How We Win Diego + Roman · recommendation model factors editable live · debate-ready

spine = BTDR-0004 9-month horizon · agentic-adjusted capacity Hedge, sequenced — fundamentals-first (BTDR-0004 G4) D1 vehicle ← LMM candidate matrix

What should we recommend? The pivotal bet is the context. The path variables are live: the LMM vehicle (resolved by the sibling candidate matrix), the AI feature scope (the BYOM build), and how the iron triangle resolves. An explicit factor model — engineering FTEs × window × onboarding drag — sets the capacity bar in real dev-weeks. The agentic coding multiplier compresses scope: work estimated at 41w takes 31.5w at 1.3×. Flip a path variable → the bar, the posture, the risk readout, and the recommended path re-shape.

Bet of record: BTDR-0004 · AI build: devensoft-ai-integration-deck (scaled per TSK-32) · Fundamentals: TSK-30 brief + F7 via TSK-33 · LMM vehicle input: lmm-candidate-matrix.html · M14 grill: 02_grill.md · M14 spec: spec.md · Pattern: decision-support-prototype.html
the fork Overarching path — Bold DevenMnA vs Hedge resource reality, never ultimatum · the unchosen path pauses · TSK-49 #4
bold DevenMnA — the greenfield standalone (Growth-focused) unchosen Hedge arm pauses · TSK-49 · fit-to-need AGREED 2026-08-18
What it is: a greenfield standalone — DevenLite + minimal DevenDiligence + minimal integration-reliant VDR; opinionated, less configurable; not quite a rewrite fit-to-need.json. The LMM vehicles are subsumed as DevenMnA components — incl. the C3/C4/C6 SDE/valuation micro-products TSK-49 #7 — and the DDD core sits at value realization, exactly where every current column is weakest (5 contexts) devenmna-greenfield-thesis. Fit-to-need (AGREED): the only column with native lifecycle stages (4 — incl. DD 93, the matrix's top score, and value-realization 87); v23 and v24 carry zero native stages. fit-to-need-canonical Scoring semantics: the fit-to-need scores = full target scope / definition of done — the 9-month commitment is the MVP: a general ~50% of each stage's need-to-fit scope, thin breadth across all 9 stages, not 100% of a few Diego 2026-08-18 · Am.2.
What runs (bold staffing — Am.1 + Am.2)
Diego + RomanDiego 0.5 build + 0.5 venture management (team, product, finances, vendors) — fully decoupled from Devensoft platform operations; Roman 1.0 — crew = 3.5 FTE TSK-49 Am.2
2 Talmatic engineers$5k each = $10k/mo (net +$10k/mo) — join the crew; "2 architects + 2 engineers × agentic" TSK-49 Am.1 5c
KAZ ~$10k/morestructured staff-aug → full hand-off: v23 maintenance + self-serve reports + builds DevenMCP (the "we have AI" signal); custom report dev curtailed TSK-49 Am.1 5b
$1.9M FY26 basecarried — provisional flag
What pauses
F1–F6fundamentals — auth/session, health, DB, CI/CD, infra, DI bridge
AI-deck P1–P3the BYOM build
Tier-2 instrumentsdemand-signal validation for C3–C6
DevenLite-standaloneabsorbed into DevenMnA TSK-49 #6
The Hedge planall D1/D2/D3 path variables, factor model + presets below
Timing reference baseline (Diego 2026-08-18): a production-ready MVP in the same 9-month window as Hedge, committing a general ~50% of each stage's need-to-fit scope — thin breadth across all 9 stages, not 100% of a few. The build itself remains unsized in dev-weeks (TSK-30 ≥1w×1.5 first-cut rule; §6 fog #1); the sizing deep-dive gate validates the ~50% slice. Capacity envelope — NOT build sizing (bold-path factors only, Am.2 tiered per-person — supersedes Am.1 5d in part): (0.5+1.0)×2.4× + 2.0×1.3× = 6.2 eff-FTE × 39w × 0.5h/8h-day overhead (93.75% eng-time) ≈ ~227 dev-wk — the gate's feasibility equation: 50% of full-scope effort ≤ ~227 dev-wk. Hedge keeps the existing model (2h/8h + current factors — 105.3/43.8/61.5 stand; committed net-new is branch-dependent: DevenLite ≈ ~47–53, DevenDiligence ≈ ~18–26, no choice = Tier-2 signal work only). No demand evidence — lifecycle-fit phrasing only ("what the market wants" is banned in the pitch). NLM 77–98% superseded — never cited. TSK-48 fit-to-need verification complete 2026-08-19 (no load-bearing cell moved). BTDR-0010 Am.2 filed (2026-08-18): Hedge Tier-1 = the LMM Vehicle Choice — deferred leadership decision, no vehicle committed, none recommended ("DevenDiligence recommended" stays pitch-side); leadership-decides; tool encoding synced (TSK-56).
