Status: Pre-engagement architectural prerequisite. Not actionable until hosted control plane MVP is committed. Owner: Vis lead + Compliance lead + Eng VP. Horizon: 18–24 months from architectural commit to ATO. Last revised: 2026-05-05.
Read this first. Enchanter-ai today is a workstation-installed CLI + VS Code surface. FedRAMP authorization applies to cloud service offerings (SaaS, PaaS, IaaS) consumed by federal agencies. There is no FedRAMP authorization to chase for a binary the customer runs locally. The unlock is the hosted control plane — without it, every dollar spent on 3PAO outreach is wasted. Section 1 is therefore the gate; sections 2–10 are conditional on it.
Cross-references:
compliance/fedramp-readiness/hosted-control-plane-prerequisite.md— architectural memo that established this gate.compliance/fedramp-readiness/conmon-plan.md— partial ConMon baseline already drafted (extended in §7).compliance/fedramp-readiness/control-coverage-matrix.md— NIST 800-53 Rev 5 control mapping seed.roadmap-2026/MACRO_ROADMAP.md— parent roadmap.
The hosted control plane is the minimum cloud-resident surface required to make FedRAMP authorization both technically meaningful and commercially valuable. Anything less, the agency cannot buy; anything more, the boundary expands and ATO cost balloons.
Six services. Nothing else inside the authorization boundary at MVP. Every additional service inflates the SAR and the annual ConMon burn.
- Audit-trail OTLP relay (per-tenant). Receives OpenTelemetry traces/metrics/logs from workstation agents over mTLS, fans into per-tenant write-isolated buckets. No mutation; relay-only. This is the load-bearing FedRAMP artifact — every action an agent took is queryable by the tenant's compliance officer.
- Pager fan-out. Receives signed alert events from agents (anomaly, policy violation, model-call cost spike), routes to tenant-configured destinations (PagerDuty, Opsgenie, agency SOC). No payload retention beyond 7 days.
- Multi-tenant rate-limiter. Token-bucket per tenant per upstream model provider. Prevents one tenant from exhausting another's quota. Persists usage counters; no agent payloads.
- SSO/SCIM. SAML 2.0 + OIDC + SCIM 2.0. Agency IdP integrations (Okta, Azure AD GCC High, Ping). User/group provisioning, just-in-time access, MFA delegation. No password storage.
- Tenant isolation boundary. The control-plane-wide enforcement layer: every request authenticates a tenant, every storage write is namespaced, every read is filter-gated. Implemented as middleware + storage-layer policy, not application code (defense in depth).
- Signed-config distribution. Workstation agents pull policy bundles (allowed models, denied tools, prompt-injection filters) signed by tenant admin keys. The control plane stores + serves; signing happens client-side. Zero secrets in transit, zero secrets at rest in the control plane.
What is not in the boundary at MVP: LLM inference (stays at customer-chosen provider, BAA/DPA managed separately), agent code (workstation-resident), source code (never leaves customer machine), git operations (local-only).
┌─────────────────────────────────────────┐
│ Agency Workstations (out-of-scope) │
│ ┌────────┐ ┌────────┐ ┌────────┐ │
│ │ Agent │ │ Agent │ │ Agent │ ... │
│ └────┬───┘ └────┬───┘ └────┬───┘ │
└───────┼───────────┼───────────┼─────────┘
│ mTLS │ mTLS │ mTLS
▼ ▼ ▼
┌───────────────────────────────────────────────────┐
│ FedRAMP Authorization Boundary │
│ ┌────────────────┐ ┌────────────────────────┐ │
│ │ Edge / WAF / │ │ IdP Integration │ │
│ │ TLS Termination│ │ (SSO, SCIM) │ │
│ └────────┬───────┘ └────────────────────────┘ │
│ │ │
│ ┌────────▼──────────────────────────────────┐ │
│ │ Tenant Isolation Middleware │ │
│ └────┬─────────────┬─────────────┬──────────┘ │
│ │ │ │ │
│ ┌────▼─────┐ ┌────▼─────┐ ┌───▼──────────┐ │
│ │ OTLP │ │ Pager │ │ Rate-limiter │ │
│ │ Relay │ │ Fan-out │ │ │ │
│ └────┬─────┘ └──────────┘ └──────────────┘ │
│ │ │
│ ┌────▼──────────────────────────────────┐ │
│ │ Per-tenant Encrypted Object Storage │ │
│ │ + Time-series DB (counters/usage) │ │
│ └───────────────────────────────────────┘ │
│ │
│ ┌──────────────────────────────────────┐ │
│ │ Config Distribution (signed bundles) │ │
│ └──────────────────────────────────────┘ │
└────────────────────────────────────────────────────┘
Build on a FedRAMP-authorized substrate. Inheriting an authorized IaaS removes ~250 of NIST 800-53 controls from our scope. Without inheritance, we re-implement physical security, hypervisor controls, DC personnel screening — economically nonviable for a small team.
