Architecture — ACRLDS

Home › Architecture

Pipeline Flow

S31 · 8 gates · curriculum_context KB-join enrichment, kb_write loopback, pacing guardrails · Click any node for schemas, prompts, contracts

User/LMS Preflight LLM Skill Tool Soft Gate Hard Gate Render/Finalize | JSON file KB table Domain rule Rubric Prompt Field contract

Cross-lane data bridges

Gate Registry

Human review checkpoints. Hard gates block downstream; soft gates auto-pass if un-reviewed.

CHAPTER_MODULE_SEQUENCE hard

Hard gate — blocks downstream pipeline until approved. Rework triggers re-run of the producing skill with human feedback.

produced by derive_chapter_sequence ch_research

CHAPTER_LO_BRIEF soft

Soft gate — informational review. Pipeline proceeds if skipped. Rework triggers re-run of the producing skill with human feedback.

produced by derive_chapter_sequence ch_research

CHAPTER_KB_SYNTHESIS soft

Soft gate — informational review. Pipeline proceeds if skipped. Rework triggers re-run of the producing skill with human feedback.

produced by research_deep ch_research

MODULE_LO_BRIEF hard

Hard gate — blocks downstream pipeline until approved. Rework triggers re-run of the producing skill with human feedback.

produced by derive_module_lo mod_review

MODULE_KB_SYNTHESIS soft

Soft gate — informational review. Pipeline proceeds if skipped. Rework triggers re-run of the producing skill with human feedback.

produced by research_deep mod_review

REFERENCE_MODULE_BLUEPRINT soft

Soft gate — informational review. Pipeline proceeds if skipped. Rework triggers re-run of the producing skill with human feedback.

produced by build_blueprint mod_review

APPLET_KB_SYNTHESIS soft

Soft gate — informational review. Pipeline proceeds if skipped. Rework triggers re-run of the producing skill with human feedback.

produced by research_deep app_review

REFERENCE_APPLET_SCREENS hard

Hard gate — blocks downstream pipeline until approved. Rework triggers re-run of the producing skill with human feedback.

produced by build_blueprint app_review

Rubrics

Scoring criteria loaded live from config/rubric/*.json. Click to see the criteria table.

applet_D Applet

Instructional Design rubric

7 criteria 25 pts Instructional Design applet_D.json

applet_L Applet

Accessibility rubric

9 criteria 25 pts Accessibility applet_L.json

applet_P Applet

Pedagogy rubric

7 criteria 25 pts Pedagogy applet_P.json

applet_X Applet

Experience rubric

8 criteria 25 pts Experience applet_X.json

applet_gates Applet

Scoring / reference file.

applet_gates.json

blueprint_guide Guide

System-specific rules for S3 build_blueprint. Operationalises module and applet structure. The LLM already knows CPA, cognitive load, ZPD — these rules are specific to this system.

blueprint_guide.json

chapter_sequence_guide Guide

System-specific rules for derive_chapter_sequence. Governs how atomic KB LOs are clustered into delivery modules. The LLM already knows curriculum theory — these rules are specific to this system's de

chapter_sequence_guide.json

design_pitfalls Shared

Run independently after domain scoring. Each violation is a named defect with a mandatory concrete fix.

design_pitfalls.json

lo_criteria Shared

LO quality rules for S1 derive_lo. System-specific — do not repeat general pedagogy theory.

lo_criteria.json

module_D Module

Instructional Design rubric

7 criteria 25 pts Instructional Design module_D.json

module_L Module

Accessibility rubric

8 criteria 25 pts Accessibility module_L.json

module_P Module

Pedagogy rubric

8 criteria 25 pts Pedagogy module_P.json

module_X Module

Experience rubric

7 criteria 25 pts Experience module_X.json

module_gates Module

Scoring / reference file.

module_gates.json

recommendation_protocol Guide

Scoring / reference file.

recommendation_protocol.json

research_synthesis_guide Guide

System-specific rules for S2 research_deep synthesis. The LLM already knows CPA, Variation Theory, cognitive load, ZPD. These rules are specific to this system.

research_synthesis_guide.json

trace_guide Guide

Traces check whether a sequence holds across content. Score by walking in order — does step N lead logically to step N+1?

trace_guide.json

Domain Rules

Injected into skill prompts via {{domain_rules}} at template render.

advisory_gate_protocol Tier B JSON

Advisory gates provide structured enhancement recommendations without blocking the pipeline. They surface opportunities for improvement that are valuable but not gatekeeping — the content can proceed without addressing them.

