# The future laboratory Luis Rodriguez · September 6, 2026 ## Direction A family of working instruments connected by one research practice: ask a question, manipulate its subject, calculate or observe a result, compare, and export evidence. The homepage preserves its personal introduction and living web. The collection now uses five research folders. The selected work now foregrounds document review and a personal watch workbench. These folders specify how the other instruments should evolve, not a claim that every future integration has shipped. ## Art that carries information Territory’s Age of Ultron account describes distinct engineering, biological and bioengineering visual identities. Sony Imageworks documents drawn linework, hatching, halftones and print-inspired effects in Spider-Verse. Those methods inform selected edges, sectional marks and brief feedback accents here; solver timestamps and interaction response are not quantized to movie animation. Sources: [Territory / Age of Ultron](https://territorystudio.com/project/marvels-avengers-age-of-ultron/), [Sony Imageworks / Spider-Verse](https://www.imageworks.com/our-craft/feature-animation/movies/spider-man-spider-verse), [Territory / Blade Runner 2049](https://territorystudio.com/project/blade-runner-2049/). These are artistic production accounts, not independent usability trials. Golden-ratio composition is optional, not a demonstrated universal usability requirement. ## Form & force Can the structure find its own shape? Start with the space you want. Let constraints and forces reveal the geometry that can hold it. ## Personal reason The web I wanted to play with became a question about architecture: how would I shape a little courtyard back home? ## Working entry [Explore how a structure carries force](/lab/loadpath/workshop/) — Working instrument · idealized truss model. ## Art and interaction A structural drawing makes the load route readable. Inside the instrument, color distinguishes force states and a comparison preserves the earlier structure. ## Instrument rebuild briefs ### Loadpath Current entry: [Remove a member and inspect where the force goes.](/lab/loadpath/workshop/). Rebuild around a repair question: which change restores a load route after a member is removed? Keep the present solver as the baseline; evaluate candidate repairs separately, show force redistribution and unsupported mechanisms, and export both cases. ## Next shared capability General inverse form-finding: specify several geometric and force goals, optimize them with JAX FDM, then use Manifold to turn a selected design into inspectable solid geometry. This is a researched direction, not an implemented integration. ## Verified foundations - [JAX FDM](https://github.com/arpastrana/jax_fdm) — MIT · Python / JAX. Differentiable equilibrium and inverse design of pin-jointed networks. Researched future engine; not integrated. Structural certification is outside scope. - [Manifold](https://github.com/elalish/manifold) — Apache-2.0 · browser WASM. Robust solid geometry, boolean cuts and model export. A geometry kernel, not a stress solver. Not integrated yet. ## Completion criteria A visitor can identify the task, act without a tutorial, inspect the resulting state and export a reproducible artifact. Sources, model limits, loading and recovery remain readable. Numerical checks, actual provider execution and browser/device checks are reported separately. ## Motion & mechanisms What makes a thing move with purpose? Open a mechanism, follow its dependencies, and compare a change against the same starting conditions. ## Personal reason My dad’s love of watches and my fascination with superhero movement meet at the same place: understanding the mechanism. ## Working entry [Open the watch workbench](/lab/weave/) — Working instrument · open movement / ideal model. ## Art and interaction Cutaway illustration and selective light make a part’s role visible. Cyan is the reference, coral is the intervention; time stays synchronized. ## Instrument rebuild briefs ### WEAVE Current entry: [Separate a watch and compare two versions of its timekeeper.](/lab/weave/). Rebuild around mechanism diagnosis and motion authoring. Preserve the documented OM10 baseline. A later MuJoCo adapter must model contacts and joint limits explicitly instead of adding inferred real-watch torque or accuracy. ### Inside Current entry: [Import an assembly and create an exploded explanation.](/lab/inside/). Make imported assemblies usable as technical explanations: verify units and hierarchy, preserve source attribution, compare named versions and use registered section views. Manifold could add real cuts where closed solid inputs permit them. ### Inside / energy Current entry: [Disconnect a fan and inspect the calculated consequence.](/lab/inside/experiment/). Grow into a thermal/electrical experiment workbench with explicit component models, calibrated data only when available, recorded interventions and energy accounting. Retain the useful fan-disconnection loop. ### Time Observatory Current entry: [Explore an open movement and save a shared study.](/lab/inside/time/). Keep the personal father-and-son exploration as the collaboration layer. Preserve actual invite/version conflict handling and experimental gesture controls; never imply that a photo reveals manufacturer CAD. ### Hero studies Current entry: [Change a swing, then dissect the forces at one moment.](/lab/hero-studies/). Treat this as speculative biomechanics: compare motion under stated assumptions, inspect time-indexed forces, and separate fictional capabilities from verified motion. A later MuJoCo character needs original rigging and tested contacts. ## Next shared capability Teach an original articulated mechanism to follow a drawn path. Use real contact, joint limits and measured tracking error; choreograph the motion without hiding failed plans. This is a researched direction, not an implemented integration. ## Verified foundations - [MuJoCo / official WASM](https://github.com/google-deepmind/mujoco/tree/main/wasm) — Apache-2.0 