SEIZE portfolio coverage
The 21 templates cover obligations in a complete mission institution. They are not a requirement to buy 21 separate paid tasks. A small mission may combine compatible obligations, while the producer, final verifier and accountable admission roles remain distinct. A template or coverage label does not prove that its work occurred.
The implementation locations are packages/successor-core/src/templates.mjs and src/underwriting.mjs in that package; compilation and the actual run must supply the evidence. Underwriting's ordered decisions are unsigned commitments, not authenticated permission. See acceptance status and the architecture decision.
| # | Obligation | Evidence expected before claiming coverage |
|---|---|---|
| 1 | Mission Constitution | Frozen purpose, principal, constraints and prohibited actions |
| 2 | Incumbent Baseline | Reproducible quality, cost, risk, time, human burden and authority |
| 3 | Alternative Set | Strong credible frontier/software/specialist/human/hybrid/no-action alternatives |
| 4 | Rights and Data Constitution | Provenance, licensing, retention, exclusions and permitted reuse |
| 5 | Mission Gym and Benchmark Constitution | Versioned state/action/transition/reward model; representative partitions; separately addressable environment and protected evaluation; basis-risk plan |
| 6 | Oracle and Teacher Study | Declared teaching intelligence, no hidden calls and transfer hypothesis |
| 7 | Failure Taxonomy | Severity, causes, abstention, adversarial cases and critical failures |
| 8 | Architecture Frontier | Materially different complete systems and preserved reserve options |
| 9 | Data and Experience Formation | Rights-cleared examples, simulations and quality controls |
| 10 | Specialist Model Formation | Exact adaptation/training or algorithm-construction artifacts |
| 11 | Tool and Retrieval Formation | Deterministic checks, retrieval, plans, tools and bounded execution |
| 12 | Verifier Formation | Calibrated evaluator, replay and independently controlled final scoring |
| 13 | Security and Privacy Proof | Threat model, isolation, credentials, data boundaries and incident evidence |
| 14 | Complete Economics | Formation, inference, evidence, review/rescue, maintenance, reproof and unwind |
| 15 | Human Burden and Agency | Escalation, contestability, supervision and retained decision rights |
| 16 | Resilience and Rollback | Tested degradation, portability, stop controls and known-good restoration |
| 17 | Protected Fresh Evaluation | Frozen system examined on fresh permitted cases under the fixed protocol |
| 18 | Independent Validation | Separated review, conflicts, limitations, full cost and critical-error findings |
| 19 | Accountable Acceptance | Purpose-specific human or institutional decision about the work |
| 20 | Chronicle and Authority Admission | Separate knowledge and authority decisions, each scoped and expiring |
| 21 | Capital-to-Capacity and Requalification | Allocation of verified available value; renewal/impairment triggers and next proof |
For the invoice rehearsal, the ten operational jobs are extraction, purchase-order matching, duplicate checking, vendor/banking evidence, completion evidence, rates/warranty, recommendation, adversarial challenge, separate verification and human decision-pack preparation. They consume the portfolio's constitution, rights, baseline, evaluation and control records. Ten successful stages do not establish real independent proof or an admitted successor.
A coverage report should state the obligation, responsible job/role, required evidence, present evidence and missing condition. Public synthetic protected-proof rehearsal must remain labelled synthetic: cases shipped in the repository are not confidential independent holdouts.
SEIZE's final output is a Promotion Request. A subsequent proof report, work acceptance, payment, Chronicle admission and Authority Envelope remain separate objects. Paying for a rigorous negative evaluation cannot change its negative verdict.