D-05 — Starter Packs for Developers

Scope: Building and activating starter packs — pack anatomy, per-pack composition, cohort billing pinning, Soft-to-LIVE rollout, and an authoring checklist (source: docs/platform-map/jijibora/D-02-PLAYBOOK-DESIGNS.md; source: docs/commercialization/playbooks/dictionary.v1.json; source: datasets/lookups/lookup.finmon.cohort_billing.v1.json) Status: as of 2026-11-09 Blueprint sources: docs/commercialization/playbooks/dictionary.v1.json; datasets/lookups/lookup.finmon.cohort_billing.v1.json; datasets/lookups/lookup.finmon.revenue_split_profile.v1.json; docs/platform-map/jijibora/D-02-PLAYBOOK-DESIGNS.md; Playbook Registry — What You Can Run, Who Runs It, What You Get Back

What this means for you

A pack is data, not code. You declare lookup keys, a scenario pack, CNS types, and consumer wiring in the dictionary, and the runtime resolves them as values (source: docs/commercialization/playbooks/dictionary.v1.json). Every pack proves itself in Soft first: Soft apply is validation plus plan with zero writes. Only a real LIVE gate passing real checks may ever flip a pack to live (source: docs/platform-map/jijibora/D-02-PLAYBOOK-DESIGNS.md). If your change needs a new table, stop and ask. A new community is a new CNS entry, catalog entry, manifest, fragment, and vocabulary — never a schema change (source: docs/jijibora-gtm/02_PRODUCT_STRATIFICATION.md).

Details

1. Pack anatomy — what a pack binds

A dictionary pack entry binds five things (source: docs/commercialization/playbooks/dictionary.v1.json):

Rules that apply to every pack: reuse the canonical criteria bus and envelope, do not invent a second bus; declare canonical_subject, canonical_event_type, and pipeline_id on pack lookups; Soft means outbox plus allowlisted subject while LIVE means JetStream plus ceiling (source: S-03 — Jijibora Playbook Registry (stakeholder + developer)).

2. Per-pack composition table

Three packs in one view. Everything here is taken from the dictionary pack entries (source: docs/commercialization/playbooks/dictionary.v1.json).

Note the honest sibling status: the claims SKU is one umbrella, and motor/medical/home/gov packs are separate pack entries sharing that SKU. The dictionary notes medical/home/gov packs stay spec_only until later dependencies land, while their pack entries carry soft_shipped maturity — flag the discrepancy, do not merge it silently (source: docs/commercialization/playbooks/dictionary.v1.json).

3. Cohort billing_ref model for targeting

• Allowlist, values only. The cohort billing lookup names exactly four pinnable offer-package ids: offer_package.jamii.starter.v1, offer_package.materra.amcos.pilot.v1, offer_package.bundle.estate_safe.v1, offer_package.claims.orchestration.pilot.v1 (source: datasets/lookups/lookup.finmon.cohort_billing.v1.json).
• Attach mechanics. A cohort definition pins a billing_ref onto one of those ids. Nothing else is attached — no new billing table per cohort (source: datasets/lookups/lookup.finmon.cohort_billing.v1.json).
• Cohort as targeting plane. The cohort decides who is targeted by an offer; it never decides amounts (source: datasets/lookups/lookup.finmon.cohort_billing.v1.json).
• Billing through existing subscriptions. Billing still runs through commercial_subscription facts plus the billing adapter. A billing_ref never invents an invoice (source: datasets/lookups/lookup.finmon.cohort_billing.v1.json).
• Split shape (skim only). Revenue split profiles are KV-only declarative profiles: parties use the FINMON-03 vocabulary (platform | pack_creator | parent_scope | community_treasury | reseller | cohort_operator | merchant), basis points must sum to 10,000, and the pack-creator share must match the SKU's creator fee basis points. The profile settles nothing by itself — the FINMON-S7 voyage distributes and the SIKASIE/Malipo plane credits (source: datasets/lookups/lookup.finmon.revenue_split_profile.v1.json). Per-party values in the lookup file are configuration, not a price list, and are not reproduced here.

