# Unreal Keyboards EC research tools > **Current publication:** Force Curve Bench `fc-3.5` and EC Parts Builder > `lib-6.1` are published from this repository. The active multi-tool site > authority is `SITE_RELEASE_MANIFEST.json` with its `SHA256SUMS` inventory. The corresponding Shopify pages present the Unreal Keyboards tools separately: - [Dome Lab](https://unrealkeyboards.com/pages/dome-lab) - [Force Curve Bench](https://unrealkeyboards.com/pages/topre-force-curve-library) - [EC Parts Builder](https://unrealkeyboards.com/pages/topre-ec-parts-library-builder) - [Beyond Snap Ratio](https://unrealkeyboards.com/pages/beyond-snap-ratio) - [Topre & EC Dome Force Curves article](https://unrealkeyboards.com/blogs/topre-mods/topre-dome-force-curves) Shopify owns the Dome Lab landing page, header, and navigation. Force Curve Bench and EC Parts Builder are embedded independently. The EC Parts Builder itself contains no Dome Lab home, tabs for other tools, or cross-tool links. ## Force Curve Bench Force Curve Bench is an interactive viewer for measured Topre-compatible dome and assembly force curves. It lets you inspect one retained test set, compare several test sets on a common scale, and browse the generated measurements behind each curve. The bench is designed to answer practical comparison questions: which dome has the higher collapse force, which one loses force more sharply after collapse, where a force wall was detected, and where a dome falls within the tested reference fleet. It does not turn those measurements into a universal good/bad rating, and it does not treat a manufacturer's nominal weight as a measured result. ## Release status | Item | Status | |---|---| | Canonical public viewer | [https://buddyog.github.io/topre-force-curves/](https://buddyog.github.io/topre-force-curves/) | | Generator integration and regeneration parity | Complete | | Force Curve release tag | `fc-3.5` | | EC Parts Builder release | `lib-6.1`; tag `ec-parts-lib-6.1`; `release_eligible: true` | | EC Parts Builder prepublication record | `lib-6.1-review.1`; completed and superseded by `lib-6.1` | | Current public package authority | `SITE_RELEASE_MANIFEST.json` and its `SHA256SUMS` | | Protected predecessor authority | `FORCE_CURVE_BENCH_RELEASE_MANIFEST.json`, bound to tag `fc-3.4` | | Open Graph image | `https://buddyog.github.io/topre-force-curves/assets/force-curve-bench-fc-3.5-og.png` | | Browser acceptance | Local interactive/runtime/export acceptance and live deployment checks recorded | | Scientific continuity | All 184 raw CSVs, `canonical-evidence/`, the retained fleet, full-precision metrics, and `perception-rank-v1` remain unchanged from `fc-3.4` | | Frozen raw-source commit | `6e86ac1955a0c566c7aae521705e51371992ba8a` | | Canonical evidence identity | `7aa8588b50856816b7fce90dd6e743c26c6f16926071123291cb054352b6cd4a` | | Metric method | `metrics-v4.2` | | Intake authority | `intake-qc-v1.4`, implementation-parity bound to `test-imp 1.1.4` | | Perception-index model | `perception-rank-v1` | | Research preprint | Version 2.1; [paper DOI](https://doi.org/10.5281/zenodo.22295223); [all versions](https://doi.org/10.5281/zenodo.22167064); [compendium DOI](https://doi.org/10.5281/zenodo.22167047) | The site manifest records the active combined publication. The retained `FORCE_CURVE_BENCH_RELEASE_MANIFEST.json` remains the immutable historical record for the `fc-3.4` viewer and its evidence surfaces. The `fc-3.5` update changes presentation and navigation only. The frozen scientific payload, run membership, measurements, method, and index equations did not change. The `fc-3.5` site overlay replaces only the Force Curve Bench and its declared supporting files. The published `lib-6.1` EC Parts Builder, every raw CSV, and the complete canonical-evidence tree remain byte-identical to the preceding combined-site release. The deterministic site builder rejects undeclared or protected-path changes. The release adds a root `.gitattributes` containing exactly `* -text`. This disables Git text and line-ending normalization for every inventoried path so a checkout cannot silently rewrite sealed bytes on Windows or another platform. It also adds an empty root `.nojekyll`, which makes GitHub Pages publish the tree directly without Jekyll filtering underscore-prefixed paths. Both files are publication infrastructure, not raw-source files or part of the canonical scientific evidence. ## EC Parts Builder The published `lib-6.1` build is one row-based assembly tool. Its catalog feeds contextual part choosers rather than a separate library page. It exposes 11 component rows, including a top-level 2u stabilizer assembly and an independently selectable 2u silencing ring. It has 28 keyboard starters, 4 manufacturer-parts shortcuts, 10 recorded 2u assemblies, and 68 exact dome-specimen measurement mappings. Eligible measured domes carry collapse force, Weight Index, Tactility Index, and detected Force-Wall onset. The starting-point defaults reflect the owner's September 3, 2026 