The Yemaya Study & Deconstruction Workspace is an evidence-grounded environment for learning filmmaking and game development by examining authorized source works from whole-work structure down to moments, frames, regions, objects, performance choices, and gameplay events. It connects observations to sources, comparisons, notebooks, exercises, and original creative decisions.
Study Workbench aliases and official name#
Search aliases: Yemaya Study Workbench, Study Workbench, Study Workspace, Deconstruction Workbench. The canonical product name is Yemaya Study & Deconstruction Workspace.
Honest status#
The canonical proposal labels itself a product and architecture proposal and does not claim the entire workspace is shipped. Since that proposal, the repository has accumulated real contracts, schemas, a service, migrations, tools, tests, evidence registers, and runbooks. Those assets prove implementation depth; they do not by themselves prove that every designed journey is user-reachable in a supported deployment.
Use these labels when discussing a capability:
- Designed — specified in the proposal, a decision record, or contract.
- Implemented — real code or schema exists and has direct tests.
- Wired — a supported UI/API path reaches the implementation.
- Operationally evidenced — the deployed path has current rollout, observability, security, recovery, and acceptance evidence.
- Withheld — the system deliberately cannot assert a result because input, rights, provider, measurement, or deployment evidence is missing.
Product thesis and non-goals#
The workspace treats study as a chain of attributable judgments, not a folder of screenshots or an automated style extractor. A learner can say what they noticed, where it occurred, how it relates to other moments, why it matters, and what original decision it informed. A reviewer can inspect the chain.
It is not:
- a piracy, ripping, DRM-bypass, or indiscriminate web-ingestion product;
- a model that converts protected works into unattributed imitation;
- a replacement for subject-matter teaching or legitimate disagreement;
- a claim that automated emotion, performance, authorship, or intent inference is objective ground truth;
- a place where “AI confidence” substitutes for evidence or review;
- a generic asset manager detached from learning and creative decisions.
Ownership and domain boundaries#
| Concern | Primary owner | Boundary |
|---|---|---|
| Workspace orchestration, projects, study sessions, assemblies | Yemaya | Owns the product workflow and composition; consumes other domains through versioned adapters. |
| Notebooks, passages, annotations, structured reference | Nisaba | Owns durable study/knowledge structures and linked reference semantics. |
| Evidence, provenance, claims, sources, grounding | Sophia | Owns evidence-bearing assertions and the ability to abstain or expose disagreement. |
| Film/visual craft metrics and analysis | Hathor | Owns its declared technical quantities; the UI must not relabel heuristics as measured fact. |
| Reference video and motion analysis | Aja | Owns source-video/motion adapters, pose/skeleton conventions, and motion lineage. |
| Audio and music analysis | Euterpe | Owns audio/music analysis and audio-specific playback/rendering concerns. |
| Visual/brand systems | Aglaea | Owns design-language and visual-system semantics. |
| Game runtime and DCC/engine integration | Bellona, Neith, Maya | Own runtime/engine adapters, replay/interchange, and engine-specific constraints. |
| Learning paths and assessment | Metis | Owns exercises, rubric/evaluation, adaptive sequencing, and learning outcomes. |
| Generative production | Isis | Owns governed generation, model/workflow policy, and generated-artifact lineage. |
| Shell, identity, tenant, policy, and member experience | Oshun / Iris | Own identity, session, memory/consent, tenant, grants, and common Studio boundaries. |
The detailed adapter promises are recorded in Domain Responsibility Contracts.
Primary users and jobs#
The workspace explicitly serves multiple expert frames rather than assuming a single “creator” persona:
| User | Core jobs |
|---|---|
| Learner filmmaker | Map structure and scenes; inspect shots, blocking, lenses, lighting, edit rhythm, sound, performance, and production design; turn observations into exercises. |
| Learner game developer | Align gameplay video with input, telemetry, state, camera, encounters, levels, and systemic causes; compare authored and emergent behavior. |
| Actor, animator, or director | Study beats, objectives, tactics, timing, gesture, pose, gaze, voice, and relationships without presenting inference as a diagnosis. |
| Character, costume, and concept artist | Trace silhouette, proportion, material, palette, shape language, costume, continuity, and visual development with rights-aware references. |
| Teacher, mentor, or team lead | Prepare guided studies, assignments, cohorts, review, rubrics, discussion, progress, and exportable evidence. |
| Researcher or critic | Build reproducible corpora, queries, comparison sets, citations, annotations, relation graphs, methodological notes, and versioned exports. |
Acceptance expectations for each are in the persona matrix.
Core journey#
Every arrow has failure and permission states. Ingest may quarantine a file; analysis may be unavailable or withheld; comparison may expose incomparable methods; sharing may be blocked by rights; export may omit restricted media; deletion may remain in an auditable pending state while derived material and backups are reclaimed.
Workspace information architecture#
A complete surface needs these connected regions, even when responsive layouts collapse them into drawers or sequential views:
- Project and source library — project scope, source status, rights, processing, versions, collaborators, and recent activity.
- Viewer and synchronized timelines — media/game playback, timecode, frames, audio, transcript, events, camera/input/state tracks, and compare alignment.
- Annotation and evidence lanes — boundaries, regions, objects, poses, performance choices, claims/observations, citations, confidence/status, and disagreement.
- Inspector — source identity, technical metadata, rights, derivation, provider/model/method, author/reviewer, and history.