1 consumer score_gates

applet_conventions Tier A MD

When reviewing module content that contains applet placeholder slides, or when

4 consumers score_gatesscore_rubric_domainscan_design_pitfalls +1

applet_scope_discipline Tier B JSON

{ "_meta": { "tier": "B", "source_template": "assess_learning_coverage.md", "consumers": ["assess_learning_coverage", "compile.py"], "description": "Scope discipline rules for applet learning coverage assessment — prevents module-level concerns from contaminatin

1 consumer assess_learning_coverage

applicable_field_protocol Tier B JSON

{ "_meta": { "tier": "B", "source_template": "score_rubric_domain.md", "consumers": ["all_scorers", "compile.py", "aggregator"], "description": "Protocol for handling rubric criteria that declare applicable_to restrictions based on content type." }, "protoco

1 consumer score_rubric_domain

concept_attribution Tier B JSON

{ "_meta": { "tier": "B", "source_template": "score_rubric_domain.md", "consumers": ["S2_structure", "S3_pedagogy", "compile.py", "assess_learning_coverage"], "description": "Taxonomy for per-slide concept attribution with action types and evidence strength."

1 consumer score_rubric_domain

convention_requirements Tier B MD

Applies when KB data includes convention flags or false-friend warnings for a topic. These arise when mathematical notation, terminology, or conventions differ across regions, curricula, or languages, creating potential confusion for students.

2 consumers build_blueprintenrich_concepts

coverage_audit_protocol Tier B JSON

{ "_meta": { "tier": "B", "source_template": "assess_learning_coverage.md", "consumers": ["assess_learning_coverage", "compile.py", "gate_evaluator"], "description": "Structured cross-reference protocol for auditing misconception, vocabulary, and visual model co

1 consumer assess_learning_coverage

delivery_language Tier A MD

When evaluating, building, or assessing educational content that belongs to a

8 consumers research_deepbuild_blueprintbuild_applet_screens +5

dev_note_anchoring Tier B MD

Applies when any scorer emits developer-facing implementation notes — recommendations about touch targets, accessibility, alt-text, animation timings, transition behaviors, or other technical specifications that a developer would action during build.

1 consumer score_rubric_domain

kb_gap_handling Tier A MD

Applies whenever a blueprint builder, scorer, or mastery assessor encounters missing data in the KB — no Visual Instructional Representation (VIR) for a topic, no misconception data, no vocabulary definitions, or sparse coverage for a learning objective.

3 consumers build_blueprintenrich_engagementbuild_applet_screens

lo_code_alignment Tier A MD

Applies when any agent must match module content to KB learning objective codes (lo_codes). This is one of the most error-prone steps in the pipeline — keyword matching produces false positives that cascade into incorrect alignment scores.

3 consumers derive_chapter_sequencederive_module_loderive_applet_lo

media_insertion_heuristics Tier A MD

Applies when any scorer or blueprint builder must decide whether to recommend inserting a video, applet, or textbook reference at a particular point in the module flow. Also governs the worked-example routing rule for determining whether multi-step content belongs on slides or in

3 consumers build_blueprintenrich_engagementscore_rubric_domain

module_annotation_schema Tier B JSON

Each module in the sequence must carry structured annotations that downstream agents consume. These annotations are set during module sequencing and refined during blueprint construction. Incomplete annotations block downstream agents from producing quality output.

1 consumer derive_chapter_sequence

module_level_tropes Tier A MD

When evaluating, building, or researching content that involves tropes (Snapshot,

7 consumers research_deepbuild_blueprintbuild_spine +4

module_sequencing_rules Tier B MD

Applies when derive_module_sequence builds the ordered list of modules for a chapter. These three constraints must all be satisfied simultaneously — they are not independent options.

