# Verixa Pilot — Execution Plan v3 (MASTER, all-inclusive)

**Prepared for:** Vimal — Founder/Chairman, Verixa · **Date:** 2026-05-31 · **Realigned:** 2026-06-18 (patent FD shift)
**Skills deliberated:** Head of QA · AI Validation & Governance · CSV/CSA Assurance · Product/Engineering Architect.
**Supersedes:** the language/timing/optimism in all prior pilot docs. This is the single source of truth; the timeline (`Verixa_Pilot_Timeline_2026.xlsx`) and flow (`Verixa_Pilot_Flow.html`) mirror it.
**Status:** Advisory. Final go/no-go + risk acceptance are human QA + Founder decisions. Verixa never claims "validated."

> 🔄 **18 Jun 2026 realignment note.** Patent FD target shifted **15 Jun → 26 Jun**. Engineering W1-W14 holds unchanged (demo lock Fri 26 Jun · Verified Go-Live Decision Tue 01 Sep). Marketing public-disclosure window shifts: founder 1-to-1 outbound to known DP targets continues pre-FD under NDA; public marketing (newsletter · pillar pages · founder LinkedIn cadence · PR note · webinar invites) unlocks **Mon 29 Jun (W5) post-FD**. All FD+N statutory patent dates re-anchor: complete specifications due **26 Jun 2027** · PCT decision **26 Dec 2026** · US/EU national-phase **26 Dec 2028** · Request for Examination **26 Jan 2029**.

---

## 1. Locked decisions

| Decision | Status |
|---|---|
| Supplier model | **Verixa verifies; design partner validates intended use.** No internal Verixa IQ/OQ/PQ. |
| Scope | **All 6 workflows + MIRA committed** to the production-verified version. |
| **Demo-ready date** | **🔒 HARD-LOCKED: Fri 26 Jun 2026.** Fixed commercial commitment (no validation dependency). G-DEMO segregation must pass by this date. |
| **Production go-live** | **TARGET 01 Sep 2026**, with a **Sep 1–15 buffer (contingency only, NOT a baseline)**. Hard backstop **15 Sep**. |
| Capacity | **Two hires are REQUIRED, not optional** (§3). |
| Language | "Verified under our SDLC; customer validates intended use" — never "validated." |
| Evidence | Everything PLANNED; implementation/verification status = `Unknown — evidence required`. |

---

## 2. The model (3 phases)

1. **Demo-ready [🔒 26 Jun]** — segregated eval env, all-six + MIRA, synthetic data; drives onboarding + design-partner LOI.
2. **Verified version [TARGET 01 Sep, buffer→15 Sep]** — controlled-SDLC build, Verixa verification only, supplier assurance pack (G-7) + signed quality agreement (G-8).
3. **Partner validation [Sep+]** — partner runs IQ/OQ/PQ + intended-use validation, leveraging G-7.

---

## 3. Required capacity (non-negotiable)

| Add | Role | Scope | Confirm by | Note |
|---|---|---|---|---|
| **+1 Full-stack build hand** | Engineer | **WP-12 inspection export + scorecard + test authoring ONLY — NOT the integrity core** | **W5** (to be productive by W7) | Existing senior devs own WP-4/WP-13b (highest risk). Budget ~60% productivity for first 2 weeks. |
| **+1 Fractional CSV/CSA + RA contractor** | Validation/RA | Verification evidence, G-7 pack, G-8 agreement, enablement kit, Part-11 framing | **W2** | 2 QA cannot own all assurance lanes alone. |

---

## 4. Critical path (7 — miss any → narrow scope per §8)

1. WP-4 e-sig substrate verification · 2. WP-13b fail-closed AI audit · 3. MIRA provider assurance pack (V7) · 4. O3 supplier qualification (Anthropic/Azure) · 5. Security + tenant isolation (P6) + DR · 6. G-7 supplier assurance pack · 7. G-8 signed quality agreement.