- Notebook and graph — notes, questions, excerpts, collections, relations, motifs, hypotheses, creative decisions, and backlinks.
- Search and comparison — filters, saved queries, similarity with declared semantics, side-by-side/overlay/timeline/graph views, and alignment state.
- Learning and practice — guided studies, exercises, rubrics, feedback, progress, accessibility adaptations, and reflection.
- Export and administration — manifests, citations, media policy, signatures, access, retention, expiry, deletion, and operational status.
The reusable Oshun shell and known gaps are cataloged in the Studio surface inventory.
Sources and granularity#
The format matrix accounts for film/television/animation/cinematics, gameplay and instrumented sessions, still images and concept art, scripts, storyboards, documents, paratexts, and creator-owned assets. Each format declares ingest, preview, time/space anchoring, extraction, analysis, export, rights, and failure expectations; unsupported formats must fail visibly.
Evidence can attach at multiple levels:
- work, edition/cut/build, sequence/level, scene/encounter, shot/take/event;
- time range, frame, region, track, object instance, pose, speaker turn, line, script span, document page/region, input or state transition;
- a relation between anchors, such as continuation, contrast, cause, response, motif recurrence, adaptation, synchronization, or uncertain association.
An anchor is not only coordinates. It carries source and version identity, coordinate/time basis, method, author, creation time, status, and enough context to detect when a new source revision invalidates or reprojects it.
Annotation tasks#
The maintained annotation guide defines task-specific instructions and quality expectations for:
- shot boundaries;
- transcript segments and speaker turns;
- object instances and regions;
- pose tracks;
- cross-reference relations;
- performance-choice functions.
Task guides distinguish observation from interpretation, require “cannot determine” paths, support disagreement/adjudication, and preserve the guide and schema version used. See the generated annotation guide.
Film and animation study#
Film-specific analysis can include structure, scene and shot segmentation, composition, camera, lens and depth cues, blocking, lighting, color, production design, edit rhythm, continuity, transitions, sound, dialogue, music, performance, animation, VFX, and text/storyboard relationships.
The UI must state how each value was obtained:
- directly read from source metadata;
- measured by a declared deterministic method;
- produced by a versioned model/provider;
- manually annotated;
- inferred as a hypothesis;
- reviewed/adjudicated;
- unavailable or withheld.
Emotion and performance labels are especially sensitive. They are annotations or hypotheses tied to observable cues and context, never medical or psychological diagnosis and never a proxy for protected traits.
Game study#
Game analysis needs more than a video player. When sources permit, it aligns:
- capture video/audio with game build, platform, level/encounter, session, and player/actor identity;
- input, camera, animation, state-machine, physics, AI, quest, combat, economy, UI, and performance events;
- authored sequences with emergent paths and repeated trials;
- spatial positions/regions with timeline moments and causal hypotheses;
- replay or telemetry schema versions with the exact build that emitted them.
When only footage is present, the workspace must not fabricate engine state. It can record visible observations and hypotheses while marking telemetry, input, or causal explanation unavailable.
Search, comparison, and synthesis#
Search spans lexical content, transcripts, tags, entities, source metadata, annotations, rights state, technical measurements, graph relations, and declared similarity features. Results preserve source/anchor identity and permission filters. A result count is never allowed to reveal a source the requester cannot access.
Comparison supports synchronized playback, side-by-side views, overlays, timelines, tables, graphs, and qualitative notes. Alignment records the method and uncertainty; comparison never silently stretches sources into apparent equivalence.
Synthesis produces original creative decisions, exercises, briefs, boards, tests, or generation requests linked back to the observations that informed them. The provenance chain keeps inspiration legible while rights and policy determine what media can be copied, embedded, exported, or used by a model.
Learning and collaboration#
Projects can organize guided studies, questions, exercises, rubrics, submissions, feedback, cohorts, and progress. Metis owns learning and assessment semantics; Yemaya assembles them around study sources. Evaluation must allow legitimate disagreement and separate annotation consistency from creative quality.
Collaboration requires project roles, scoped source access, comments/review, assignment, presence where supported, revision history, conflict behavior, and attributable acceptance/rejection of suggestions. A collaborator's access to a notebook does not imply access to every underlying source.
Accessibility, internationalization, and offline/partial behavior#
The workspace targets keyboard-complete operation, visible focus, semantic regions and controls, accessible timelines and alternatives for graphical views, caption/transcript workflows, reduced motion, zoom/reflow, contrast, screen-reader status, touch targets, and error recovery. Media analysis must not require color, audio, pointing, drag, or precise timing as the only path.
Locale affects text, number/date/time, reading direction, transcript language, tokenization, search, caption rendering, and export. Supported desktop, browser, hardware, decoder, assistive-technology, locale, network, and runtime baselines are explicit in Supported Baselines.
Offline and partial states must say what is cached, editable, queued, unavailable, stale, or awaiting synchronization. “Saved” is reserved for a durable acknowledged write, not a local optimistic label.
Continue reading#
- Study data, evidence, learning, and operations — schemas, rights, provenance, evaluation, runbooks, and gates.
- Data, rights, security, and safety — the cross-workbench trust model.
- Development, testing, and change map — exact code/evidence roots and verification lanes.
- Full proposal — exhaustive product, system, rollout, and requirement detail.