source · browser WASM. Google DeepMind’s articulated dynamics engine now has maintained JavaScript bindings. Bindings are still work in progress. They are researched, not installed here; planning and control still require implementation. - [Openmovement OM10](https://openmovement.org/documents/OM10/OM10_Technical_Information_20210314.pdf) — CC BY-SA 3.0 documentation. Published tooth counts, frequency and inertia ground the current watch study. Original schematic geometry; not Rolex or Omega manufacturer CAD. ## Completion criteria A visitor can identify the task, act without a tutorial, inspect the resulting state and export a reproducible artifact. Sources, model limits, loading and recovery remain readable. Numerical checks, actual provider execution and browser/device checks are reported separately. ## Light, sound & patterns What could we understand if we could see it? Give invisible fields a readable form. Listen, measure, isolate a pattern, and keep observation separate from prediction. ## Personal reason Physics gets interesting when something ordinary—a room, a moving point, a growing pattern—reveals a hidden rule. ## Working entry [Listen to a room change](/lab/resonance/room/) — Working instrument · calculated early reflections. ## Art and interaction Contour lines, signed fields and translucent layers reveal structure. Print-style hatching distinguishes a diagram layer from the observed subject. ## Instrument rebuild briefs ### Resonance Current entry: [Move a source and listener; compare the sound paths.](/lab/resonance/). Evolve from one sound scene into a room-design investigation: compare source/listener placements against a defined acoustic objective, import measured impulse responses separately, and use a difference view that preserves units and acquisition conditions. ### Afterimage Current entry: [Measure motion in a video and inspect the observations.](/lab/afterimage/). Make measurement uncertainty first class: camera calibration, tracking confidence, re-labeling, lens distortion and exported coordinates. Keep predictions visually distinct from observations. ### Living web laboratory Current entry: [Probe a strand, grow a chemical pattern, or change an orbit.](/lab/web/). Keep mechanics, reaction-diffusion and gravity as independent scientific adapters. Add explicitly defined inverse goals one at a time rather than pretending the same physics explains every organism and system. ## Next shared capability Light written by matter: draw a target image and optimize a simulated glass surface that projects it. A separate molecular study could inspect real structures and their provenance. This is a researched direction, not an implemented integration. ## Verified foundations - [Mitsuba / inverse caustics](https://mitsuba.readthedocs.io/en/stable/src/inverse_rendering/caustics_optimization.html) — Free research renderer · BSD-style terms. Optimizes a glass heightfield to produce a target image through simulated light. Requires a local/server Python/C++ job. It is not a browser shader trick or an implemented feature here. - [Mol*](https://github.com/molstar/molstar) — MIT · browser WebGL. Real macromolecular structure inspection, selections and saved views. A researched biology path. Molecular visualization does not predict folding or validate biological function. ## Completion criteria A visitor can identify the task, act without a tutorial, inspect the resulting state and export a reproducible artifact. Sources, model limits, loading and recovery remain readable. Numerical checks, actual provider execution and browser/device checks are reported separately. ## Earth & place How does technology learn to care for a place? Make local information understandable, traceable and useful—even when the network or the information is incomplete. ## Personal reason Puerto Rico is home. Its coast, architecture and weather make these questions personal. ## Working entry [Explore the landmark atlas](/lab/landmarks/) — Working instrument · sourced places and geometric studies. ## Art and interaction Topographic contours establish place. Dates, gaps and source boundaries have visible form; water-colored marks never turn missing data into certainty. ## Instrument rebuild briefs ### Landmark Loom Current entry: [Explore sourced landmarks, including Puerto Rico, and their geometric studies.](/lab/landmarks/). Grow into a place-and-form notebook: preserve sourced coordinates, identify original interpretations versus scans, and compare site constraints. Do not label all landmarks as reconstructable 3D models. ### Pocket Relay Current entry: [Prepare a note for offline transfer between devices.](/lab/relay/). Make resilient transfer the useful invention: paired devices, integrity receipts and recoverable segments. Physical device-to-device transfer needs actual device testing; file and local QR preparation alone do not prove it. ### After the Rain Current entry: [Restore the modeled systems of a coastal botanical station.](/lab/after-rain/). Retain the original place as a spatial systems case study. Make each repair reveal the circuit, thermal state and causal history; use it as a demonstration of the same inspectable models, not a promise of AAA production parity. ## Next shared capability A replayable place study: compare changes in a coastal dataset through time, preserving coverage gaps, source revisions and uncertainty as visible layers. This is a researched direction, not an implemented integration. ## Verified foundations - [VTK.js](https://github.com/Kitware/vtk-js) — BSD-3-Clause · browser. Scientific geometry and scalar-volume visualization, clipping and data readers. Rendering support is not an environmental model. New datasets need coverage, provenance and unit checks. - [Wikidata data access](https://www.wikidata.org/wiki/Wikidata:Data_access) — Structured data CC0. Sourced identifiers and coordinates for an open place atlas. Current atlas already uses Wikidata. External media and CAD keep their own licenses. ## Completion