# Force Curve Bench data and provenance > Provenance documentation for Force Curve Bench `fc-3.5`. The canonical viewer is > [https://buddyog.github.io/topre-force-curves/](https://buddyog.github.io/topre-force-curves/) > and the release tag is `fc-3.5`. Exact current-package hashes are recorded by > `SITE_RELEASE_MANIFEST.json`, `SHA256SUMS`, and > `FORCE_CURVE_BENCH_RELEASE_MANIFEST_fc-3.5.json`. > `FORCE_CURVE_BENCH_RELEASE_MANIFEST.json` remains the immutable > historical `fc-3.4` record. EC Parts Builder > `lib-6.1` is published from tag `ec-parts-lib-6.1`. The active > `SITE_RELEASE_MANIFEST.json` and `SHA256SUMS` inventory cover the combined > site while retaining the `fc-3.4` manifest as a predecessor record. The Force Curve Bench is a presentation layer over a frozen evidence epoch. Its source measurements, retained-run membership, full-precision results, and metadata are tied to one Git commit and one independently verifiable evidence identity. Viewer changes do not silently change that evidence. ## Frozen evidence identity | Property | Frozen value | |---|---| | Repository | `BuddyOG/topre-force-curves` | | Git commit | `6e86ac1955a0c566c7aae521705e51371992ba8a` | | Git root tree | `e43d25a3fed8c5177467bba96a5abbdd5a3442b0` | | Canonical epoch | `canonical-evidence-2026-08-15-6e86ac19` | | Canonical evidence identity | `7aa8588b50856816b7fce90dd6e743c26c6f16926071123291cb054352b6cd4a` | | Evidence package | `canonical-evidence-epoch-6e86ac19-step2` | | Metrics method | `metrics-v4.2` | | Intake policy | `intake-qc-v1.4` | | Importer parity authority | `test-imp 1.1.4` | The Git commit identifies the repository snapshot. The Git tree identifies the complete directory tree at that commit. The evidence identity seals the canonical inputs, decisions, and generated evidence surfaces. These identities serve different purposes and should be reported together. ## Population and count definitions Several correct counts describe different layers of the dataset. They must not be treated as interchangeable. | Count | Meaning | |---:|---| | 189 | All files in the frozen repository snapshot. | | 184 | Raw CSV paths in that snapshot. All 184 are retained. This is also the number of semantic run bindings used by the viewer. | | 180 | Unique physical acquisitions after shared evidence is counted once. | | 76 | Semantic records shown by the generated dataset. A semantic record is one stated interpretation of evidence, not necessarily an independent physical test. | | 75 | Independent measurement cohorts after the shared-evidence alias is counted once. | | 76 | Derived curve packs, one for each semantic record. | | 68 | Release-eligible dome-baseline cohorts in the frozen reference fleet used to calibrate the two perception indices. Part-assembly records are excluded from this calibration population. | ### Why 184 paths become 180 acquisitions Four raw acquisitions appear under both `Topre_HHKB_Pro2_45g` and `Topre_Slider_Black`. They are one physical test with two legitimate semantic interpretations: the assembled HHKB Pro2 test and the black-slider parts test. The epoch therefore keeps both semantic records and all eight path bindings, while counting the four underlying acquisitions only once for independence. This shared-evidence ruling produces: - 76 semantic records; - 75 independent measurement cohorts; - 184 retained raw paths or semantic run bindings; and - 180 unique physical acquisitions. The two records may be displayed separately, but they must never be described as eight independent observations. ## Publication-tree continuity ### Historical fc-3.4 publication The fc-3.4 publication tree keeps 187 of the frozen snapshot's 189 files byte for byte: all 184 raw CSV files at their existing repository-relative paths, plus three legacy HTML endpoints for URL continuity. The prior root `index.html` and `README.md` are intentionally replaced by the fc-3.4 viewer and public overview. The three preserved HTML endpoints are not part of the fc-3.4 viewer validation surface and were not scientifically revalidated by this release. Their preservation is a compatibility decision. The final 752-file repository-root inventory, including those 187 preserved files, is sealed by `FORCE_CURVE_BENCH_RELEASE_MANIFEST.json` and `SHA256SUMS`. ### Historical lib-6.0 EC Parts Builder overlay The `lib-6.0` release replaces `dome-lab-parts.html` with the generator-owned standalone EC Parts Builder. Shopify owns the Dome Lab home and navigation; the builder contains no in-tool tabs or sibling-tool links, and its catalog is contextual chooser data. The released builder projects 9 component rows, 13 source-backed keyboard starters, 68 exact dome-specimen measurements carrying measured collapse force plus released Weight Index and Tactility Index values, and 2 public keycap additions. Its browser payload contains 338 decision-changing compatibility edges; the complete 3,403-edge configuration remains the audit authority. Public compatibility labels are **Works**, **Works with conditions**, **Does not work**, and **Not verified**; domes are **Not evaluated**. That one endpoint is intentionally no longer byte-identical to the fc-3.4 predecessor. The overlay keeps every other protected predecessor endpoint, all 184 raw CSV paths, and the complete canonical-evidence tree byte-identical to fc-3.4. The overlay policy declares every source path that may be introduced or replaced. The release verifier restricts generated differences to the EC Parts Builder presentation/provenance set and rejects changes to the Force Curve viewer or scientific evidence. `SITE_RELEASE_MANIFEST.json` plus `SHA256SUMS` are the active combined-site authority; the fc-3.4 release manifest remains the protected predecessor record. ### Published lib-6.1 EC Parts Builder overlay The `lib-6.1` release supersedes lib-6.0 at `dome-lab-parts.html` without altering the protected fc-3.4 scientific surfaces. Its complete compatibility closure contains 3,570 edges, and