4. Pack COLD vs LIVE rollout

• COLD = Soft posture (validation, zero writes). Soft apply validates and plans: it checks the digest pin, resolves references, and must write zero live rows. Soft-never-LIVE is a dictionary principle (soft_billing_ne_live_paid, fail_closed_soft_ne_live) (source: docs/commercialization/playbooks/dictionary.v1.json).
• Soft apply validation plan: confirm the soft pin writes the soft. key with invent_refuse_live: true and live_gate_met: false, confirm the outbox row is written for the pack emit, and confirm the pack's Soft acceptance tests pass (for the starter: Soft ticket end-to-end, outbox row for the work-order emit, and the billing-adapter Soft binding check; for claims: the Soft fixture suite with no verdict invented) (source: docs/commercialization/playbooks/dictionary.v1.json).
• Writes: 0. A Soft apply that writes anything live is a failure. If an external service (solver, Camel, Kestra) is down in Soft, refuse honestly instead of inventing a result — Soft gates treat an honest 503 as a pass (source: docs/platform-map/jijibora/D-02-PLAYBOOK-DESIGNS.md).
• LIVE = JetStream + ceiling + real probes. Flip to LIVE only when the live path's real checks pass: the ceiling is enforced, the digest pin matches, and each LIVE probe reports genuinely green. Env flags alone never satisfy live_gate_met (source: Honesty Invariants). A pack never declares itself live (source: Playbook Registry — What You Can Run, Who Runs It, What You Get Back).

5. Authoring checklist — start from the maintenance desk template

Copy the maintenance desk as your template, then declare the new pack in the dictionary. Check each item before asking for activation:

• [ ] Intents declared. Intake taxonomy (for maintenance: plumbing, electrical, security, facilities) plus location, urgency, and photo extraction are named as intent inputs in the entry (source: docs/jijibora-gtm/03_ENTRY_POINT_AND_STARTER_PACK.md).
• [ ] Gate policy set. Approval gate (manager one-tap / Decision Gate), evidence gate (photo + GPS, inspector signature where required), and payout gate (escrow released only on proof) are refused-by-default gates, not comments (source: docs/jijibora-gtm/03_ENTRY_POINT_AND_STARTER_PACK.md; source: Playbook Registry — What You Can Run, Who Runs It, What You Get Back).
• [ ] SLA / bands / VAS composition added. Claims packs add service-level clocks, amount/risk approval bands, and value-added-service eligibility. All three are lookup-composed (threshold, SLA, and VAS-eligibility lookups) and change without code (source: docs/commercialization/playbooks/dictionary.v1.json; source: Playbook Registry — What You Can Run, Who Runs It, What You Get Back).
• [ ] AI hook allowlists start off-by-default. The assistant proposes and narrates only; hooks are advisory/tracking-only, the core path works with hooks empty, and nothing auto-approves or auto-pays (source: docs/platform-map/jijibora/D-02-PLAYBOOK-DESIGNS.md).
• [ ] Pack is fully bound. Lookup keys, scenario pack, CNS types, consumer wiring, and the SKU monetization ref (offer package + split profile + creator fee matching) are present and consistent (source: docs/commercialization/playbooks/dictionary.v1.json).
• [ ] SoD + anti-self-approval wired. Creator/resolver cannot approve their own item; unattested providers are refused; missing evidence refuses (source: Playbook Registry — What You Can Run, Who Runs It, What You Get Back).
• [ ] Soft-first proof attached. Soft validation plan with writes: 0 has been run, rollback re-pin rehearsed, and the LIVE gate still reports dark (source: docs/platform-map/jijibora/D-02-PLAYBOOK-DESIGNS.md).
• [ ] No schema change smuggled in. New behaviour is CNS entries + interface catalog + manifest + fragments + APF vocabulary + KV lookups only. If a new table is needed, stop and ask (source: docs/jijibora-gtm/02_PRODUCT_STRATIFICATION.md).

Open items / gaps

• canonical_subject / canonical_event_type / pipeline_id declarations on pack lookups (ESP-02), ontology-normalizer generalization (ESP-03), and Soft ceiling assert (ESP-04) are TODO — design with them in mind, do not assume them live (source: docs/platform-map/jijibora/D-02-PLAYBOOK-DESIGNS.md).
• Real money movement gates (timestamp authority, public-safety answering point, billing-adapter production migration) remain OPEN-OPS and need an external party; live money stays Soft (source: docs/platform-map/jijibora/D-02-PLAYBOOK-DESIGNS.md).
• SoR table/shape for several packs beyond resource / claim / truth_node is UNKNOWN in the read set; locate the canonical mapping before building (source: docs/platform-map/jijibora/D-02-PLAYBOOK-DESIGNS.md).
• Claims pilot pricing is unpublished and no savings figure may be quoted; do not put price-guide numbers into pack docs as promised prices (source: Playbook Registry — What You Can Run, Who Runs It, What You Get Back).
• Cohort billing is an allowlist of four ids today; adding a fifth package is a lookup change, not a schema change — but it must still pass the digest pin before it takes effect (source: datasets/lookups/lookup.finmon.cohort_billing.v1.json; source: docs/jijibora-gtm/09_GAP_ANALYSIS_AND_RECOMMENDATIONS.md).