review. Manufacturer shortcuts intentionally load no dome, and the Realforce 106U, 108U, 89U, and 91U intentionally load no separate spacebar stabilizer because their short spacebars use the selected 2u assembly. Starting-point defaults are assembly aids, not compatibility verdicts. The lib-6.1 consumer payload contains 241 decision-changing pair edges after generic matrix-expanded edges for three part-scoped conical-spring findings are suppressed. The affected spring is flagged once on its own row, unrelated parts are not blamed, and the whole-build status inherits the finding; genuine pair-specific findings remain. The complete 3,570-edge configuration remains the audit authority outside the browser. Public results use **Works**, **Works with conditions**, **Does not work**, **Not verified**, and the declared **Manufacturer matched** fallback. Dome compatibility is outside this engine and is labeled **Not evaluated**. The historical lib-6.0 release exposed 9 component rows, 13 keyboard starters, and 338 decision-changing browser edges against its 3,403-edge audit closure. Its tag and release record remain available for exact historical citation. See the [EC Parts Builder guide](documentation/EC_PARTS_LIBRARY.md), [lib-6.1 prepublication acceptance record](documentation/EC_PARTS_LIBRARY_LIB61_REVIEW.md), [lib-6.1 release notes](documentation/RELEASE_NOTES_lib-6.1.md), [lib-6.0 release notes](documentation/RELEASE_NOTES_lib-6.0.md), and [Shopify embed contract](documentation/SHOPIFY_EMBED.md). ![Representative fc-3.5 Force Curve Bench](assets/force-curve-bench-fc-3.5-og.png) ## What the viewer can do - Plot the averaged press and return traces for a retained test set. - Use one shared fleet-wide absolute displacement axis for every curve comparison. The normal endpoint is 4.0 mm; this fleet uses 4.5 mm because one released force-wall marker is slightly beyond 4.0 mm. Recorded turnaround does not enlarge the axis. - Browse dome tests and part-assembly tests in separate **Domes** and **Parts** workspaces, and overlay up to 10 compatible force curves at once. - Use the same collapsed name-and-value filter system in **Domes**, **Parts**, and **Tests**. Filtered sidebar matches appear as a flat, left-aligned list; the unfiltered fleet remains organized under collapsible test groups. - Show weight-family and tactility-family comparison profiles on a shared percentile scale. - Compare detected force-wall onset across individual domes and assembly-family groups against the measured Topre production reference context. - Plot Weight Index against Tactility Index for calibrated domes. - Browse and sort all generated test records, with name, category, and numeric range filters. - Show exact chart readings with a mouse, touch screen, or pen. - Keep the current comparison in the page URL so it can be shared. - Export an evidence-stamped PNG of any available chart mode with a faint, centered **UNREAL KEYBOARDS** watermark and export date/time. - Choose the default **Simplified** force-curve display or the full **Detailed** drafting-style measurement overlay, and independently show or hide measurement labels and visuals. - Fall back to embedded curve snapshots if the pinned raw files cannot be read from the network. See the [Viewer Guide](documentation/VIEWER_GUIDE.md) for step-by-step use. ## Quick start 1. Open [Force Curve Bench](https://buddyog.github.io/topre-force-curves/). 2. Choose **Domes** or **Parts**, expand **Tests**, and select a test. Expand **Filters** to search by name or set numeric ranges. 3. Choose additional tests to compare. Switching between **Domes** and **Parts** clears an incompatible selection because the two record kinds are intentionally kept separate. 4. Switch among **Force curves**, **Weight profile**, **Tactility profile**, **Force-wall onset**, and **Weight vs tactility**. Opening a comparison tab with nothing selected loads the complete compatible fleet; use **Clear selection** or **Load all** to reset that comparison. 5. Hover over the chart for exact readings, or tap it on a touch device. 6. Copy the address bar to share the current selection, or choose **Export PNG** to save the current chart. The viewer limits direct force-curve overlays to 10 tests to keep the color-coded series and readout legible. Fleet-wide profile, force-wall, and scatter comparisons can load all compatible records. ## What the main measurements mean The viewer groups mechanical descriptors by the aspect of the pilot in which they had the strongest observed associations. That grouping is useful for comparison, but it does not prove that every descriptor independently causes a particular perception. ### Weight-family descriptors - **Collapse force** is the detected mechanical collapse peak, measured in gram-force (gf). - **Ramp** describes the pre-collapse force buildup over the baseline-relative 10–90% rise, after excluding the seating transient, in gf/mm. - **Pre-collapse work** is the one-way mechanical work integral from the start of the press to collapse, in