SFT Part III

The SFT Core

Part III defines the theorem-bearing fragment: `SoftwareField`, artifact-mediated change, operation support, policy, governance interventions, `ForecastCone`, `ConsequenceEnvelope`, and closed-loop field update.

SoftwareField and architecture projection

`SoftwareField` is a computable slice of a development field. It is not itself an AAT `ArchitectureObject`; the AAT object is obtained through architecture projection.

Field projection
arch : SoftwareField -> ArchitectureObject
Computable core
ComputableCore
  = SoftwareField
  + arch projection
  + ArtifactMediatedChange
  + OperationSupport
  + OperationPolicy
  + StepRelation
  + ObservationRecord
  + GovernanceIntervention
  + ForecastCone
  + FieldUpdate
  • Contextual state History, constraints, observations, governance, and exogenous artifact inputs belong to the field.
  • Architecture projection AAT reasoning applies to the projected architecture object, not to every field component at once.

Artifact-mediated change

The formal term is `ArtifactAction` or `ArtifactMediatedChange`. Force remains an informal field-level reading. PRDs, specs, issues, and AI proposals can produce multiple candidate updates rather than a single next field.

Set-valued action
ArtifactAction a
  := alpha_a : FieldSpace -> Set FieldSpaceUpdateHypothesis

Forecasted effects are outputs of forecasting, not part of the raw action definition. This prevents an artifact from already containing the forecast that it is supposed to generate.

  • Raw action Source artifact, target components, interpretation boundary, candidate updates, action boundary, and observable boundary.
  • Force descriptor Action class, target regions, candidate support families, affected policy or governance components, and uncertainty boundary.
  • Observed effect Transition, before / after signatures, comparable deltas, unexpected witnesses, and review / CI / runtime outcome.

Operation support and policy

SFT asks which operations are lawful, but also which operations appear natural, possible, low cost, or likely to be selected in a field.

  • OperationSupport The admissible operation set after constraints and governance interventions.
  • OperationPolicy A preorder, cost, selection relation, or explicitly probabilistic distribution on supported operations.
  • Attractor engineering Support and policy shaping, not a hidden physical force or automatic stability theorem.

ForecastCone and ConsequenceEnvelope

`ForecastCone` is the formal set of reachable field paths from a starting field, support relation, and horizon. For set-valued artifact actions, SFT forms a family of cones and summarizes them for practice as a `ConsequenceEnvelope`.

ConsequenceEnvelope
ConsequenceEnvelope
  = selected ForecastCones
  + reachable path classes
  + affected architecture regions
  + comparable signature axes
  + obstruction witness candidates
  + missing invariants / boundaries
  + forecast boundary
  + unknown / unmodeled remainder
Action-indexed cone family
u in alpha_a(F)
F_u := applyUpdate(F, u)
U_u := U(F_u)

ForecastConeFamilyAfterAction(F, a, h)
  = { ForecastCone(F_u, U_u, h) | u in alpha_a(F) }

Observation, governance, and FieldUpdate

Review, CI, type checking, architecture rules, AI policy, and runtime guards are governance interventions because they can change support, policy, observation, and feedback update.

`FieldUpdate` records the difference between forecast and observation: signature delta, unexpected witnesses, review or CI outcome, runtime feedback, missing evidence, and non-conclusions.

  • Restrictive intervention Removes unsafe support from later accepted operations.
  • Redirective intervention Raises shortcut cost and lowers lawful path cost inside the selected field model.
  • Instrumenting intervention Adds observation axes, tests, runtime checks, or evidence records.
  • Learning intervention Preserves observed outcomes and forecast error in posterior field memory.

Field update preserves forecast error and boundary evidence. It does not automatically prove that future forecasts become more accurate.

Core boundary and non-conclusions

Part III defines the theorem-bearing fragment, not the whole social reality of a development organization. `ForecastCone` is a reachable path set under selected support and horizon; `ConsequenceEnvelope` is the reader-facing report that explains the cone family, unknown remainder, and forecast boundary.

The core does not assert that artifacts have unique effects, that policy is probabilistic unless a distribution is supplied, or that a field update automatically improves later forecast quality.