Organization-wide practice / reusable across partnersPhase 1 · proposed global mandate · owner approval pending
Forward-deployed engineering
Make payment infrastructure Agentic Ready.
Decentra’s specialized FDE practice makes payment infrastructure Agentic Ready through bounded integration, validation and operational transfer. The customer outcome stands on its own; rights-cleared field learning strengthens Decentra’s proprietary platform. Every company inherits the same gates, never another company’s commitments.
LANE A / FDE DELIVERYIntegrate. Validate. Transfer.
One scoped payment journey, independently accepted by the customer.
LANE B / DECENTRA PLATFORMProductize what we may reuse.
Separate rights, product approval and commercial evidence; no automatic adoption.
Partner reality determines the integration. Decentra’s global gates determine how we work.Global core ≠ company deal
01
Mandate and boundary
Payments-only FDE enables agentic commerce at payment-infrastructure boundaries; it is not generic AI consulting.
We do
Bounded, embedded delivery
Translate a named partner outcome into one testable journey and an agreed integration boundary.
Serve processors, gateways, acquirers, merchant infrastructure and ecosystems where work advances agentic-payment readiness. Decline unrelated AI or software engineering work.
Work with the partner’s APIs, controls, people and operating constraints.
Build or adapt the smallest useful primitive; validate adverse as well as happy paths.
Document ownership, evidence and the reusable learning before handoff.
We do not
Unbounded or implied commitments
Treat a demo, protocol adapter or patent thesis as production certification.
Take custody, regulatory obligations or irreversible payment authority by assumption.
Let one partner silently define the global product or another partner’s contract.
Equate successful FDE delivery with platform adoption, a signed license, production transaction volume or downstream merchant activation.
Call a pilot successful because the slide or UI works while recovery and operations do not.
Two ledgers, always. Track partner-specific delivery separately from reusable platform/IP work. An integration may consume a Decentra primitive; the partner does not adopt the whole technology stack unless that is expressly scoped and validated. FDE deliverables and acceptance stand on their own; licensing, usage and distribution each require separate evidence.
Agentic Readiness is scoped, not a company-wide badge. For each selected workflow capability—agent identity, delegated authority, transaction authorization, spend policy, payment interoperability, risk, audit evidence, recovery or operations—record not assessed, assessed, implemented, validated or production-ready, with environment, owner and proof. Not every engagement needs every capability. Production-ready requires separate release and operations approval.
IP has an explicit rights gate. Identify Decentra background IP, customer background IP and data, engagement foreground IP, generalized inventions and possible residual know-how. Flag candidates promptly; no cross-customer reuse of customer-specific work until contract and confidentiality review, security clearance and product ownership approval. A reusable idea is not a reuse license.
02
Qualify before embedding
An interesting account is not an FDE engagement until the inputs and authority exist.
Named problem and sponsor
What specific adoption or transaction journey is blocked? Who owns the outcome and can decide?
Accessible boundary
Is there a real system owner, test environment, contract/API view and lawful data path?
Evidence-shaped success
Can both sides state acceptance tests, failure modes and an observable baseline?
Economic and reusable case
Is the effort funded or intentionally exploratory, with a bounded timebox and a learning that could recur?
Payments boundary
Who owns the payment instruction, funds, merchant order, risk decision and reconciliation? Which rail, data class, jurisdiction and network rules may apply?
Product and distribution fit
Could the work yield a rights-cleared reusable integration or channel opportunity? Mark both as hypotheses until demonstrated or contracted.
If any item is unknown, run discovery rather than calling the work an implementation. “No-go” and “not yet” are valid decisions.
03
Seven stages. One gate at a time.
Each stage has an owner, a tangible artifact and an exit decision. Open a stage to see its minimum bar.
Account and sponsor, problem statement, candidate journey and known systems.
Reason to involve FDE rather than ordinary sales, product support or a generic SDK; confirm direct payments-mission fit.
Unknowns, immediate blockers and discovery timebox.
Exit: accept discovery, defer or decline. Partner owner and FDE lead record the decision and next owner. No engineering commitment is inferred from a meeting.
01Discover · map reality and baseline
Produce
Current journey, actors, trust boundaries, systems of record and integration surface.
Access and data inventory; legal/security and regulated/network questions; baseline pain and failure paths.
Initial Agentic Readiness status for only the capabilities relevant to this journey, with environment and evidence.
Options: configure existing feature, adapt primitive, build adapter, or do nothing.
