feat(rules): slot-grouped brain injection + graduation-time classifier

[Gradata]

Summary

Lays the groundwork for wiring prompt_synthesizer into the session-start injection path (the legacy inject_brain_rules.py N-rules block will be replaced in a follow-up PR).

• synthesize_brain_injection(lessons, *, budget_tokens, max_per_slot, persona_baseline) — new slot-grouped entry point that orders rules by Preston Rhodes' 6-step checklist (task → context → examples → persona → format → tone), enforces a 400-token default budget, and preserves inline r:xxxx anchors for capture_learning.py attribution.
• classify_slot(item) — resolves explicit slot → example pair → category lookup → context fallback. Exposed at import time so graduation can tag lessons once, at promotion time.
• Lesson.slot — new optional dataclass field (default ""). Format/parse round-trip through lessons.md as Slot: <value>.
• Graduation: _ensure_slot(lesson) runs once per lesson at INSTINCT→PATTERN and PATTERN→RULE promotion. Idempotent; classifier errors are swallowed (slot is optional metadata).

Legacy synthesize_rules_prompt is untouched for back-compat. All existing tests remain green.

Test plan

• tests/test_prompt_synthesizer.py — 29 tests: classify_slot resolution order, Preston-Rhodes slot ordering, persona baseline (str vs Path), budget-drops-lowest-priority-first, examples-slot promotion on example_draft, Lesson-like object support.
• tests/test_slot_graduation.py — 7 tests: default slot, format omits empty, format+parse round-trip, INSTINCT→PATTERN assigns slot, existing slot preserved, no-promotion leaves slot unset.
• Full suite: 3931 pass, 2 skip (pytest tests/).

Generated with Gradata

[greptile-apps]

Your free trial has ended. If you'd like to continue receiving code reviews, you can add a payment method here.

[coderabbitai]

Caution

Review failed

The pull request is closed.

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: ASSERTIVE

Plan: Pro

Run ID: 08b5c29b-58fb-498d-8cb5-9dd163e6c14c

📥 Commits

Reviewing files that changed from the base of the PR and between ff8c053 and c1575b5.

📒 Files selected for processing (6)

• Gradata/src/gradata/_types.py
• Gradata/src/gradata/enhancements/prompt_synthesizer.py
• Gradata/src/gradata/enhancements/self_improvement/_confidence.py
• Gradata/src/gradata/enhancements/self_improvement/_graduation.py
• Gradata/tests/test_prompt_synthesizer.py
• Gradata/tests/test_slot_graduation.py

---

📝 Walkthrough

Summary

• New Lesson.slot field: Added optional string field (default "") to store Preston-Rhodes synthesis slot identifier; serialized/parsed through lessons.md as Slot: <value> metadata
• New public API synthesize_brain_injection(): Slot-grouped brain-injection entry point implementing Preston-Rhodes 6-step ordering (task → context → examples → persona → format → tone) with configurable token budget (default 400, env var GRADATA_SYNTH_BUDGET) and per-slot caps
• New public API classify_slot(): Slot classifier using explicit slot → example pair → category mapping → context fallback; exposed at import time for graduation-time assignment
• Graduation-time slot assignment: _ensure_slot() idempotently assigns slots during INSTINCT→PATTERN and PATTERN→RULE promotions; swallows classifier errors (slot is optional metadata)
• Persona baseline support: synthesize_brain_injection() accepts optional persona baseline (raw string or Markdown file path) with "Overrides" clause support
• Backward compatibility preserved: Legacy synthesize_rules_prompt() unchanged; all existing tests green (3931 pass, 2 skip)
• Exports added: DEFAULT_BUDGET_TOKENS, SLOT_ORDER, classify_slot, synthesize_brain_injection
• Markdown round-trip: parse_lessons() and format_lessons() extended to handle optional Slot: metadata lines

Walkthrough

The PR introduces Preston-Rhodes slot-based lesson synthesis and classification. A new slot field is added to lessons, slot classification logic is implemented with fallback precedence rules, a slot-ordered prompt synthesizer with token budgeting is created, the graduation pipeline is extended to auto-assign slots via classification, and markdown parsing/formatting are updated to handle slot metadata.