1 consumer derive_chapter_sequence

scope_discipline Tier A MD

When reviewing, scoring, building, or assessing content for a module or applet

10 consumers research_deepbuild_blueprintbuild_spine +7

slide_role_observations Tier B JSON

{ "_meta": { "tier": "B", "source_template": "score_rubric_domain.md", "consumers": ["S6_detail", "S6_express", "S7_vision", "compile.py", "render.py"], "description": "Canonical slide-role taxonomy with confidence levels and design pitfall suspension rules."

2 consumers score_rubric_domainscan_design_pitfalls

slide_tagging_protocol Tier B JSON

Every slide in the blueprint must carry explicit pedagogical commitment tags that downstream scorers can independently verify. These tags are the contract between the blueprint author and the scorer — if a tag is present, the scorer will check for it.

2 consumers build_blueprintbuild_spine

storyboard_read_rules Tier A MD

When vision-scoring or evaluating applet storyboard PPTX files where multiple

3 consumers score_gatesscore_rubric_domainscan_design_pitfalls

textbook_floor_principle Tier B MD

Applies when building module sequences, blueprints, or assessing mastery against textbook coverage. The textbook is the starting reference but not the limit of what must be taught.

1 consumer derive_chapter_sequence

visual_hierarchy_rules Tier A MD

When vision-scoring slides or screens that may suffer from organisational

3 consumers score_gatesscore_rubric_domainscan_design_pitfalls

visual_model_protocol Tier A MD

Applies whenever a vision scorer (S6 Detail, S6 Express, S7) inspects slide thumbnails and must emit `observed_visuals` entries. This protocol governs what counts as a reportable visual model and how to describe it.

4 consumers build_blueprintenrich_engagementbuild_applet_screens +1

vocabulary_status_protocol Tier B JSON

During deep research, every vocabulary term must be classified by its relationship to KB data. This classification flows downstream to blueprint builders and scorers so they know which terms are KB-backed and which need attention.

1 consumer research_deep

vocabulary_tracking Tier B JSON

Every mathematical term that the module introduces must be tracked with its introduction point and formalization level. This plan is set during blueprint and verified during scoring and learning coverage assessment.

2 consumers build_blueprintenrich_concepts

Rule \ Skill	assess_learning_coverage	build_applet_screens	build_blueprint	build_spine	derive_applet_lo	derive_chapter_sequence	derive_module_lo	enrich_concepts	enrich_engagement	enrich_experience	generate_action_summaries	research_deep	scan_design_pitfalls	score_gates	score_rubric_domain
advisory_gate_protocol B	·	·	·	·	·	·	·	·	·	·	·	·	·	●	·
applet_conventions A	●	·	·	·	·	·	·	·	·	·	·	·	●	●	●
applet_scope_discipline B	●	·	·	·	·	·	·	·	·	·	·	·	·	·	·
applicable_field_protocol B	·	·	·	·	·	·	·	·	·	·	·	·	·	·	●
concept_attribution B	·	·	·	·	·	·	·	·	·	·	·	·	·	·	●
convention_requirements B	·	·	●	·	·	·	·	●	·	·	·	·	·	·	·
coverage_audit_protocol B	●	·	·	·	·	·	·	·	·	·	·	·	·	·	·
delivery_language A	●	●	●	·	·	·	·	·	·	·	●	●	●	●	●
dev_note_anchoring B	·	·	·	·	·	·	·	·	·	·	·	·	·	·	●
kb_gap_handling A	·	●	●	·	·	·	·	·	●	·	·	·	·	·	·
lo_code_alignment A	·	·	·	·	●	●	●	·	·	·	·	·	·	·	·
media_insertion_heuristics A	·	·	●	·	·	·	·	·	●	·	·	·	·	·	●
module_annotation_schema B	·	·	·	·	·	●	·	·	·	·	·	·	·	·	·
module_level_tropes A	·	●	●	●	·	·	·	·	·	·	·	●	●	●	●
module_sequencing_rules B	·	·	·	·	·	●	·	·	·	·	·	·	·	·	·
scope_discipline A	●	●	●	●	·	·	·	●	·	●	·	●	●	●	●
slide_role_observations B	·	·	·	·	·	·	·	·	·	·	·	·	●	·	●
slide_tagging_protocol B	·	·	●	●	·	·	·	·	·	·	·	·	·	·	·
storyboard_read_rules A	·	·	·	·	·	·	·	·	·	·	·	·	●	●	●
textbook_floor_principle B	·	·	·	·	·	●	·	·	·	·	·	·	·	·	·
visual_hierarchy_rules A	·	·	·	·	·	·	·	·	·	·	·	·	●	●	●
visual_model_protocol A	·	●	●	·	·	·	·	·	●	·	·	·	·	·	●
vocabulary_status_protocol B	·	·	·	·	·	·	·	·	·	·	·	●	·	·	·
vocabulary_tracking B	·	·	●	·	·	·	·	●	·	·	·	·	·	·	·