gf·mm. ### Tactility-family descriptors - **Drop** is collapse force minus valley force, in gf. - **Steepest drop** is the fastest force loss over any 0.10 mm window between collapse and valley, in gf/mm. - **Drop rate** divides force drop by the collapse-to-valley distance, in gf/mm. - **Snap %** is force drop divided by collapse force, multiplied by 100. ### Detected force-wall onset Detected force-wall onset is an operational result from the complete measured assembly. It is not nominal dome height, physical switch travel, actuation travel, or a universal contact position. The viewer reports the measured onset and marks it on the curve. The **recorded turnaround range (test limit)** is separate. It only reports how far the retained acquisitions were recorded. It is never substituted for a force wall. If the wall rule is not satisfied, the viewer says **Not detected**. Displayed mechanical scalar values come from full-precision per-run results and the declared cohort-aggregation rule. The two perception indices use full-precision cohort aggregates and frozen reference arrays. Rounding occurs only for presentation. The displayed visualization trace is averaged separately, so a scalar marker does not have to land exactly on a feature of the averaged line. ## Weight and tactility indices **Weight Index** is the percentile of a dome's collapse force compared with the other domes tested, from 0 (lightest) to 100 (heaviest). **Tactility Index** is the percentile of a dome's force drop compared with the other domes tested, from 0 (least sharp) to 100 (sharpest). They are not predicted 1–10 ratings, universal perceptual units, or proof that one dome will feel the same to every person or in every keyboard. The full calculation is specified in [`documentation/METHOD.md`](documentation/METHOD.md#pilot-derived-perception-indices). The mechanical inputs were selected because collapse force and force drop had the strongest observed rank associations with the pilot's perceived-weight and tactility-sharpness scores. The pilot used one rater, 25 domes, and three fixed-order sessions. It contained no linear/off observation, so a low Tactility Index does not classify a dome as linear. Part-assembly records are outside the calibrated reference population and display **Not calibrated**. Force-wall onset is not part of either index. ## Evidence and trace sources The current frozen evidence contains: - 76 semantic test records; - 75 independent measurement cohorts; - 184 semantic raw-run bindings representing 180 unique acquisitions; and - 68 release-eligible dome-baseline cohorts in the index reference fleet. One group of acquisitions intentionally has both a dome interpretation and a part-assembly interpretation. Those records describe different meanings but are not independent physical tests. The viewer exposes record IDs, source sets, and measurement cohorts so this relationship is not hidden. At runtime the viewer first attempts to fetch retained raw CSV traces from the exact frozen commit. Here, “live” means a current read of immutable, commit-pinned bytes; it does not mean the latest mutable branch. If that read fails or times out, the viewer uses its embedded snapshot traces and displays a snapshot notice. Generated scalar measurements are the same in both modes and are never recomputed from the downloaded or embedded averaged curve. The selectable fleet is generated in advance. Uploading a new folder or changing a branch cannot silently add a test to an existing viewer build. ## Data acceptance in brief Raw runs are admitted through `intake-qc-v1.4`, which is bound to the accepted `test-imp 1.1.4` implementation. The intake sequence checks parsing and trajectory integrity, signal integrity, hard speed and cadence limits, landmark validity, mixed replicate-supported populations, replicate agreement, and a minimum of two matching retained runs. A retained-cohort RAMP review is advisory only; it does not alter membership or measurements. These checks are engineering intake guardrails. They are not manufacturer tolerances, formal instrument-uncertainty bounds, or perceptual thresholds. ## Status labels - **Not detected** — the force-wall rule did not find a wall. The viewer does not replace it with the last sample. - **Not available** — that measurement is missing or cannot be defined for the record. The viewer does not invent a value. - **Not calibrated** — a perception index is intentionally not assigned to a part-assembly record. ## Documentation - [Viewer Guide](documentation/VIEWER_GUIDE.md) — selecting, comparing, filtering, sharing, reading, and exporting. - [Measurement and calculation method](documentation/METHOD.md) — acquisition, intake, aggregation, metric, and index rules. - [Limitations and interpretation boundaries](documentation/LIMITATIONS.md) — what the measurements and indices do and do not establish. - [Data and provenance](documentation/DATA_PROVENANCE.md) — frozen identities, population counts, metadata lineage, aliases, and authoritative