Devensoft · Business transformation · Executive brief — the one thing

Keep the Lights On — The RETAIN Critical Path

The transformation plan trimmed to the least minimally necessary content: one serial chain, three parallel items, nothing else. If everything else slips, the business survives. If this chain slips, it does not.

~27.5–40.5 dev-wkserial chain — F6 → F1 → F5 → F3 → F7 (cluster-sum)
~10–12 dev-wk moreverticals — AI posturing · Zapier · (self-serve reporting rides F7, in the chain)
1 hard gateF6 → F1 — every other arrow is crew sequencing

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

F6
DI / runtime bridge completion
5–8 dev-wk left already started
  • 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
hard gate
F1
Auth-provider migration (Clerk / Auth0)
9–13 dev-wk * audit & compliance mandated
  • 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)
crew seq.
F5
Infra resilience
5–8 dev-wk compliance mandated
  • 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
crew seq.
F3
Database reliability
3–6 dev-wk performance necessary
  • 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
crew seq.
F7
Reporting-data foundation
~5.5 dev-wk minimal — creates bandwidth
  • 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.

G5a
DevenMCP sliver
~4–6 dev-wk · HIGH confidence

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.

R4
Self-serve reporting
rides F7 · in the chain

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.

G4
Zapier app
~6 dev-wk · LOW confidence

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.

Why these can run in parallel: none of them depends on F1. They ride the existing API surface, not the F1-gated future one. That independence is the slip absorber (§4): if the chain's auth migration runs long, the verticals keep shipping visible progress.

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.

MetricValue
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.
One honest caveat: the crew-sequencing arrows mean the chain's calendar length depends on allocation discipline — protecting the chain crew from context-switching is what keeps 27.5–40.5 dev-weeks from becoming more calendar than it should.
Envelope + ramp + ~4–7 months + ±1-month band — the locked roadmap's phase-gate, critical-path, and sizing-dispatch sections · compression rule (floor protected first) — the pivotal-bet decision's sequencing rule as carried in the roadmap · slip-panel structure — Diego-confirmed qualitative scope.

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 exclusion is a scoping call, not a safety claim: F4 carries the flagged audit risk — KeyVault is disabled and secrets sit in plaintext in committed config files, "a security audit failure waiting to happen" — and F2's broken legacy health endpoint silently fails the probes the release pipeline depends on. Leadership should know both are outside the bare minimum and inside the full plan's floor.

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.