| Layer | Choice | Rationale |
|---|---|---|
| IaaS | AWS GovCloud (US) — FedRAMP High authorized | Inherits ~250 controls; widest agency familiarity; native KMS, IAM, CloudTrail integrations |
| Compute | EKS on GovCloud + Fargate for serverless tiers | Container isolation, no node-level patching burden on us |
| Storage | S3 (per-tenant prefix + bucket policy + KMS-CMK per tenant) | Cryptographic isolation, FedRAMP-blessed |
| Time-series | Timestream or Managed Prometheus (GovCloud) | Counters/rate-limit state; no agent payload data |
| Edge | CloudFront + WAF (GovCloud regional) | DDoS, OWASP Top 10 baseline |
| IdP | Cognito Federation or PingFederate (FedRAMP High) | SAML/OIDC broker; integrates with agency IdPs |
| Secrets | AWS Secrets Manager + KMS-CMK | FIPS 140-2 validated |
| Logging | CloudTrail + CloudWatch Logs → tenant-isolated S3 export | Auditor-required immutable log retention |
| CI/CD | GovCloud-resident CodePipeline + signed artifacts | Build provenance into the boundary |
Alternatives considered:
- Azure Government — equivalent FedRAMP posture; viable if the sponsor agency is Azure-standardized (DoD-adjacent, some DHS components).
- GCP Assured Workloads — narrower agency footprint, fewer reference customers; rejected at MVP, revisit at scale.
Three concentric layers; each independently sufficient to prevent cross-tenant leak, all three required for defense in depth:
- Logical (application). Every API surface is tenant-scoped via a
tenant_idclaim on the SSO JWT; middleware rejects any request whose claim doesn't match the URL path tenant. Unit-tested, fuzz-tested, red-teamed. - Physical/storage (infrastructure). Each tenant gets a dedicated S3 prefix + dedicated KMS Customer Managed Key. IAM policies deny cross-tenant decrypt at the KMS API level — even a compromised application can't read another tenant's data without KMS authorization.
- Network (boundary). VPC-level isolation per environment (gov-prod, gov-staging); no shared subnets with non-fed workloads.
Tenant key model: one KMS-CMK per tenant, rotated annually, deletable on contract termination (BYOK option available for agencies with internal KMS).
| Surface | Mechanism |
|---|---|
| Transit (agent ↔ control plane) | TLS 1.3 mandatory, mTLS for OTLP and pager fan-out, certificate pinning client-side |
| Transit (control-plane internal) | mTLS via service mesh (App Mesh or Istio), TLS 1.3 only |
| Rest (object storage) | SSE-KMS with per-tenant CMK, FIPS 140-2 Level 2 validated HSMs |
| Rest (databases) | TDE with KMS-CMK, automated backup encryption |
| Rest (logs) | CloudWatch Logs → S3 export with SSE-KMS, immutability via Object Lock (Compliance mode, agency-configurable retention) |
| Secrets | Secrets Manager + KMS-CMK, no plaintext at rest anywhere |
| Key management | AWS KMS in GovCloud (FIPS 140-2 validated); BYOK supported via External Key Store for agencies with internal HSMs |
Three baselines (Low / Moderate / High) corresponding to FIPS 199 impact categorization.
| Baseline | Controls | Suited to | Typical SaaS examples |
|---|---|---|---|
| Low | ~125 | Public-facing, no CUI | Marketing sites, public APIs |
| Moderate | ~325 | CUI, non-NSS, most agency adoption | Slack Gov, GitHub Enterprise Cloud Gov, most SaaS |
| High | ~425 | Law enforcement, emergency services, financial | Salesforce Government Cloud Plus, AWS GovCloud itself |
Recommendation: Moderate baseline.