Prompt Templates

config/prompts/skill_templates/*.md — sent to the LLM per skill call.

assess_learning_coverage MD

Used by mod_review+app_review · 34 lines · 2 variables

{{anti_bias}}{{domain_rules}}

build_blueprint MD

Used by mod_review+app_review · 242 lines · 3 variables

{{anti_bias}}{{domain_rules}}{{rubric_section}}

build_spine MD

Used by mod_review · 115 lines · 2 variables

{{anti_bias}}{{domain_rules}}

derive_applet_lo MD

Used by app_review · 54 lines · 3 variables

{{anti_bias}}{{domain_rules}}{{rubric_section}}

derive_lo_brief MD

Used by ch_research · 59 lines · 2 variables

{{anti_bias}}{{rubric_section}}

derive_module_lo MD

Used by mod_review · 41 lines · 3 variables

{{anti_bias}}{{domain_rules}}{{rubric_section}}

derive_module_sequence MD

Used by ch_research · 129 lines · 3 variables

{{anti_bias}}{{domain_rules}}{{rubric_section}}

enrich_concepts MD

Used by mod_review · 37 lines · 2 variables

{{anti_bias}}{{domain_rules}}

enrich_engagement MD

Used by mod_review · 30 lines · 2 variables

{{anti_bias}}{{domain_rules}}

enrich_experience MD

Used by mod_review · 61 lines · 3 variables

{{anti_bias}}{{domain_rules}}{{pacing_state}}

extract_applet_summary MD

Used by mod_review · 28 lines · 1 variables

{{domain_rules}}

generate_action_summaries MD

Used by mod_review+app_review · 34 lines · 2 variables

{{anti_bias}}{{domain_rules}}

research_deep MD

Used by mod_review+app_review+ch_research · 193 lines · 3 variables

{{anti_bias}}{{domain_rules}}{{rubric_section}}

scan_design_pitfalls MD

Used by mod_review+app_review · 17 lines · 4 variables

{{anti_bias}}{{domain_rules}}{{recommendation_protocol}}{{rubric_section}}

score_gates MD

Used by mod_review+app_review · 14 lines · 3 variables

{{anti_bias}}{{domain_rules}}{{rubric_section}}

score_rubric_domain MD

Used by mod_review+app_review · 37 lines · 3 variables

{{anti_bias}}{{domain_rules}}{{rubric_section}}

Field Contracts

config/field_contracts.json — declarative produces/consumed_by contracts per skill.

module_lo_brief_enriched 5 fields 5 consumers

S28_reshape — enriched MODULE_LO_BRIEF.json. Wraps derive_module_lo output with compiled chapter ground truth (slimmed: lo_components, vocabulary_terms, misconception_ids, pedagogical_models), ±1 siblings, and deterministic drift flags.

Field	Type	Required	Note
derived	object	●	LLM derive_module_lo output: module_lo, lo_components, prerequisite_chain, content_type, lo_codes, notes
planned	object	●	Slimmed compile from CHAPTER_MODULE_SEQUENCE: position, module_title, content_type, lo_components, cpa_phase, lo_codes, vocabulary_terms, misconception_ids, pedagogical_models, flags
predecessor	object	○	Position-1 sibling (condensed: position, module_title, lo_components, content_type, cpa_phase, vocabulary_terms, misconception_ids)
successor	object	○	Position+1 sibling (same condensed shape)
scope_drift_flags	array	○	Deterministic drift flags: lo_components overlap, content_type mismatch, lo_codes delta

consumed by: research_deep_synthesisbuild_module_blueprintrender_module_lo_brief_htmlrender_module_research_htmlderive_applet_lo

