Start with the decision
A feature request is not a problem statement. Identify who decides what, using which evidence, under what deadline, and at what cost if wrong.
OPEN FIELD MANUAL
Principles for work that survives contact.
A concise operating reference for turning technical capability into adopted, measurable outcomes.
01 / PRINCIPLES
Twelve rules for durable deployment.
These principles apply across forward deployed engineering, production AI, solutions engineering, consulting, and implementation.
Map the current workaround
Spreadsheets, chat messages, private scripts, and expert memory contain the real operating logic. Treat them as evidence, not embarrassment.
Separate evidence from inference
Label what is observed, calculated, assumed, predicted, and recommended. Trust collapses when these categories blur.
Ship the thinnest credible slice
Narrow the scope, not the completeness. A slice still needs data, logic, interface, permissions, support, and a measurable decision.
Design the failure states first
Stale data, missing identity, model uncertainty, integration failure, and user disagreement need visible behavior—not silent fallback.
Keep humans in authority where risk demands it
Human review must have clear inputs, time, expertise, escalation, and accountability. A button labeled approve is not a control system.
Treat adoption as system design
Trust, incentives, workflow fit, training, ownership, and exceptions shape real performance as much as code.
Instrument use before claiming value
Measure whether people use the intervention, where they override it, how work changes, and whether the intended outcome moves.
Do not automate unresolved policy
Software will make ambiguity faster and harder to see. Resolve ownership and decision rules before encoding them.
Write for the next operator
Architecture diagrams, runbooks, data contracts, and decision records should let someone else operate and change the system safely.
Make rollback a product feature
A rollback plan includes data, workflow, communications, ownership, and reconciliation—not only a previous container image.
Scale only after the mechanism is understood
A positive pilot is evidence, not proof. Know what created the result, what conditions it depends on, and what changes at scale.
02 / WORKING ARTIFACTS
Eight documents that keep the work coherent.
The templates are intentionally short. Their job is to expose weak reasoning and preserve decisions, not generate documentation volume.
Discovery record
Capture the decision, workflow, stakeholders, baseline, assumptions, constraints, evidence, and open questions.
Decision map
Show the actor, trigger, inputs, decision rule, action, feedback, and downstream consequence.
Data contract
Document source, owner, freshness, identifiers, validity rules, access, lineage, and failure behavior.
Solution thesis
State how the intervention changes the decision, why it should work, and what must be validated.
Risk and control register
Connect each failure mode to prevention, detection, response, owner, and evidence.
Pilot charter
Define users, scope, duration, training, support, metrics, gates, rollback, and decision authority.
Operating runbook
Explain normal operation, alerts, exceptions, escalation, recovery, ownership, and changes.
Value proof
Compare baseline and observed outcomes, adoption, costs, confounders, and the scale/revise/stop recommendation.
03 / USE
Use the manual inside real work.
Choose one live project. Complete the discovery record before redesigning the solution. Revisit the decision map at every scope change. Use the value proof to decide whether the intervention deserves more investment.
START WITH A MISSION
Apply the principles under controlled pressure.
Atlas Freight provides a complete task sequence and staged complication.