tegwick
82becd0fda
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_014KmVxhJ35tCo7rE7UnLwWu Assistant: claude-code Assistant-Model: opus Assistant-Process: 1116572@bnt-lap001 Assistant-Session: 8ba9bb93-a72a-4883-b189-2499cce5c400
125 lines
2.8 KiB
Markdown
125 lines
2.8 KiB
Markdown
---
|
|
id: FLUID-WP-0006
|
|
type: workplan
|
|
title: "FLUID Science - the adaptation loop closes (Blueprint Phase C)"
|
|
domain: infotech
|
|
repo: fluid-core
|
|
status: active
|
|
owner: worsch
|
|
topic_slug: fluid-core
|
|
created: "2026-09-04"
|
|
updated: "2026-09-04"
|
|
planning_priority: high
|
|
planning_order: 5
|
|
depends_on:
|
|
- FLUID-WP-0005
|
|
---
|
|
|
|
# FLUID-WP-0006 - FLUID Science
|
|
|
|
Hypotheses, experiments, fitness and promotion. Entirely human-driven — the
|
|
Daimon is deliberately absent. Blueprint §50 says the success criterion is
|
|
proving this loop works cleanly and safely, not autonomous coding.
|
|
|
|
## T01 - Hypothesis records and lifecycle
|
|
|
|
```task
|
|
id: FLUID-WP-0006-T01
|
|
status: todo
|
|
priority: high
|
|
```
|
|
|
|
Schema doc §2: DRAFT, READY, PRIORITIZED, DESIGNING, EXPERIMENTING, EVALUATING,
|
|
then ACCEPTED, REJECTED, SUPERSEDED or DEFERRED. Keep observation, explanation,
|
|
prediction, intervention and result separate — schema doc §18 warns that
|
|
collapsing them into one narrative destroys criticism and auditability.
|
|
|
|
## T02 - Competition groups
|
|
|
|
```task
|
|
id: FLUID-WP-0006-T02
|
|
status: todo
|
|
priority: medium
|
|
```
|
|
|
|
Rival explanations of one pressure coexist (schema doc §6). Preserving
|
|
uncertainty is a feature.
|
|
|
|
## T03 - Hypothesis API
|
|
|
|
```task
|
|
id: FLUID-WP-0006-T03
|
|
status: todo
|
|
priority: medium
|
|
```
|
|
|
|
Blueprint §44.3: create, compare, prioritize, attach candidate, record outcome.
|
|
|
|
## T04 - Experiment controller
|
|
|
|
```task
|
|
id: FLUID-WP-0006-T04
|
|
status: todo
|
|
priority: high
|
|
```
|
|
|
|
Blueprint §16. Declares hypothesis, control, candidate, cohorts, allocation,
|
|
metrics, guardrails, start and stop conditions, maximum duration. Per §17 it
|
|
must not touch traffic directly: it writes routing policy the router consumes.
|
|
Experiments are interruptible.
|
|
|
|
## T05 - Experiment API
|
|
|
|
```task
|
|
id: FLUID-WP-0006-T05
|
|
status: todo
|
|
priority: medium
|
|
```
|
|
|
|
Blueprint §44.4: create, start, change bounded allocation, stop, record metrics,
|
|
finalize.
|
|
|
|
## T06 - Fitness evaluation against hypotheses
|
|
|
|
```task
|
|
id: FLUID-WP-0006-T06
|
|
status: todo
|
|
priority: high
|
|
```
|
|
|
|
Wire the fitness engine to experiments. Success criteria may not be changed
|
|
after results are visible without recording the amendment (§18).
|
|
|
|
## T07 - Promotion controller
|
|
|
|
```task
|
|
id: FLUID-WP-0006-T07
|
|
status: todo
|
|
priority: high
|
|
```
|
|
|
|
PROMOTE, EXPAND_EXPERIMENT, RETAIN_AS_OPTION, MODIFY, REVERT, ABANDON, DEFER
|
|
recorded as decisions with an authorizing actor (§19). Human authority only.
|
|
|
|
## T08 - Append-only event log
|
|
|
|
```task
|
|
id: FLUID-WP-0006-T08
|
|
status: todo
|
|
priority: high
|
|
```
|
|
|
|
Every transition above as a `fluid_event` (schema doc §15). FLUID evolution is
|
|
itself a system behavior that must remain reconstructable.
|
|
|
|
## T09 - Audit trace command
|
|
|
|
```task
|
|
id: FLUID-WP-0006-T09
|
|
status: todo
|
|
priority: high
|
|
```
|
|
|
|
`fluid audit trace <revision>` reconstructs pressure, hypothesis, revision,
|
|
experiment, fitness and decision, answering all eleven questions in API
|
|
Standards §25.
|