research_deep_synthesis 14 fields 8 consumers

Field	Type	Required	Note
scope	string	●
lo	string	●
lo_misconceptions	array	●
misconceptions	array	●
visual_models	array	●
ias_patterns	array	●
conventions_flags	array	●
gaps	array	●
synthesis_narrative	string	●
textbook_references	array	●
vocabulary_status	array	●
translation_notes	array	○	S20 addition — first-class bilingual pairs
prerequisites_to_reactivate	array	○	S22 addition — module scope only, feeds knowledge trace prerequisite tracking
lo_components_enrichment	array	●	S29: per-component pedagogical reference material. One entry per input lo_component (preserving index order) at module/applet scope. Empty list acceptable at chapter scope. Each entry: {component_index, component_statement (echo of input), visual_treatment_reference (research-grade multi-paragraph reference describing proven visual strategies), resource_bindings [{interactive_id, url, activity_title, pedagogical_value ≥120 chars, confidence: high\|medium\|low}] grounded STRICTLY in kb_evidence.interactive_landscape input, derivation_evidence (optional), rationale (optional)}. This is the module/applet-layer enrichment producer — kb_write_agent consumes this directly.

consumed by: build_module_blueprintbuild_applet_screensscore_gates_resultscore_rubric_domain_resultscan_design_pitfalls_resultassess_learning_coverage_resultwrite_tracekb_write_agent

build_module_blueprint 21 fields 6 consumers

Field	Type	Required	Note
lo	string	●
lo_components	array	○	S28_reshape: inherited lo_components[{statement, bloom_verb, visual_treatment}] for traceability
content_type	string	●
slides	array	●
applet_placements	array	●
video_placements	array	●
visual_models	array	●
trope_map	object	●
cpa_summary	object	●
textbook_alignment	array	●
vocabulary_introduction_plan	array	●
convention_requirements	array	○
scaffolding_level_per_slide	string	○	S23: per-slide scaffolding intensity (full/partial/guided/independent)
vocabulary_on_this_slide	object	○	S23: per-slide vocab state with definitions, conventions, false-friend warnings
visual_model_description	string	○	S23: per-slide visual layout description beyond VIR id
tgar_role_per_slide	string	○	S23: per-slide TGAR cross-module arc role
major_idea_per_slide	string	○	S23: per-slide key idea (one sentence)
transition_from_previous	string	○	S23: per-slide cognitive/narrative bridge from previous slide
asset_context_role	string	○	S29: structural role — assigned deterministically by compile pass from applet_placements, video_placements, slide.tropes, section. 13-value taxonomy (content/applet_/video_/trope_*/worked_example/exit_question/summary). Legacy S23 values (context/placeholder/consolidation) retained for backwards compat.
teaching_sufficiency_note	string	○	S23: content clarity and deliverability assessment

consumed by: score_gates_resultscore_rubric_domain_resultscan_design_pitfalls_resultassess_learning_coverage_resultcompile_reviewrender_gate

build_applet_screens 24 fields 6 consumers

Field	Type	Required	Note
lo	string	●
primary_ias_archetype	string	●
misconception_targeted	string	●
screens	array	●	S23: may exceed 6 (multiple screens per IAS stage)
scaffolding_arc	string	●
visual_models_referenced	array	●
convention_requirements	array	○
screen_label	string	●	S23: distinguishes screens within same stage
student_sees	string	●	S23: visual layout at screen entry
student_does	string	●	S23: physical interaction
student_thinks	string	●	S23: reasoning demand
intermediate_states	array	●	S23: if-then interaction sequence
visual_model_description	string	●	S23: layout/colours/spatial arrangement
visual_model_purpose	string	●	S23: pedagogical rationale for visual
feedback	object	●	S23: on_correct/incorrect/partial/misconception + timing
screen_zones	object	●	S23: action_instruction/scaffolding_text/dev_notes
affordances	array	●	S23: element + visual_cue pairs
variants_and_invariants	object	●	S23: mathematical structure (variants/invariants)
time_estimate	object	●	S23: thinking/interaction/feedback seconds
difficulty_level	string	●	S23: entry/developing/proficient/extending
difficulty_rationale	string	●	S23: why this difficulty level
transition_from_previous	object	●	S23: previous_end_state/transition_purpose/this_screen_entry_state
teaching_sufficiency_note	string	●	S23: content clarity and deliverability