Categorization rationale per NIST SP 800-60 Vol. 2 Rev. 1:
- Information types stored: agent activity logs, usage telemetry, tenant policy bundles, anomaly alerts. No raw source code; no agency mission data; no PII beyond user identifiers.
- Confidentiality: Moderate. Agent telemetry, if exposed, could reveal development priorities and tooling configuration — non-trivial but not catastrophic.
- Integrity: Moderate. Policy bundle tampering could lead to allowed-tool bypass; tamper-evident signing mitigates but doesn't reduce the categorization.
- Availability: Moderate. Control-plane outage degrades agency dev velocity but doesn't halt mission operations (workstation agents continue with cached policy).
Overall categorization: Moderate (M-M-M). High is over-scoped (and roughly 3× more expensive); Low is under-scoped (most agencies reject Low for any system holding agency telemetry).
Two paths to FedRAMP authorization.
| Path | Authorizer | Reach | Difficulty | Time | Cost |
|---|---|---|---|---|---|
| JAB P-ATO | Joint Authorization Board (DOD + DHS + GSA CIOs) | Government-wide reusable | Hardest; ~12 CSPs/year admitted | 18–24mo plus ~6mo queue | +30–50% premium |
| Agency ATO | Single sponsoring agency | Reusable but requires per-agency package review | Moderate; gated on sponsor | 12–18mo | Baseline |
Recommendation: Agency ATO with sponsoring agency identification as a Sales motion.
Trade-off:
- JAB P-ATO gives ecosystem reach (one auth, all agencies inherit) but JAB only admits ~12 CSPs/year via the FedRAMP Connect program; admission itself is competitive and gated on demonstrated demand across multiple agencies. For a pre-revenue government segment, JAB admission is unlikely.
- Agency ATO is achievable with one design-partner agency. Other agencies can then reuse the package (a "FedRAMP Authorized" listing in the Marketplace), shortening their internal ATO from 12mo to ~3mo. This is the standard path for small/midcap SaaS.
Strategy: pursue Agency ATO with the sponsor identified in §5; once authorized and listed, market the reusable package to subsequent agencies as a 90-day on-ramp.
Third-Party Assessment Organizations conduct the SAR. They must be FedRAMP-accredited (A2LA) and independent of the CSP.
| Firm | 3PAO-Accredited | AI/Dev-Tools Experience | Initial Assessment | Annual ConMon | Lead Time | Notable References |
|---|---|---|---|---|---|---|
| A-LIGN | Y | Strong — GitHub, Snowflake (commercial), several MLOps SaaS | $350–500k | $80–150k | 6–9mo | Largest 3PAO by volume; thorough but slower scheduling |
| Coalfire | Y | Strong — many cloud-native SaaS, some AI tooling | $300–450k | $70–130k | 4–8mo | Best-known FedRAMP brand; mature ConMon practice |
| Schellman | Y | Moderate — broad SaaS, less dev-tooling specific | $275–425k | $60–120k | 4–7mo | Aggressive pricing; SOC 2 + FedRAMP bundled discount possible |
| BAI Security | Y | Moderate — boutique, more agency-side relationships | $200–350k | $50–100k | 3–6mo | Smaller, faster, less brand weight with skeptical sponsors |
| Kratos | Y | Light on commercial SaaS; strong defense/intel | $325–475k | $75–140k | 5–8mo | Good for defense-adjacent sponsors (DoD components) |
| Aprio | Y | Light — generalist; new entrant to FedRAMP | $225–375k | $55–110k | 3–6mo | Lowest cost; reference base thin; risk on first FedRAMP assessment |
Shortlist for RFP: A-LIGN, Coalfire, Schellman. Issue parallel RFPs (see §4.1) with identical statement of work; select on price + earliest available kickoff + reference-call results.
- System scope. Hosted control plane MVP per §1, FedRAMP Moderate baseline, AWS GovCloud substrate.
- Deliverables. SAP, SAR, Readiness Assessment Report (RAR) optional pre-engagement, ConMon attestation framework.
- Timeline. Kickoff within 60 days of award; SAP within 30 days of kickoff; SAR within 90 days of testing start.
- Pricing structure. Fixed-fee initial + per-month ConMon retainer; not T&M.
- Independence. No advisory engagement with us in the preceding 24 months (FedRAMP-required separation).