---

## 5. Engineering refinements (folded in — Architect lens)

1. **Decide the PDF library in Week 1** — shared by WP-12 (inspection pack) and WP-13 (AI governance pack). It was a flagged TBD; un-blocks both before W7.
2. **WP-10 Findings→CAPA gate pulled to W7** — WP-12's export depends on the observation→finding→CAPA chain it emits. (Alternative if WP-10 slips: build WP-12 export-core — hashing/manifest/PDF — first, wire linkage events when WP-10 lands.)
3. **Build hand ring-fenced** to WP-12/scorecard/tests; the integrity core (WP-4, WP-13b, WP-2) stays with the senior devs.
4. **Stand up the CI verification-gate (`CLAUDE.md` gates) in Week 1** so verification evidence is **auto-captured**, making the W7→W14 evidence burndown cheap rather than a manual August tax.

Agentic-coding split: agents drive bounded new work (WP-12, scorecard, tests); humans-with-agent-assist on WP-4/WP-13b/WP-2 where audit/e-sig/RLS correctness is unforgiving.

---

## 6. Schedule (key dates)

| When | Item |
|---|---|
| **W1 (Jun 01)** | Approve V1/V2 · start MIRA V7 · start O3 · draft G-7/G-8 · **PDF-lib decision** · **CI gate up** · secure design-partner LOI · confirm both hires |
| W2 (Jun 08) | WP-4 + WP-13b build · contractor onboarded |
| W3 (Jun 15) | WP-2 subset · WP-1 single-tenant lock |
| W4 (Jun 22–26) | G-DEMO gate · **🔒 DEMO-READY 26 Jun (LOCKED)** |
| W5 (Jun 30) | **G-JUN** Go/No-Go · **build hand confirmed** |
| **W6 (Jul 06)** | **KILL-CHECK** (≥2 red → narrow/re-baseline) |
| **W7 (Jul 13)** | **WP-12 + WP-13 front-loaded start** · WP-10 · WP-8 · WP-9 · evidence burndown begins (weekly) |
| W8 (Jul 20) | WP-12/WP-13 complete; **MIRA guardrail verification starts (as each guardrail lands, W8/W9→W12)**; **G-7 populated continuously from W7** |
| W9 (Jul 31) | **RC freeze + G-8 SIGNED (hard gate)** |
| W10–W13 (Aug) | Verification **closure** (not first authoring) · G-7 **closure** · MIRA provider pack · O3 · security/P6/DR · enablement kit · defect closure |
| **W14 (01 Sep)** | **TARGET go-live** — per-workflow gate (§8) |
| **W15–16 (Sep 1–15)** | **BUFFER (contingency only)** — slip / defect / verification catch-up |
| **15 Sep** | **HARD BACKSTOP** — unverified workflows defer to a change-controlled wave |

---

## 7. W6 kill-check (decide before July build expands)

WP-4 + WP-13b built/evidence-started · design-partner LOI signed-or-in-legal · G-8 draft in review · **build hand confirmed** · **contractor confirmed** · WP-12/WP-13 design ready for W7. **≥2 red ⇒ all-six by Sept 1 is unrealistic; narrow or re-baseline.**

**Enforcement (binding):** the kill-check is a control, not a discussion. **If two or more items are red at W6, the scope narrows or the baseline moves — automatically, no debate.** Recognition of the top-5 risks (WP-12, MIRA verification, G-8 by W9, O3, August QA load) is not control; the W6 gate is the control.

---

## 8. Per-workflow go-live rule (binding, no soft wording)

At the W14 gate each workflow is classified:

| Status | Meaning |
|---|---|
| **Production-verified** | Build complete · risk-based verification complete · evidence in G-7 · covered by signed G-8 · no open critical (integrity/AI/security/supplier) blocker |
| **Not production-verified** | Any required evidence missing |
| **Demo-only** | Shown in the evaluation environment only |
| **Deferred — controlled wave** | Ships later under change control |

Workflows not production-verified by W14 use the **Sep 1–15 buffer**; anything still short at **15 Sep** is **Deferred**. QA/CSA will not label a workflow "verified" without its evidence — the buffer does not change that.