artifacts. - [Reproduction and verification](documentation/REPRODUCING.md) — package verification and conditional full regeneration. - [Browser acceptance](documentation/BROWSER_ACCEPTANCE.md) — interactive, runtime-battery, export, and responsive acceptance evidence. - [Future data updates](documentation/DATA_UPDATES.md) — the versioned path for adding tests, retests, corrections, or folder renames without silently changing a released dataset. - [Dome testing and new-dome intake SOP](documentation/testing-sop/README.md) — the frozen, plain-language workflow for collecting a new cohort, running the exact importer, handling exceptions, and handing it into a new evidence epoch. - [Third-party font notice](documentation/THIRD_PARTY_NOTICES.md) — embedded font credits and SIL Open Font License 1.1 terms. - [Subjective pilot](documentation/subjective-pilot/README.md) — protocol, source identity, observations, summaries, correlations, and limitations. - [fc-3.5 release notes](documentation/RELEASE_NOTES_fc-3.5.md) — current viewer presentation, comparison, filtering, export, and continuity changes. - [fc-3.4 release notes](documentation/RELEASE_NOTES_fc-3.4.md) — historical predecessor viewer and publication identity. - [EC Parts Builder guide](documentation/EC_PARTS_LIBRARY.md) — standalone workflow, contextual catalog, Evidence-beta labels, and interpretation limits. - [lib-6.1 release notes](documentation/RELEASE_NOTES_lib-6.1.md) — current Parts Builder changes, release identity, and scientific-continuity boundary. - [lib-6.1 prepublication acceptance record](documentation/EC_PARTS_LIBRARY_LIB61_REVIEW.md) — completed 2u-assembly, starter, catalog, and acceptance scope. - [lib-6.0 release notes](documentation/RELEASE_NOTES_lib-6.0.md) — historical release chronology, target identity, changes, limitations, and verification contract. - [Shopify embed contract](documentation/SHOPIFY_EMBED.md) — exact-origin parent/child messaging and production acceptance checks. - [Public release record](documentation/PUBLIC_RELEASE.md) — final public URLs, released identities, paper hashes, licensing boundary, and rollback retention. - [Citation guide](documentation/CITATION.md) — citations for the paper, compendium, tools, and measurement records. - [Prospective multi-rater study](documentation/prospective-study/README.md) — frozen `perception-validation-v1` protocol, plain-language SOP, participant script, analysis plan, data templates, and blinded schedule generator. - [Dome testing SOP v1 freeze notes](documentation/RELEASE_NOTES_dome-testing-sop-v1.md) — importer identity, operating boundaries, and release continuity. - [Full-paper roadmap](documentation/research-paper/README.md) — how Preprint 2.1, the exploratory pilot, and future validation results become the full manuscript without mixing evidence classes. - [Perception-validation v1 freeze notes](documentation/RELEASE_NOTES_perception-validation-v1.md) — frozen design, operator materials, readiness boundary, and scientific continuity. ## Scope and limitations - Results describe the tested specimen and measured assembly. Identical-looking domes can differ. - A family average is a visualization aid, not an independently measured dome record, and therefore has no generated scalar measurements. - Individual retained runs are preserved in the evidence, but the consumer viewer treats the averaged retained set as the displayed result. - Percentile indices are relative to a frozen fleet. A future fleet or method version requires a new versioned reference and may change ranks. - This tool supports comparison; it does not recommend a universally “best” dome. ## Rights and reuse © 2026 Brian “BuddyOG” Gebo — Unreal Keyboards. All rights reserved. The owner permits sharing unmodified charts exported by Force Curve Bench and unmodified data files published with the repository when the share includes attribution to **BuddyOG / Unreal Keyboards** and a link to the repository or the canonical live viewer. This is limited sharing permission, not an open license or a broader grant for the software, documentation, or data. It does not permit relicensing or imply endorsement. Inter and IBM Plex Mono are embedded third-party fonts licensed separately under the SIL Open Font License 1.1. See the [third-party font notice](documentation/THIRD_PARTY_NOTICES.md). Their licenses do not change the project terms above. ## Credits - **Measurement-rig design:** [bluepylons/Open-Switch-Curve-Meter](https://github.com/bluepylons/Open-Switch-Curve-Meter). - **Testing, data stewardship, analysis, and Force Curve Bench:** Brian “BuddyOG” Gebo / [Unreal Keyboards](https://unrealkeyboards.com). The released EC Parts Builder and Force Curve Bench consume the same fc-3.4 canonical evidence without changing it. Beyond Snap Ratio Preprint 2.1 records the released methods and exploratory pilot. The prospective blinded multi-rater protocol is frozen as `perception-validation-v1`; until the study is completed, no confirmatory multi-rater perception claim is made.