Exit: one narrow opportunity is technically credible. Technical lead and partner system owner confirm facts versus assumptions; unknown critical permissions block charter approval.
02Charter · authorize a bounded engagement
Produce
One journey, in/out-of-scope, parties, DRI, timebox, commercial basis and change route.
Acceptance tests including denial, replay, timeout, partial success and manual recovery.
Data/security obligations, customer-approved environment and tools, named personnel, test access path and stop conditions.
Applicable signed SOW and relationship terms; counsel decides whether a DPA, Security Addendum, AI Addendum or Payments Regulatory Flow-Down Schedule is required. Proposed documents are not assumed executed.
Exit: sponsor, delivery lead and required risk owners sign off. If price, staffing, access or liability is unresolved, label it explicitly and do not start committed implementation.
03Design · freeze interfaces and control placement
Produce
Sequence diagram, request/response and events, idempotency and reconciliation plan.
ADR for bespoke versus potentially reusable code, with background/foreground provenance; test plan, telemetry and rollback.
Map human authorization, data flows (including prompts/logs), security tests and incident route before sensitive access or live effects.
Exit: reviewable design and no unresolved critical control gap. Partner technical owner signs the boundary; security/compliance approves relevant risk before sensitive access or live effects.
04Deliver · implement behind a controlled boundary
Produce
Versioned code/config, integration and security tests and traceable changes; synthetic data first, customer information only in its approved environment.
Record access roster, approved AI tools and potential reusable inventions as they arise; material scope or data changes return to Charter.
Observability for every irreversible action and a documented pause/rollback path.
Updated decision log whenever actual partner behavior differs from the design.
Exit: test environment demonstrates the scoped journey. A demo is not acceptance. Scope change goes back through charter and control review.
Linked traces, payment/order status reconciliation, residual risks and incident exercise.
Explicit pass/fail against charter criteria and documented exceptions; use-case-specific AI metrics, datasets and human review only where relevant.
Record validated capabilities separately from the later production release decision and any platform license.
Exit: partner owner accepts the named scope, or it remains a pilot. Production change requires separate operational, security, legal and commercial approval where applicable.
06Transfer · operate, learn, productize
Produce
Runbook, support owner, monitoring, incident route, version/maintenance obligations and SOW-scoped training so customer operators can run the agreed deliverable.
Reusable primitive candidate with invention log, contractual rights review, confidentiality screen, product owner, generalization test and known limits; keep restricted forks separate.
Retrospective: actual effort, partner outcome, failure patterns and next-company applicability.
Exit: operations accepts ownership and reusable learning is recorded. A partner fork stays labeled a fork until a product owner accepts it into the common core.
04
Decision rights, not title theater
Name actual people per engagement. One DRI per decision; reviewers cannot silently substitute for an approver.
Decision
Accountable role
Required input / record
Priority & sponsor
Partner / commercial owner
Partner outcome, economic basis, internal priority and sponsor confirmation.
Engagement scope
FDE engagement lead
Charter, staffing, timebox, acceptance, in/out-of-scope and change log.
Technical boundary
FDE technical lead + partner system owner
Interfaces, system of record, controls, observability and ADR.
Reusable product
Decentra product owner
Generalization criteria, maintenance owner, version policy and tests.
Security / privacy
Designated risk owner(s)
Threat model, data inventory, access/custody review and exceptions.
Operational acceptance
Partner service owner + Decentra support owner
Runbook, incident path, SLOs if agreed, on-call and handoff record.
Contract and IP rights
Authorized signatory + counsel / IP owner
Executed terms, scope, data and foreground allocation; artifact-specific reuse clearance.
Regulatory and network applicability
Counsel + security + customer counterpart
Entity, activity, data and network classification; approved obligations and distinct notice triggers.
Separate release evidence; verify any partner, patent, certification, license or revenue claim before use.
A small team may hold multiple roles, but the decision and its accountable signer remain explicit. Escalate disagreements to a named executive sponsor before widening scope or assuming risk.
05
Non-negotiable guardrails
These apply before a company-specific overlay, subject to stricter legal and partner requirements.
Authority and data
Protect the boundary
Use least-privilege, revocable partner access. No shared production secrets in notes, repos or demos.
Define who may authorize what; an agent recommendation is not buyer consent or a payment instruction by itself.
Prefer synthetic data and sandbox rails; production data and live funds require explicit approval and control review. Customer information remains in customer-controlled or expressly approved environments; production-data approval is not permission to export it to Decentra systems, personal devices or external AI tools.