Decision mechanics (pitch §7): Eric + Thom decide the fork. What would change the recommendation: a DevenMnA sizing deep-dive gate (TSK-48 verification of the agreed matrix completed 2026-08-19 — no cell moved, gate passed). The unchosen path pauses — resource reality, never ultimatum.
Reference basis — comparable v1 builds (Am.2 · provisional / from-memory benchmarks / confidence Medium)
DealRoomM&A deal room — ~1 yr, 2–4 people ≈ 1–3 py to credible v1
AffinityPE/deal CRM — ~2 yrs, 3–8 eng ≈ 3–8 py
Attio~3 yrs to GA
Basecamp 2004~0.5 py
Baremetricssolo — ~1 mo
Agency consensusB2B SaaS MVP 0.5–2 py; production-grade v1 2–9 py
Takeaway: at blended ~2.35× agentic leverage the plan deploys conventionally ~10 py of output — 2–4× what comparable v1 products needed — aggressive, not pessimistic.
input LMM vehicle — from the candidate matrix this pick replaces the "DD-as-micro-product" option in D1
The pick above is what lmm-candidate-matrix.html settled on. It flows into D1 below as the option under test. C1 + C2 = the Tier-1-live options under the LMM Vehicle Choice (BTDR-0010 Am.2 — see panel below); C3–C6 = Tier-2 validate-first (C5/DevenFiles rides the choice). Tip: open with ?pick=Cx on this URL to pre-set the pick; the sibling matrix chains here.
M14 · structure Diversify — two-tier recommendation choice + validate-then-build · BTDR-0010 Am.2
Tier 1 · LMM Vehicle Choice deferred — leadership-decides (Eric + Thom) · BTDR-0010 Am.2
C2 — DevenLite (Targets+Tasking) ~32w min-guess (LOW) · coupling-dominated · front-scoped · late-window if chosen · needs DevenDiligence + DevenFiles follow-ons
C1 — DD carve-out (DevenDiligence) ~3–5w (LOW) · Clarity gate retired
The plan commits to no vehicle and recommends none (the pitch's "DevenDiligence recommended" stays pitch-side). Capacity is branch-dependent — envelope unchanged (105.3 total / ~43.8 committed / ~61.5 headroom): choice → DevenLite ≈ ~47–53 of ~61.5 ("fits, barely"); choice → DevenDiligence ≈ ~18–26 incl. follow-ons; no choice = Tier-2 signal work only. BTDR-0010 Am.2
Tier 2 · validate then build need external signal — conditional
C4 — SDE Quick-Valuation priority demand-signal instrument
C3 — Valuation Enginedemand-signal instrument
C5 — DevenFiles (self-serve VDR)rides the choice — follow-on on the DevenLite branch
C6 — Sourcing pipelinedemand-signal instrument
Graduate to build only if a demand-signal instrument (fake-door/survey) confirms demand. Negative signal → defer to Beyond the Horizon (kill-condition). D3 targets C3/C4/C6; C5/DevenFiles rides the choice. grill Q13 · spec D3/D4 · BTDR-0010 Am.2
Phase 0 workstream Lost-lead census — systematic HubSpot closed-lost analysis. Re-anchored: Retain churn signal + preferred (non-required) demand input informing the vehicle choice — gates nothing. grill Q10 · spec D5 · BTDR-0010 Am.1/Am.2
context DevenConnect — tiered API pricing revenue model · amends BTDR-0007 · grill Q14
Retain Free for current customers — value-add; they already pay the seat license
Grow Consumption-based monthly for new clients — lower barrier than $75k; scales with usage
Diversify Consumption-based monthly (if needed) — LMM products are frontend-only on the API
Zapier integration (Phase 1) = the integration-enabler story — advertised and distributed once available. The connective tissue that makes the API sellable to non-developer buyers. grill Q14 · spec G4
context The pivotal bet (BTDR-0004) — hedge, sequenced the shape of the plan · not toggleable
Floor · table-stakes
Quick-win AI (Phase 1)
Inline ✨ text-assist on any field + document summarization. Broadest, lowest-risk "yes we have AI." devensoft-ai-integration-deck
Satisfy · prospect ask
Domain-value AI (Phase 2)
Chat-with-doc · DD doc Q&A · Ask-the-platform (NL Q&A + citations). The "integrations + AI" prospect ask. devensoft-ai-integration-deck
Win · differentiation
Integration / PMI / seam-AI
DevenMCP at our center-of-gravity seam — greenfield; no rival ships it competitor-feature-parity-matrix. Roadmap, not in-horizon.