criteria A visitor can identify the task, act without a tutorial, inspect the resulting state and export a reproducible artifact. Sources, model limits, loading and recovery remain readable. Numerical checks, actual provider execution and browser/device checks are reported separately. ## Software & agency Can we rehearse a decision before trusting it? Make the software’s behavior visible: evidence, dependencies, retries, costs and the decision a person actually approved. ## Personal reason This is where the laboratory meets my daily work: making complicated systems easier to understand and change responsibly. ## Working entry [Investigate a document](/lab/fondeo/) — Working local workflow · source, correction and approval. ## Art and interaction Time runs in one direction. Branches show alternatives, edges show dependencies, and a changed state gets a brief localized signal. ## Instrument rebuild briefs ### Fondeo Current entry: [Correct a document field and approve the exact reviewed revision.](/lab/fondeo/). Keep document evidence and exact-revision approval as the core. Add provenance-preserving extraction adapters and compare evaluation sets before changing a model; AI output remains a proposal. ### Wind Tunnel Current entry: [Lose a reply, retry a request, and inspect duplicate behavior.](/lab/wind-tunnel/). Unify retry policy experimentation, reproducible fault scenarios and actual execution receipts. Connect observed traces as evidence while preserving the difference between simulated and executed runs. ### Follow the Thread Current entry: [Trace a field back to its source and acceptance rule.](/lab/thread/). Nest field-to-source explanation inside the document workflow. Its useful future is a portable evidence contract that survives extraction, edits, model changes and human review. ### Causality Current entry: [Change traffic and find a simulated bottleneck.](/lab/causality/). Use observed traces and declared dependencies for explanation, then test interventions in separately labeled models. Correlation and duration are not proof of causation. ### Foundry Current entry: [Compare hypothetical model-routing tradeoffs.](/lab/foundry/). Replace hypothetical profiles gradually with versioned evaluation results, measured latency and provider costs. Compare routing choices against a declared dataset and quality threshold before proposing automatic routing. ### Continuum Current entry: [Save an investigation another engineer can reproduce.](/lab/continuum/). Make saved experiments replayable across workers and agents: inputs, model versions, tool receipts, limits and branch lineage. Verify replay before claiming reliable autonomous handoff. ### Early notebooks Current entry: [Inspect the earlier programs and what I corrected.](/lab/notebooks/). Preserve the learning history as a supporting archive. Show the original question, the failure and the corrected result; it should support the main inventions rather than compete with them. ## Next shared capability Turn real traces and declared system constraints into diagrams that stay readable while you change them. Keep AI proposals separate from the checks that accept them. This is a researched direction, not an implemented integration. ## Verified foundations - [Penrose / Bloom](https://github.com/penrose/penrose) — MIT · browser / WASM. Constraint-based technical diagrams and interactive optimization. A diagram optimizer, not a physical solver or proof that an architecture is correct. Researched, not integrated yet. - [OpenTelemetry OTLP](https://opentelemetry.io/docs/specs/otlp/) — Open specification. Recorded operation timing and parent relationships for inspectable software. WEAVE already reads bounded local OTLP files. Recorded duration alone does not establish causality. ## Completion criteria A visitor can identify the task, act without a tutorial, inspect the resulting state and export a reproducible artifact. Sources, model limits, loading and recovery remain readable. Numerical checks, actual provider execution and browser/device checks are reported separately. ## Implementation boundary Fondeo is the current practical engineering focus. Its review interface runs with a local Node service or a labeled browser fallback. The Python/AWS adapters are a separate, locally tested source kit; no AWS execution is claimed. The open scientific engines listed here are researched foundations for supporting experiments, not newly installed product dependencies. The source catalog is lib/research-folders.ts. Product folders in projects/ organize ownership and rebuild briefs; app/lab/ contains the active routes listed in the catalog. No synced sources/ material was edited. Research and artist claims were checked against primary pages; future capability statements are recommendations. ## Perception and wayfinding: what the references support [Kevin Lynch's city research](https://betterworld.mit.edu/images-of-the-city/) identifies paths, edges, districts, nodes and landmarks. Applying that to stable folder order, breadcrumbs and recognizable project drawings is a design inference, not proof that a 3D portfolio is easier to use. [Fitts's original 1954 motor-control study](https://doi.org/10.1037/h0055392) connects aimed-movement performance to task amplitude and tolerance. For this interface the practical response is nearby, legible controls and larger targets, including HTML alternatives to small 3D supports. [W3C's target-size guidance](https://www.w3.org/WAI/WCAG22/Understanding/target-size-minimum.html) specifies a 24 CSS-pixel minimum with exceptions; this design generally uses larger controls. No accessibility conformance claim follows from CSS alone. [Tufte's own sparkline discussion](https://www.edwardtufte.com/notebook/sparklines-history-by-tufte-1324-to-now/) emphasizes compact graphics interpreted with their nearby context. Here the force, unit, source and limit sit beside the selected strand. Reducing decoration is a design judgment, not a universal instruction to remove every label or orienting mark.