- References. Minimum 3 FedRAMP Moderate SaaS authorizations completed in prior 36 months.
- AI workload familiarity. Question on the firm's experience assessing systems with LLM-augmented agents — pricing variance often signals comfort level.
Agency ATO requires a federal agency willing to issue the authorization. The agency carries assessment-package liability and gets first-mover access to the tool. This is a Sales motion, not a compliance motion — treat it as such.
| Agency | AI Tooling Adoption Posture | Entry Point | Why a fit |
|---|---|---|---|
| DHS | Active — DHS AI Task Force; Silicon Valley Innovation Program (SVIP) funds early-stage tools | DHS S&T SVIP solicitations; CISA secure-dev initiatives | Mission alignment with secure software development; existing SaaS authorization pipeline |
| GSA | Active — TTS (10x, Centers of Excellence) explicitly funds gov-software-modernization tools | TTS 10x program; GSA's AI Center of Excellence | TTS authorizes its own tools and shares packages with sister agencies; lowest-friction sponsor |
| DOE | Moderate — national labs adopt AI dev tools aggressively (ORNL, LLNL); HQ slower | Lab CIO offices; DOE Office of the CIO | Labs run their own ATO boundaries; can sponsor for HQ if lab adoption demonstrates value |
| DOT | Light — limited AI dev-tooling adoption to date | Volpe Center; FAA AI initiatives | Lower priority; revisit after first sponsor |
Recommended primary: GSA TTS. Lowest friction; TTS routinely sponsors small SaaS for internal use, then makes the package available; well-understood ATO workflow.
Recommended secondary (parallel pursuit): DHS SVIP. Funding + sponsor potential combined; explicitly targets early-stage tools.
- Week 0–4: Compile capability brief (2-pager): what Enchanter does, why it matters for federal dev velocity + security, hosted-MVP architecture summary, FedRAMP roadmap commitment.
- Week 4–8: Warm intros via existing federal-adjacent network (Coalition for Sensible Safeguards, federal-contractor advisory firms, ex-USDS network). Cold outreach via SAM.gov + agency innovation-lab public contacts.
- Week 8–16: Discovery calls. Demonstrate workstation agent + describe hosted control plane on the roadmap. Ask: would your office sponsor an Agency ATO if MVP ships by [date]?
- Week 16–24: MOU or LOI with sponsor. Co-define use case + pilot terms. Sponsor agrees to issue ATO conditional on SAR clearance.
- Week 24+: Begin 3PAO engagement (§4) with sponsor named in the SAP.
Enchanter is an vis platform that lets federal development teams use AI coding assistants without source code, prompts, or developer activity ever leaving the workstation. The hosted control plane — currently in design for FedRAMP Moderate authorization on AWS GovCloud — provides agency CIOs the audit trail, tenant isolation, and policy enforcement they require, while keeping the LLM-inference surface at the developer's provider of choice. We are seeking a sponsoring agency for our Agency ATO pursuit and offer pilot access in exchange for sponsorship.
The three documents that move a FedRAMP package from "in assessment" to "Authorized".
| Doc | Produced by | Purpose | Typical length |
|---|---|---|---|
| SAP (Security Assessment Plan) | 3PAO | Scope, methodology, testing schedule, sample sizes for control testing | 50–80 pages |
| SAR (Security Assessment Report) | 3PAO | Findings from executing the SAP; per-control pass/fail/risk-accepted | 200–400 pages |
| POAM (Plan of Action & Milestones) | CSP (us) | Tracking of every SAR finding to remediation or risk-acceptance | Living spreadsheet/JSON |
Cadence (mature state):
- SAP: revisited annually as part of ConMon annual assessment.
- SAR: produced once at initial ATO; full re-test annually under ConMon (a subset re-tested each year on a rotating schedule).
- POAM: updated monthly with remediation progress. Submitted to authorizing agency monthly. New findings (from ConMon scans, incident response, etc.) appended within 30 days of discovery.