BYOM = the build, not a routing layer. We build the inference-dependent features (the surfaces above). The customer configures their LLM provider — base URL + API key — so features run on their own models + their own compliance posture. Devensoft routes calls; it never hosts models → stays out of the sub-processor compliance path. devensoft-ai-integration-deck (deck calls this BYOK)
Model-governance rule (risk-tiered): a hosted "general LLM" exists only in the LMM standalone micro-product (lower legal/MNPI stakes). At enterprise, BYOM is mandated — the customer's own model serves all platform AI. BTDR-0004 G1
Deliberately NOT built: full governance moat (→BYOM) · non-demand feature-AI · hosted/native LLM at enterprise · commodity enterprise diligence-AI · splitoff in the near term. BTDR-0004 G3  ·  Demand reality: 0/108 direct AI demand youtrack-demand-sweep; #1 pain = access/reliability, #1 demand = reporting hubspot.
factor model Capacity inputs all editable — tune the model to test paths
Engineering FTEs
1.5 FTE — existing team (Roman + Diego), no onboarding drag
available headcount · hard floor 1.5 (existing team) · cap 17.5
Window (weeks)
~9 months ≈ 39w
Onboarding drag
% lost to contractor transition / teaching — applies to FTE above the 1.5 existing-team baseline
Productivity drag past 3 FTE
pizza-team: each FTE > 3 loses ×this productivity (higher = more drag → less capacity)
Agentic coding multiplier
work completes this × faster (1.3 = scope ÷ 1.3; 2.0 = half the dev-weeks)
Non-Engineering Overhead
hours/day subtracted from each FTE's 8-hr day (default 2 → 6 hrs/day, 0.75× engineering time)
Effective capacity
= eff-FTE × weeks × eng-time · dev-weeks (drag baked into eff-FTE tiers)
Capacity is the honest constraint. The agentic multiplier is the lever we control. Per-feature dev-week estimates are evidence-grounded where possible; those marked assumption are editable planning placeholders. The vehicle row's dev-weeks stay editable — size them to the chosen candidate (carve-outs cheaper, net-new costlier; per-candidate sizing not yet evidence-grounded).
path variables Three decisions to recommend flip a toggle → readout re-shapes live
D1 — LMM vehicle (the LMM Vehicle Choice)
Tier-1 = the LMM Vehicle Choice — deferred; Eric + Thom decide (no vehicle committed, none recommended; BTDR-0010 Am.2). The pick below is the option under test, not a commitment. DevenLite rides as candidate C2 from the matrix (static toggle removed — TSK-51). BTDR-0010 Am.2 · BTDR-0004 G2 · pick from candidate matrix
D2 — AI feature scope (BYOM build)
How much of the deck do we commit to in-window?
D3 — Iron triangle
If scope exceeds capacity, which constraint gives? 9-mo horizon
Capacity vs committed scope (dev-weeks)
"We have AI" posture · risk-appetite readout
    Recommended Path (what we'd propose to Eric)
    Preset scenarios — stress-test the 9-month window
    Evidence chips: devensoft-ai-integration-deck AI surfaces; phase estimates scaled ×1.42 per TSK-32 → 13/13/18w · tsk30 Fundamentals F1–F6 (32–53w; mid 43) · tsk33 F7 reporting-data cluster (~5.5w) · tsk32 BYOM corrected sizing (35–54w, mid ~44) · youtrack-demand-sweep 0/108 AI demand · hubspot #1 pain access/reliability; #1 demand reporting · competitor-feature-parity-matrix AI commodity-in-diligence, greenfield-in-integration · grill Q13 two-tier Diversify · BTDR-0010 Am.2 Tier-1 = LMM Vehicle Choice (deferred — Eric + Thom) · grill Q14 tiered API pricing · spec D5 lost-lead census (Phase 0) · BTDR-0004 pivotal bet · assumption editable planning placeholder (Vehicle 10w · Power-BI-MCP 6w).