-next hkipo-next params compare 2476 v0002 v0003 --format json

Output Cues

• active_version is the default version used by later scoring commands.
• action_changed in compare output tells you whether a tuning change would alter the recommendation.
• factor_deltas shows which factor scores moved between versions.

Companion Skills

• Use $hkipo-decision-engine after an active version is ready.
• Use $hkipo-review-optimizer when tuning changes come from real-world review feedback.