# missionMATCH Finish Plan (presentation readiness)

Goal: take the platform from feature complete backend to a coherent, demonstrable, client ready product,
then audit and debug the whole application, then produce a full presentation PDF. No Python anywhere, no
dashes in written output, the frozen v1.4.0 engine stays the only per candidate scorer, EEO walled off, every
step genuinely built, tested, and verified. Anything blocked on external credentials is called out honestly as
integration ready, never faked.

Live state confirmed (prod demo org 11111111...1111 on M4 missionmatch, goose v42): 1511 candidates, 134
available TS/SCI active, 15 requirements, 6 open opportunities, 3349 global opportunity records, 10 slates, but
only 1 persisted candidate evaluation and 0 staffing vendors, and the AI brain has no frontend surface.

## Track A: demo coherence (the story ties out end to end)
- [ ] A1 Fix the demo seed files so the gap to filler story is reproducible on a clean DB: align the rare
      skills (SIGINT, Satellite C2, Cross Domain Solutions) across cleared_skills.sql, demo_base.sql, and
      demo_gapfillers.sql, assign each rare skill to at least three candidates, and point the three gap
      requirements at those rare skills so the gap the Command Center names is the gap the fillers close.
- [ ] A2 Populate candidate evaluations for the LIVE demo org through the real engine (slate generation or a
      batch screen), so WhoCanSupport and the Candidates Win Impact panel show ranked qualified people with the
      audit trail present. This is the single biggest live gap (only 1 evaluation exists).
- [ ] A3 Seed Phase 2 vendor demo data (staffing_vendors plus vendor_submissions linked to the seeded
      evaluations) so the vendor scorecard and quality distribution have a live scenario.
- [ ] A4 Verify the screenMATCH opportunity selector and the Command Center opportunity panels populate from
      the live catalog; add representative demo opportunity records if a segment is empty.
- [ ] A5 Confirm a coherent pre generated slate with decisions so slateMATCH opens populated and the four
      fifths readiness readout renders.
- [ ] A6 Produce a fairness snapshot and a curation proposal so the compliance page renders real content.

## Track B: surface the differentiators in the UI
- [ ] B1 AI brain Ask panel: a nav item and route that posts a natural language question to POST /ai/query,
      renders the grounded answer with its citations, and offers example prompts mapped to the seeded data.
- [ ] B2 Command Center panels for Phase 1 Capture Readiness and Phase 2 Vendor Intelligence so the two newest
      backend phases are visible in the demo.

## Track C: demo operations
- [ ] C1 One command demo bootstrap (make demo) and docs/DEMO_WALKTHROUGH.md with the exact click path and the
      expected on screen numbers.
- [ ] C2 Provision the Galapagos demo org owner with TOTP, confirm the deployed binary is the auth plus brain
      build, so a login backed demo works from a clean browser.

## Track D: full audit and debug
- [ ] D1 Comprehensive multi dimension adversarial audit across the whole application (AEDT compliance,
      security and tenancy and RLS, correctness, the demo story, the frontend, dead code and TODOs). Fix every
      confirmed issue.
- [ ] D2 Verify the full unit and integration suites are green and the live demo is unbroken after all changes.

## Track E: presentation PDF (on the Desktop)
- [ ] E1 How it works: a step by step data flow narrative.
- [ ] E2 Sales presentation: the client facing pitch.
- [ ] E3 Full application architectural write up.
- [ ] E4 A data flow chart (visual diagram).
- [ ] E5 Confidentiality and intellectual property disclaimers so the material cannot be repurposed.
- [ ] One comprehensive branded PDF via hand authored HTML rendered by headless Chrome (no Python).

## Blocked on external credentials (present as integration ready, not built here)
- iCIMS live ATS connector (adapter and connector_sync_state exist; needs a pilot OAuth grant).
- Workday and SharePoint resume ingestion (needs live tenant API grants).
- GovWin IQ, GovTribe, eBuy opportunity sources (redistribution licenses, per tenant vault).
- AWS GovCloud or Azure Government production deploy (Terraform exists; needs cloud creds and a container
  runtime to build the distroless FIPS images).
- Live FedRAMP embedding and parsing endpoint (needs a provisioned GPU or a managed Bedrock endpoint).