consumed by: score_gates_resultscore_rubric_domain_resultscan_design_pitfalls_resultassess_learning_coverage_resultcompile_reviewrender_gate

score_gates_result 3 fields 1 consumers

Field	Type	Required
gate_set	string	●
content_type	string	●
gates	array	●

consumed by: compile_review

score_rubric_domain_result 10 fields 1 consumers

Field	Type	Required	Note
domain	string	●
domain_code	string	●
scope	string	●
content_type	string	●
criteria	array	●
domain_total	integer	●
domain_verdict	string	●
observed_visuals	array	○	Vision scorer only (D/X domains)
concepts_on_slide	array	○	P domain only
slide_role_observations	array	○	Vision scorers — feeds design pitfall suspension in compile

consumed by: compile_review

scan_design_pitfalls_result 3 fields 1 consumers

Field	Type	Required	Note
pattern_set	string	●
violations	array	●
slide_role_observations	array	○	Feeds compile design pitfall suspension

consumed by: compile_review

assess_learning_coverage_result 2 fields 1 consumers

Field	Type	Required	Note
content_type	string	●
mastery_estimate	object	●

consumed by: compile_review

derive_module_lo 7 fields 2 consumers

S28_reshape: lo_components with SCOPE-INTENT visual_treatment (creator direction, editable at MODULE_LO_BRIEF hard gate). S29 correction: visual_treatment here is scope intent, NOT pedagogical reference. The reference-grade enrichment is produced LATER by research_deep(module) as lo_components_enrichment. derive_module_lo does NOT feed kb_write_agent — that trigger moved to post-research.

Field	Type	Required	Note
module_lo	string	●
lo_components	array	●	Validated lo_components[{statement, bloom_verb, visual_treatment}] — visual_treatment is SCOPE INTENT (creator-editable at gate), NOT enrichment. The research_deep(module) step produces the reference-grade visual_treatment_reference downstream.
prerequisite_chain	array	●
content_type	string	●
lo_codes	array	●
derivation_evidence	array	●	Source trail per component: {component_statement, source, source_text}. Consumed by MODULE_LO_BRIEF gate render; not written to KB by this skill (research_deep does the KB write).
notes	string	○

consumed by: research_deep_synthesisbuild_module_blueprint

derive_applet_lo 6 fields 2 consumers

S28_reshape: receives parent_lo_components with visual_treatment inheritance from module. S29 correction: derive_applet_lo produces applet scope intent, NOT enrichment. lo_components / derivation_evidence fields may be emitted as narrowing refinements, but reference-grade enrichment is produced by research_deep(applet) downstream. derive_applet_lo does NOT feed kb_write_agent.

Field	Type	Required	Note
applet_lo	string	●
applet_type	string	●
lo_codes	array	●
lo_components	array	○	OPTIONAL refinement — applet-scoped subset of parent (scope intent only; not pedagogical enrichment).
derivation_evidence	array	○	OPTIONAL — per-component source trail for scope-intent refinement.
notes	string	○

consumed by: research_deep_synthesisbuild_applet_screens

derive_lo_brief 5 fields 1 consumers

S28_reshape: added surfaced_lo_codes for curriculum LO surfacing at chapter level

Field	Type	Required	Note
chapter_lo	string	●
prerequisite_chain	array	●
curriculum_alignment	array	●
surfaced_lo_codes	array	●	All KB LO codes relevant to chapter from target + sister curricula, classified as core/extension/prerequisite
notes	string	●

consumed by: derive_module_sequence

derive_module_sequence 3 fields 3 consumers

S28_reshape: per-module schema slimmed — lo_components replaces free-text lo, vocabulary/misconceptions/models slimmed to terms/ids/names, deferred fields removed

Field	Type	Required	Note
module_sequence	array	●	Per module: position, module_title, content_type, lo_components[{statement,bloom_verb}], lo_codes, cpa_phase, vocabulary_terms[], misconception_ids[], pedagogical_models[], flags[]
textbook_validation	object	○
reconciliation_map	array	○

consumed by: render_gatefinalize_chapter_researcherkb_write_agent

kb_write_agent

S29 (corrected): Append-only enrichment history on lo_components. Three layer-specific entry points: run_kb_write_chapter, run_kb_write_module, run_kb_write_applet. Triggered at the RIGHT producer moments — chapter at finalize (skeleton only), module AFTER research_deep(module) produces lo_components_enrichment (rich), applet AFTER research_deep(applet). derive_module_lo / derive_applet_lo do NOT trigger KB writes — they produce scope intent, not enrichment. Every call appends exactly one diff-only event to the row's enrichments JSONB array. Identity fields (master_lo_code, statement, bloom_verb, embedding) are set on first INSERT and never modified.