Sample POAM entry shape:
poam_id: POAM-2027-014
control: AC-2(7) # Account Management - Role-Based Schemes
finding: SCIM provisioning does not auto-deprovision on agency IdP user disable within 4 hours
severity: Moderate
discovered: 2027-03-14
discovered_by: 3PAO annual assessment
remediation_plan: Implement SCIM event listener with 1h reconciliation window
milestones:
- 2027-04-15: Design complete
- 2027-05-15: Implementation complete
- 2027-06-01: Re-test by 3PAO
status: in-progress
risk_accepted: falseConMon makes FedRAMP authorization a continuous obligation, not a point-in-time check. The baseline lives in compliance/fedramp-readiness/conmon-plan.md; extensions specific to the hosted control plane:
| Monitor | Frequency | Source | Alert destination |
|---|---|---|---|
| Per-tenant OTLP relay throughput + error rate | Real-time | CloudWatch metrics on OTLP endpoints | Pager fan-out + monthly POAM |
| Cross-tenant access attempts (denied) | Real-time | Application audit log → SIEM | SOC + immediate POAM entry if > 0 |
| KMS decrypt-denial events | Real-time | CloudTrail | SOC + POAM if > threshold |
| Tenant isolation middleware test suite | Daily CI | Automated red-team fuzzing | Eng + monthly POAM rollup |
| Vulnerability scans (containers + dependencies) | Weekly | ECR scan + Snyk/Trivy | Eng + monthly POAM |
| Penetration test | Annual | Independent firm (not the 3PAO) | Full POAM entry per finding |
| Configuration drift | Daily | AWS Config + custom Conftest policies | Eng + monthly POAM |
| Anomaly detection on usage patterns | Real-time | ML-based outlier detection on rate-limiter counters | Pager fan-out for confirmed anomalies |
- Aggregate findings from all monitors above for the prior month.
- Classify: new / in-progress / resolved / risk-accepted.
- Compute remediation aging: any High open > 30 days, any Moderate > 90 days, any Low > 180 days requires explicit risk-acceptance signature from CSO + sponsoring agency AO.
- Submit POAM update by 5th business day of following month to sponsoring agency.
Be honest. FedRAMP is expensive and slow. The numbers below are realistic for a small/midcap SaaS; under-budgeting here is the most common cause of mid-engagement abandonment.
| Phase | Duration | Spend Range | Major line items |
|---|---|---|---|
| Year 0 — Architectural commit + sponsor pursuit | 6–9 months | $300–500k | Hosted MVP design + early build (eng), compliance lead hire (FTE or fractional), sponsor outreach travel + capability briefs, FedRAMP advisor retainer (~$15k/mo × 6mo), readiness gap assessment (optional RAR from 3PAO ~$50k) |
| Year 1 — Hosted build + 3PAO engagement | 12 months | $400–800k | 3PAO initial assessment ($300–500k), continued hosted-MVP eng (~$200k incremental), GovCloud infra ($30–80k/yr), penetration test ($50–80k), security tools (SIEM, vuln scanning) ($50–100k/yr) |
| Year 2 — ATO finalization + ConMon ramp | 12 months | $200–400k | 3PAO ConMon retainer ($60–150k/yr), GovCloud infra growth, FedRAMP PMO fees (~$5k/yr), compliance FTE fully loaded, remediation eng for SAR findings |
| Year 3+ — Steady state | annual | $250–500k/yr | Annual ConMon assessment, infra, compliance team, penetration test, remediation eng |
Realistic time-to-ATO:
- Best case: 18 months from architectural commit, if sponsor is identified by month 6, hosted MVP ships by month 12, 3PAO kickoff at month 12, no major SAR findings.
- Typical: 24 months. SAR finds ≥ 30 items requiring remediation; remediation + retest pushes ATO 6+ months.
- Slip case: 30+ months. Sponsor agency turnover, 3PAO schedule slip, significant architectural rework post-SAR.
Three-year ATO cost envelope: $900k–$1.7M before steady-state ConMon. Most small SaaS underestimate by ~40%.
3PAO engagement is rational only when all of the following hold. Engaging earlier wastes 3PAO fees on a system that isn't assessable yet.
- Hosted control plane MVP (§1) deployed to GovCloud in a dedicated
gov-prodaccount with at least one design-partner tenant onboarded. - Sponsoring agency (§5) has issued an LOI or MOU naming us as the system under assessment.
- FIPS 199 categorization document signed by CSO and sponsor AO designee (Moderate per §2).
- System Security Plan (SSP) drafted to 80%+ — every control in the NIST 800-53 Rev 5 Moderate baseline addressed, even if "planned" rather than "implemented" for some.
- Compliance lead in place (FTE or fractional with ≥ 5 prior FedRAMP authorizations).