Record where credentials, payment tokens and personal data reside, transit and expire.
Delivery and claims
Protect the truth
Keep decision logs, test evidence and versioned artifacts tied to the scoped environment.
Do not conflate protocol compatibility, local demo, pilot, certified integration and production availability.
Do not announce partner, network, IP, compliance or commercial status without the owner’s supporting evidence and approval.
Stop and re-charter on unapproved irreversible effects, new data classes or material control changes.
Keep a named access roster; obtain internal and any contractual customer consent before other teams or subcontractors touch client work. Preserve evidence and report suspected incidents immediately to security and engagement leadership; a within-one-hour internal objective requires owner approval. External notice follows the applicable contract and counsel-reviewed rule, not a universal clock.
Exception rule. An exception names the violated control, risk, mitigation, approver, expiry and rollback. No “temporary” exception without a recorded owner and review date.
06
Minimum evidence pack
The handoff is a set of inspectable artifacts, not a retrospective claim of readiness.
Artifact
Minimum contents
When
Opportunity record
Problem, sponsor, candidate journey, qualification and no-go conditions.
Intake
Discovery map
Actors, flow, APIs, control boundaries, assumptions, baseline and options.
Discovery
Engagement charter
Scope, acceptance tests, owners, timebox, commercials, environment and changes.
Before build
Architecture / ADR
Interface contracts, authority/custody, failure modes, integration placement and reuse choice.
Before build
Risk & access record
Threat/data review, approvals, exceptions, credentials inventory and revocation.
Before sensitive access
Acceptance dossier
Test cases, linked traces, failures, recovery, partner decision and residual risks.
Acceptance
Runbook & reuse note
Operational owners, alerts, incident path, known limits, maintenance and productization decision.
Transfer
Contract & access matrix
Effective signed terms, obligations, named team, approved tools, data flow and any regulatory/network applicability decision.
Before build or sensitive access
Readiness & rights register
Selected capability/status/environment/evidence; IP provenance, invention flag and legal/security/product reuse disposition.
Design through Transfer
Commercial evidence ledger
FDE acceptance apart from executed platform license, realized usage and measurable merchant activation; model outputs are not results.
Acceptance and review
Metrics: time from intake to charter and to validated readiness; time to first reproducible test; acceptance pass/fail; approved production releases; recovery performance; handoff and customer self-sufficiency; escaped incidents and support load. Track rights-cleared reuse and reduced repeat effort separately from verified platform licensing, usage and merchant enablement. Set denominators and baselines before claims. Do not convert the draft financial model into targets.
07
How each company inherits the core
A separate company Foldy narrows this standard. It does not edit the global rules in place.
Company overlay / required fields
Duplicate this structure for Payroc later, and for every subsequent company. Keep the global standard linked and mark local facts, assumptions and approvals distinctly.
Company, sponsor, owners and decision authority
Specific business outcome and one first journey
Current-state systems and system-of-record map
Integration surfaces, credentials and data permissions
Scope, exclusions, timebox and commercial status
Partner-specific security, legal and operational constraints
Acceptance evidence, failure tests and go/no-go decision
Selected readiness dimensions and evidence; separate FDE, production and optional platform/commercial milestones
Background/foreground rights, potential reuse clearance and downstream distribution assumptions
Inheritance rule. Global principles are the default; a company overlay may add constraints or request a documented exception, but cannot silently delete a global gate. Changes useful across multiple companies are proposed back to the global standard with an owner, evidence and version increment.
08
Adopt, review, revise
This is a proposed organization-wide standard, not evidence that every team has already approved or followed it.
To activate v1
Assign the stewards
Name an executive sponsor, FDE practice owner, product owner and security/compliance reviewer. Agree who can approve charters, risk exceptions and production transitions. Ratify the evidence pack and store it in an accessible system of record.
Operating cadence
Review the pattern, not just the deal
Review active engagement gates and blockers regularly; inspect acceptance and incidents at handoff; revisit the global standard after early partner learnings. Record a dated version and change rationale. Future company Foldys cite the version they inherit.
Source basis: v0.1 and the private Phase 1 source review of the FDE playbook, September funding report, 14-tab draft pro forma and available product context. The legal positions require counsel; the High-case model and pitch assertions are not approved targets or proof of partner commitments. Executive, legal, security and finance owners must approve their respective provisions before adoption. This Foldy remains proposed and unpublished.