1 The One Thing ▾
The keep-the-lights-on chain: F6 → F1 → F5 → F3 → F7 — plus three items that run alongside it. Nothing else is absolutely mandatory for RETAIN.
Five workstreams, ~27.5–40.5 dev-weeks of serial effort (cluster-sum of the locked sizings). F6 is already started. F1 and F5 are audit-and-compliance mandated. F3 is the performance work the platform genuinely needs. F7 is the minimal reporting foundation that creates bandwidth back.
Alongside the chain — parallelizable, with no F1 dependency: the DevenMCP sliver (~4–6 dev-wk, HIGH confidence) for necessary AI posturing, and the Zapier app (~6 dev-wk) to make the API story sellable — together ~10–12 dev-weeks. The third vertical, self-serve reporting, arrives with F7 inside the chain: F7 is the self-serve reporting foundation.
Everything else in the plan — the Grow AI tiers, the Diversify vehicles, the GTM motion — rides above or after this chain and is deliberately out of this view (§5). The framing is blunt on purpose: the first stretch of the plan is spent on prerequisites, not on visible product. That is not a failure of ambition; it is what keeping the lights on costs.
Chain selection + node reasons — Diego directive, 2026-08-19 · sizings — the fundamentals sizing brief (grilled & locked) · structure — the locked roadmap.2 The Chain — F6 → F1 → F5 → F3 → F7 ▾
- Static → Non-Static Phase 2: ~70% of the epic spent — the unspent back half is the harder one
- 6 unstarted children, all unestimated (incl. high-blast-radius EmaControllers)
- Sized by analogy to completed children × the 1.5× large-work calibration
- Today: token-presence-only OAuth — any valid token reaches all endpoints, no scopes
- Buy-not-build: managed provider replaces Forms Auth + client-credentials
- * Pre-re-sizing — scope grew to full provider integration; the live dispatch item that moves the whole tail (§4)
- Auto-healing first: zero Auto Scaling Groups today — no recovery on hardware failure
- ~20 open small infra-debt items (tag hygiene, public-IP tightening, uptime-check port)
- DR re-enablement excluded — a network re-architecture, a separate strategic call
- Root cause of midday multi-tenant timeouts: hardcoded 30s CommandTimeout, still in place
- Timeout hardening, EF resiliency, ReadOnlyDbContext, pool protection, circuit breaker
- 11 of 16 remediation children still unstarted
- Dataset curation for 11 entities; per-tenant CoreReport provisioning (TOM/XMLA)
- 4-template starting catalog: Financials · Activity · RAID · Deal Pipeline
- Medium confidence (near-perfect analog, zero build actuals) — the chain's payload
Arrow semantics: one small crew walks the chain in order — the arrows are crew sequencing (priority order: compliance-mandated → performance → bandwidth), except F6 → F1, the one hard structural gate: auth/session refactoring requires clean dependency injection — the 325-symbol / 69-dependent session root cannot be untangled while the static→non-static bridge is half-built.
Sum: ~27.5–40.5 dev-weeks serial (cluster-sum; the full F1–F7 contractual floor runs ~32.5–49.5 — the chain is a strict subset; see §5 for what was left out). F6 and F1 alone consume roughly four to seven of the nine months at realistic allocation. The chain is not headcount-fixable: each node's start is its predecessor's finish.
Sizings + evidence — the fundamentals sizing brief's cluster sections (incl. the reporting-foundation addendum for F7) · F1 scope change — the auth-migration decision record (buy-not-build) · hard gate + ~4–7 months — the locked roadmap · arrow semantics + node reasons — Diego directive, 2026-08-19.3 The Verticals — what runs alongside ▾
Three items are mandatory for RETAIN: two run off-chain in parallel, and the third — self-serve reporting — arrives with F7 inside the chain.
Necessary AI posturing. The in-horizon slice of the DevenMCP runtime — enough MCP surface to make the "we have AI" claim true and demonstrable. The full runtime core (~22–34 dev-wk) is post-horizon and F1-gated — deliberately out of this view. The highest-confidence item on this page.
Creates more time. Delivered with F7: the foundation node (dataset curation, per-tenant provisioning, 4-template catalog) is itself the self-serve reporting base, already counted in the chain. The recommendations' authoring layer above it is deliberately unsized separately in the locked record — treat F7's number as the reporting bet's cost. Every self-served report is engineer time returned to the chain — the lowest-effort answer to the #1 capability demand in the evidence base.
