# Sequential feature roadmap

Prepared for a November 9, 2026 role start. All examples are synthetic and the product is independent of the employer. Build one small feature increment at a time and collect Kyle's preferred approach before expanding it.

| Priority | Feature | Approach A | Approach B | Approach C | Implemented now | Next useful increment |
|---|---|---|---|---|---|---|
| 1 | Data quality | Column completeness | Identifier integrity | Field/domain rules | All three, arbitrary-column mapping, preview, required fields, amount bounds, allowed categories, source references | Reusable named rule sets and comparison against prior review dispositions |
| 2 | Reconciliation | Individual key occurrences | Sum by key or unique-only | All-record totals | Two mapped extracts, full outer join, explicit duplicate policies, unmatched/duplicate inventories, source references; paired mode retained | Optional composite identifiers and user-reviewed transaction match rules |
| 3 | Loss development | Volume-weighted links | Mean origin links | Latest-origin links | All three, selected tail, projected ultimate/unpaid | Exclusion notes and selected factors by age; residual diagnostics |
| 4 | Reserve scenarios | Paid development | Paid BF | Expected loss | All three, unpaid and case difference | Side-by-side assumption sensitivity grid and linked triangle CDFs |
| 5 | P&C ratio review | Underwriting ratios | Surplus/growth screening | Portfolio mix | All three with denominators disclosed | Period comparison bridge and user-defined peer references |
| 6 | Control evidence | Gap inventory | Risk-weighted queue | Evidence age | All three, as-of date and staleness setting | Evidence-file hashes and reviewer sign-off stored in workbook |
| 7 | Deadline planning | Countdown | Owner workload | Backward milestones | All three from user-entered dates | Authoritative jurisdiction/source register with reviewed effective dates |
| 8 | Data lineage | Metadata inventory | Downstream impact | Cycle detection | All three on declared dependency graph | Optional workbook formula lineage with explicit scope preview |
| 9 | Anomaly triage | IQR fences | MAD robust z | Group concentration | All three, configurable sensitivity | Peer-group baselines and reason-coded review feedback |
| 10 | Review package | Draft narrative | Action register | Evidence matrix | All three, text/CSV/JSON/new-sheet export | Persistent version comparison and reviewer disposition |
| 11 | Reinsurance review | Quota-share allocation | Per-claim excess layer | Aggregate stop-loss | Planned only | Add explicit limits, reinstatement assumptions and gross/net reconciliation |
| 12 | Weather/P&C join | Location exposure overlay | Event-to-claim timeline | Portfolio event roll-up | Planned only; weather API catalog is a separate app deliverable | Join stable locations/event IDs; distinguish hazard proxy from insured loss |
| 13 | Regulatory source tracker | Manual primary-source register | Saved-link change review | Jurisdiction checklist | Planned only | Source checks with retrieval/effective dates; no inferred legal deadlines |

## Current delivery state

- Implemented: ten modules, thirty distinct computational approaches, synthetic cases, paste/file input, three-approach comparison, exports, Office selected-range read and new-sheet write code.
- Tested: pure parser/calculation/export suite; record actual browser and hosted checks in the delivery notes.
- Requires environment validation: actual Excel-host sideloading and workbook read/write, enterprise add-in policy. Public HTTPS/DNS deployment is live; hosted smoke evidence is in LIVE_SMOKE.md.
- Not implemented: official RBC formulas, statutory deadline database, automated workbook-wide discovery, confidential client data integrations, persistent/cloud storage, actuarial opinion/sign-off, Marketplace distribution.

The three input modes (synthetic, CSV, Excel) are additional ways to use **each** approach; they are not being counted as the three analytical alternatives.