extract_applet_summary 7 fields 1 consumers

S26: new entry for applet PPTX summary extraction

Field	Type	Required
applet_title	string	●
lo_served	string	●
ias_archetype	string	●
visual_model	string	●
interaction_modes	array	●
screen_summaries	array	●
alignment	object	●

consumed by: compile_review

refine_module_result 2 fields 1 consumers

S26: new entry for gate refine/split/merge module operations

Field	Type	Required	Note
modules	array	●
reasoning	string	●

consumed by: render_gate

insert_module_result 2 fields 1 consumers

S26: new entry for gate insert-module operation

Field	Type	Required	Note
modules	array	●
reasoning	string	●

consumed by: render_gate

generate_action_summaries_result 6 fields 1 consumers

S26: new entry for action summary blurbs (Haiku)

Field	Type	Required
summary_blurb	string	●
gate_summary	string	●
recommendation_summary	string	●
mastery_narrative	string	●
mastery_action_line	string	●
slide_actions	array	●

consumed by: compile_review

refine_gate_entry 1 fields 1 consumers

S26: new entry for individual gate entry refine/edit. Polymorphic schema — entries shape varies by entry_type.

Field	Type	Required	Note
reasoning	string	●

consumed by: render_gate

This page is a live, queryable view of the ACRLDS pipeline — every node, schema, prompt, rule, gate, and contract as they exist on disk right now. Everything is read-only; edits happen in the config/ tree or the agent source.

The six tabs

Pipeline — the node graph (chapter research → module review → applet review). Each node is a skill, tool, gate, or render step with inputs, outputs, model/cost, and click-through to schemas + prompts + constraints. Filter chips at the top narrow by agent, type, or gate kind.
Gates — every hard / critical / advisory gate with the scorer that raises it, consumer that reads it, and the rubric entry defining it. Click a gate id for its JSON definition.
Rubrics — the scoring rubric files under config/rubric/. Grouped as Module / Applet / Shared / Guide. Click any card for Overview, Criteria table, and Raw JSON (inline, pretty-printed).
Rules — the domain rule files under config/rubric/rules/ injected into skill prompts via . Tier A rules are shared across 3+ skills; Tier B are narrower. Each card shows its consumers and co-injected rules.
Prompts — the skill prompt templates under config/prompts/skill_templates/. Full template body + list per prompt.
Contracts — the field-level producer/consumer map (config/field_contracts.json). Useful for tracing which skill's output field feeds which downstream skill's input.

Modals

Cards open a tabbed modal with a dark panel and a blurred halo separating it from the page. Each modal has topic-specific tabs: Overview (metadata + summary), Criteria / Content / Variables (body), Raw JSON (inline, pretty-printed, no second-modal indirection).

Dismiss with the × button, the ESC key, or by clicking the backdrop.

Search

The search box in the top bar matches node labels, rule names, rubric ids, criterion names, and field contract keys across all tabs. Switching tabs keeps the filter active.

What this is good for

Onboarding — browse the Pipeline tab end-to-end to see the full chapter → module → applet flow.
Debugging a scoring verdict — find the rubric criterion in the Rubrics tab, follow its rules + prompt to the originating skill, check field contracts for upstream drops.
Adding a new signal — use Contracts to find where the new field needs to flow, then Prompts to add the production instruction, then Rules if the signal should be cross-injected.
Reviewing gate behaviour — Gates tab gives the scorer → consumer chain per gate, with the rubric entry linked.

Note

This page mirrors what's on disk. If a card shows stale data, reload the page — the server re-reads config/ on each request. Edits persist immediately; no restart needed.

Pipeline Flow

Gate Registry

Rubrics

Domain Rules

Prompt Templates

Field Contracts

—

How to use this page

The six tabs

Modals

Search

What this is good for

Note