Changes

Cohort / File(s)	Summary
Core Type Definition Gradata/src/gradata/_types.py	Added slot: str = "" field to Lesson class to store Preston-Rhodes synthesis slot identifier, intended for assignment at graduation.
Slot Synthesis & Classification Gradata/src/gradata/enhancements/prompt_synthesizer.py	Introduced classify_slot() for 6-step Preston Rhodes slot inference (explicit slot → examples → category mapping → context fallback) and new public synthesize_brain_injection() function that renders slot-grouped prompts with token budgeting, persona baseline support, and per-slot caps. Module documentation updated to identify brain-injection as canonical session-start generator.
Markdown Serialization Gradata/src/gradata/enhancements/self_improvement/_confidence.py	Extended parse_lessons() to parse case-insensitive Slot: metadata lines and format_lessons() to emit Slot: lines when slot is non-empty, enabling round-trip preservation of slot metadata in markdown.
Graduation Integration Gradata/src/gradata/enhancements/self_improvement/_graduation.py	Added _ensure_slot() to attempt best-effort slot assignment via classify_slot() during lesson promotion, hooked after rule refinement (ToT) and before state transitions, with silent failure on classification errors.
Test Coverage Gradata/tests/test_prompt_synthesizer.py, Gradata/tests/test_slot_graduation.py	Added tests validating slot classification precedence, brain-injection synthesis ordering/budgeting/persona handling, empty input handling, per-slot caps, and round-trip markdown serialization. New test module verifies slot assignment during graduation transitions and preservation of existing slots.

Sequence Diagram(s)

sequenceDiagram
    participant User as Caller
    participant Synth as synthesize_brain_injection()
    participant Classify as classify_slot()
    participant Render as Slot Renderer
    participant Budget as Token Budgeter

    User->>Synth: lessons[], budget_tokens
    loop For each lesson
        Synth->>Classify: lesson
        Classify-->>Synth: slot (explicit→examples→category→context)
    end
    Synth->>Render: slot→sentence groups (SLOT_ORDER)
    Render-->>Synth: formatted sentences
    Synth->>Budget: sentences, budget_tokens
    Budget-->>Synth: truncated sentences (drop lowest-priority)
    Synth-->>User: SynthesizedPrompt (text, anchors_used)

Loading

sequenceDiagram
    participant Grad as graduate()
    participant Lesson as Lesson (INSTINCT→PATTERN)
    participant Ensure as _ensure_slot()
    participant Classify as classify_slot()

    Grad->>Lesson: refine rule wording (ToT)
    Grad->>Ensure: lesson
    alt slot empty?
        Ensure->>Classify: lesson
        Classify-->>Ensure: inferred slot
        Ensure->>Lesson: assign slot
    else slot already set
        Ensure-->>Grad: (no-op)
    end
    note over Ensure: failures swallowed
    Grad->>Lesson: promote INSTINCT→PATTERN
    Grad-->>Grad: complete

Loading

Estimated Code Review Effort

🎯 4 (Complex) | ⏱️ ~75 minutes

Possibly Related PRs

• feat: S102 — MiroFish P0-P2 roadmap implementation (9 features) #24: Modifies Lesson dataclass with additional metadata fields (alpha/beta), sharing the same target class as this PR's slot field addition.
• feat: Sim 9 engine hardening — rule scoping, convergence gate, efficiency metric #15: Adds domain_scores metadata field to Lesson, creating a related pattern of lesson metadata extension.

Suggested Labels

feature

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch feat/synthesizer-wire

---

_{Comment @coderabbitai help to get the list of available commands and usage tips.}