- FedRAMP advisor retainer in place (independent of the chosen 3PAO).
- Year-1 budget approved by board / executive sponsor; not contingent on funding round.
- Penetration test completed by an independent firm in the prior 6 months; findings remediated or risk-accepted.
- ConMon tooling (SIEM, vuln scan, config drift) operational and producing signal for ≥ 90 days.
- All inheritable controls from AWS GovCloud documented in the SSP with the CSP-customer responsibility matrix.
- Tabletop incident-response exercise completed; IR runbook reviewed by sponsor.
If any item is "in progress" rather than "complete," delay 3PAO kickoff. The cost of a 60-day delay is ~$25k; the cost of failing initial SAR is ~$150–300k in remediation + retest.
FedRAMP is a strategic commitment, not an opportunistic one. Pursue only if all four are true. Any "no" → defer; revisit in 6 months.
| Gate | Required state | Source of truth |
|---|---|---|
| (a) Government design partner | ≥ 1 federal agency has expressed concrete intent to use Enchanter in a hosted-control-plane configuration, ideally with LOI | Sponsor outreach log per §5 |
| (b) Hosted MVP roadmap commitment | §1 hosted control plane on the engineering roadmap with quarterly milestones and named tech lead | roadmap-2026/MACRO_ROADMAP.md |
| (c) Year-1 budget approval | ≥ $400k allocated in year-1 budget specifically for hosted build + 3PAO; not fungible with other lines | CFO sign-off in budget doc |
| (d) C-suite mandate | CEO + CTO + CSO have all signed off on the 24-month commitment in writing | Board memo or signed strategy doc |
If 4/4 → green-light, execute §1 → §5 → §4 → §6 in order. If 3/4 → fix the gap before any 3PAO spend. Sponsor outreach (§5) and architectural work (§1) are no-regret; 3PAO RFP (§4) is regret-heavy if any gate is unmet. If ≤ 2/4 → defer. Park this package; revisit quarterly.
- ATO — Authority to Operate. The authorization a federal agency issues a system before agency users may operate it.
- AO — Authorizing Official. The agency executive who signs the ATO.
- CSP — Cloud Service Provider. Us, in this context.
- ConMon — Continuous Monitoring. The ongoing security posture maintenance required post-ATO.
- CUI — Controlled Unclassified Information. Information requiring safeguarding under federal law but not classified.
- JAB — Joint Authorization Board. DOD + DHS + GSA CIOs; issues P-ATOs reusable across agencies.
- P-ATO — Provisional ATO. JAB-issued; agencies still issue their own ATO citing the P-ATO but with lower friction.
- POAM — Plan of Action & Milestones. Tracking of remediation work for SAR findings.
- 3PAO — Third-Party Assessment Organization. A2LA-accredited firm that performs the SAR.
- SAP — Security Assessment Plan. 3PAO's testing plan.
- SAR — Security Assessment Report. 3PAO's findings.
- SSP — System Security Plan. CSP-authored description of the system + how each control is met.
- SVIP — Silicon Valley Innovation Program. DHS S&T's startup engagement vehicle.
- TTS — Technology Transformation Services. GSA's modern-dev arm; runs 10x and the COEs.
This package depends on, and extends, these compliance artifacts:
compliance/fedramp-readiness/hosted-control-plane-prerequisite.md— establishes the §1 gate.compliance/fedramp-readiness/conmon-plan.md— extended by §7.compliance/fedramp-readiness/control-coverage-matrix.md— NIST 800-53 Rev 5 mapping seed; expanded into the SSP.roadmap-2026/MACRO_ROADMAP.md— must reflect the hosted-MVP commitment for §10(b) to evaluate true.
- This is hard. Most pre-revenue SaaS that start FedRAMP pursuit do not finish. The dropout point is usually month 14–18, mid-3PAO, when the SAR comes back with 60+ findings and the cap table revisits whether the federal segment is worth the burn.
- The sponsor is the unlock. No sponsor → no agency ATO → no marketplace listing → no other agency reuses → zero ROI on the spend. §5 outranks §4 in sequencing risk.
- The hosted MVP is the bigger unlock. Without it there is no system to assess. §1 outranks §5 in technical risk.
- Estimates assume favorable execution. A real engagement adds 20–40% to numbers above for unforeseen remediation. Plan accordingly.