---

## 9. Verification approach (CSV/CSA lens)

Risk-based, not exhaustive. **HIGH-risk → direct evidence:** e-sig/approval/closure, audit trail, RLS/tenant isolation, MIRA-to-record, inspection-export hash reproducibility. **MEDIUM → scripted/targeted.** **LOW → review/automated/supplier evidence.** Hard stop: no high-risk function ships without direct test evidence.

**Two binding operating rules (added at v3 review):**
- **G-7 is populated continuously from W7** as evidence is generated; W10–W13 is **closure, not first authoring** — otherwise QA is buried in August.
- **MIRA guardrail verification starts W8/W9** (as each guardrail lands) and runs through W12 — not deferred to August. MIRA remains a top-2 risk with QA-1 + AI-Owner concentration; front-load and automate the guardrail tests.

---

## 10. Top risks

| # | Risk | Mitigation |
|---|---|---|
| 1 | WP-12 not verification-ready (late build) | Build hand + front-load W7 + WP-10 pulled to W7 + weekly burndown |
| 2 | MIRA verification concentrated on QA-1 + AI Owner | Front-load guardrail tests; automate them; contractor takes non-AI verification |
| 3 | Design-partner LOI slips → G-8 can't sign by W9 | Commercial owner drives LOI in June; counsel template ready W1 |
| 4 | New hire ramp drag in Cat-5 codebase | Confirm build hand by W5; ring-fence to bounded work |
| 5 | Buffer treated as planning slack | Buffer is contingency only; teams plan to Sep 1 |

---

## 11. Final disposition

| Seat | Position |
|---|---|
| Head of QA | **Approved with conditions** — commit all six + the two required hires + §5 refinements + buffer; per-workflow gate (§8) intact; QA signs at each gate |
| AI Validation | **Acceptable with conditions** — MIRA advisory/limited-risk; WP-13b + WP-13 + V7 pack are hard conditions |
| CSV/CSA | **Approved with conditions** — all-six verification achievable under risk-based CSA + §3 capacity + §5/§6 front-loading; no high-risk function ships without evidence |
| Product/Eng Architect | **Ready with conditions** — §5 refinements adopted; WP-10→WP-12 dependency + PDF-lib decision resolved; proposed, requires engineering confirmation at W6 |

**Conditions (all must hold):** two hires secured · LOI by end June · G-8 signed by W9 · WP-12/WP-13 front-loaded to W7 · evidence burndown from W7 · CI gate W1 · PDF-lib decided W1 · QA hard stop (no evidence, no verified label).

---

## Source Ledger
| Source | Date | Supports | Limitation |
|---|---|---|---|
| Skills: head-of-qa · ai-validation-governance · csv-csa-assurance · product-engineering-architect (loaded) | 2026-05-31 | Multi-seat deliberation, conditions, refinements | Workbenches; human sign-off required |
| Repo audit `dev-vimal-audit-2 @ 3bd2fe47` (read-only) | 2026-05-31 | WP sizing; WP-10→WP-12 dependency; WP-12/MIRA latest poles | Static audit; `Proposed — requires engineering review` |
| FDA CSA (03-Feb-2026, supersedes 24-Sep-2025); GAMP 5 2nd Ed; EU AI Act 2024/1689 (GPAI Aug-2025) | verified | Risk-based CSA; supplier-verify/customer-validate; AI provider duties | Guidance/interpretation, not legal opinion |
| Status: hires · LOI · WP execution | — | — | `Unknown — evidence required` |

*Advisory only. Final go/no-go + risk acceptance are human QA + Founder decisions. Verixa never claims "validated." Read-only — no Verixa repo files modified.*