Distribution. The Zapier integration exposing the DevenConnect API to non-developers — "all integrations enabled by Zapier" once available. Explicitly not on any critical path: ships when ready.
Sum: ~10–12 dev-weeks of additional parallel effort (sliver + Zapier), with self-serve reporting arriving as part of F7 in the chain rather than as a separate line. The sequencing logic of this whole brief in one sentence: the chain defends the revenue base; the verticals make the base defensible — AI-postured, self-serving, and integrable — while the chain grinds.
Sliver split + sizings — the recommendations' grow inventory (sliver ~4–6w H in-horizon; full core ~22–34w post-horizon) · self-serve reporting — the sizing brief's F7 (~5.5 dev-wk, the self-serve report-creation foundation) + the recommendations floor (authoring layer above F7, unsized separately) · Zapier ~6w L, not on critical path — the net-new sizing table · no-F1-dependency — the locked roadmap · vertical selection — Diego directive, 2026-08-19.4 Capacity & slip — does it fit, and what if it doesn't ▾
The capacity frame, carried from the locked plan — not re-derived here.
| Metric | Value |
|---|---|
| Total effective capacity (9-month envelope) | 105.3 dev-wk |
| Committed to the Retain / fundamentals floor | ~43.8 dev-wk |
| Headroom for everything else (Grow / Diversify) | ~61.5 dev-wk |
| Capacity ramp (effective FTE) | ~2.0 (mo 1–2) → ~3.5 (mo 3–6) → ~4.5 (mo 7–9) |
| Keep-the-lights-on minimum (chain + sliver + Zapier, cluster-sum; reporting rides F7) | ~37.5–52.5 dev-wk |
The minimum sits inside the committed-floor neighborhood of the envelope. The rest of the plan's ambition lives in the headroom; this rendering spends none of it.
The slip panel — qualitative, no calculator
- F1 is the swing factor. Mid-chain, scope-changed (full provider integration vs. the sizing brief's narrow enforcement surface), re-sizing live. F1 running long pushes F5, F3, F7 right — month for month; F1 running short pulls the whole tail earlier. Until the dispatch closes, phase-gate windows carry an implicit ±1-month band.
- The chain slides; the verticals keep running. The verticals' F1-independence is not a convenience — it is the slip absorber. Auth migration runs long → the sliver, self-serve reporting, and Zapier continue shipping visible progress while the serial tail slides.
- The minimum has no give. Everything on this page is mandatory for keeping the lights on. There is no scope below the minimum to trade away — anything cut from the chain is lights-on risk by definition. Under compression in the full plan, the fundamentals floor is protected first and the Grow/Diversify scope gives ground before it.
- F7's already-minimal scope is the last defense. If the late window compresses, F7 is the node that can shed weight last — it is scoped "minimal to create bandwidth" precisely so that a hard landing still lands something.
5 Boundaries & pointers — what was deliberately left out ▾
Floor items excluded from the minimum
The contractual floor runs F1–F7 (~32.5–49.5 dev-wk cluster-sum); the chain uses five of the seven. Deliberately excluded: F2 — health checks + observability (3–6 dev-wk) and F4 — CI/CD + secret hardening (2–3 dev-wk).
The full plan's critical path continues past this view
The plan-level critical path (the roadmap module's) runs F6 → F1 → T2 (DevenConnect) → T3 (extraction → Diversify) → T7 (GTM) — the serial chain out to the growth bets. This rendering stops at RETAIN by design; the Diversify tier-1 vehicle question, the Grow AI tiers, and the GTM motion live in the main plan.
Dispatch items that move this page's numbers
F1 re-sizing (scope changed to full provider integration; the auth-migration decision record) is the one live measurement that rewrites the tail. T2 sizing (unsized) matters only to the full plan's path, not this one.
Provenance basis
Chain selection, node reasons, and vertical selection: Diego directive 2026-08-19 (this rendering's scope confirmation). All sizings and evidence: the fundamentals sizing brief (grilled & locked; F7 = the self-serve reporting foundation at ~5.5 dev-wk), the roadmap module, the recommendations module, and the auth-migration decision record (buy-not-build). Provisional flags carried from source: the FY26 financial context behind the RETAIN frame is an internal projection (unaudited); F7 confidence is Medium (zero build actuals); F6's remaining children are unestimated (sized by analogy × 1.5).
The one thing, restated: F6 → F1 → F5 → F3 → F7, plus the three verticals. Nothing else keeps the lights on.