its browser projection contains 241 decision-changing pair edges. It omits 97 generic matrix-expanded edges that repeat three part-scoped conical-spring findings: Deskeys and KLC are part-scoped **Does not work**; MetaPulse is part-scoped **Not verified**. The consumer displays each issue once on the culprit spring, leaves unrelated parts unflagged, and carries the part issue into the whole-build result. Pair-specific findings not subsumed by a part issue remain in the projection. The release also publishes 11 component rows, 28 keyboard starters, 4 manufacturer-parts shortcuts, 10 recorded 2u assemblies, and Force-Wall / Travel / Dome compression presentation for eligible measured domes. The controlled overlay keeps every other protected predecessor endpoint, all 184 raw CSV paths, and the canonical-evidence tree byte-identical to fc-3.4. The historical `ec-parts-lib-6.0` tag retains its 338-edge consumer projection and 3,403-edge audit record. ### Current fc-3.5 Force Curve Bench overlay The `fc-3.5` release replaces the root Force Curve Bench and its declared viewer-generated/supporting files on top of the `lib-6.1` combined site. It does not create a new scientific epoch: the raw-source commit, 184 raw CSV paths, canonical-evidence tree, retained cohort, full-precision mechanical results, 68-dome percentile reference arrays, and `perception-rank-v1` scores remain unchanged. The overlay preserves both `dome-lab-parts.html` and `generated/packs/picker.staged.html` byte for byte at their published `lib-6.1` SHA-256. It rebuilds the active generated/site inventories around the new viewer rather than promoting an unrelated regenerated picker provenance change. `FORCE_CURVE_BENCH_RELEASE_MANIFEST_fc-3.5.json` records this component boundary, and the complete `SITE_RELEASE_MANIFEST.json` plus `SHA256SUMS` remain the active combined-site authority. Three files in the sealed subjective-pilot presentation bundle receive a documented prose-only correction: `README.md` and `index.html` now state the actual 25-assembly fixed-grid apparatus and describe `observations.csv` as 75 session-by-dome rows containing 150 numeric ratings; the directory's `SHA256SUMS.json` is resealed to those bytes. The observations, grid mapping, dome summaries, analysis, source identity, workbook hash, and all released index values remain unchanged. The publication root also adds `.gitattributes` with exact content `* -text`. This disables Git text and line-ending normalization so platform-specific checkouts do not rewrite sealed bytes. The file is publication byte-preservation infrastructure; it is not part of the frozen raw-source snapshot or canonical scientific evidence. An empty root `.nojekyll` is also added so GitHub Pages publishes the exact static tree without Jekyll filtering underscore-prefixed paths. It too is release infrastructure, not source evidence. The raw-source commit and Git root tree above continue to identify the scientific input snapshot. Later publication commits and tags identify expanded release trees. Keeping those identities distinct prevents a viewer or documentation update from being mistaken for a new raw-data epoch. ## Metadata authority The owner-supplied workbook contains 79 metadata rows: - 65 rows marked `YES`, which are authoritative for the corresponding tested records; - 14 rows marked `NO`, which remain documented but are not treated as tested records in the canonical fleet; and - 11 additional released semantic records that are not represented by a workbook `YES` row. Those 11 records retain explicit predecessor-dataset lineage. Spreadsheet values were not invented to fill the gap. The 65 workbook-authoritative records plus the 11 predecessor-lineage records produce the 76 semantic records. The frozen workbook is `epoch_inputs/source_workbook/untested-domes_v2.xlsx`, with SHA-256 `02e85a2fbdc8e025f1445ab8cb7bd830501f755b79d0a755d6ee31e53bca45a4`. Its normalized metadata and reconciliation records are preserved alongside the source workbook. ## Provenance chain The evidence can be followed from source files to the viewer: 1. **Frozen repository snapshot.** The exact commit is materialized into an immutable raw cache. The repository inventory records each path, byte count, SHA-256 digest, and Git blob OID; the root tree is reconstructed recursively. 2. **Metadata and history.** The workbook extraction, normalized metadata, predecessor lineage, exclusions, retests, and evidence-event ledger preserve where every released record came from and what happened before the frozen epoch. 3. **Intake and retention.** The retained-run decision artifact binds every raw path to its semantic record and records intake decisions under `intake-qc-v1.4`, checked independently against `test-imp 1.1.4`. 4. **Canonical calculations.** Per-run and aggregate measurements are generated at full precision under `metrics-v4.2`. 5. **Presentation derivatives.** Curve packs and the single-file viewer are generated from the canonical data. They support visualization; they do not replace the full-precision authority. At the frozen epoch, all 184 raw paths are retained. Canonical intake reports no mixed-population stop, no required ramp review, and no unresolved membership warning. Historical exclusions and nine historically pruned runs remain in the lineage record; they are not silently reintroduced as frozen raw evidence. ## Which artifacts are authoritative? Paths below are relative to the root of the sealed canonical-evidence package. | Artifact | Role | |---|---| | `EPOCH_MANIFEST.json` | Package identity, frozen counts, component identities, and canonical artifact map. | | `SHA256SUMS.csv` | Complete package inventory and SHA-256 checksums. | | `epoch_inputs/manifest/raw_file_inventory.json` | Raw path, blob, byte-count, and hash authority. | | `epoch_inputs/metadata/dome_metadata_registry.json` | Reconciled record metadata and its source. | | `epoch_inputs/history/exclusions_and_retests.json` | Exclusion, replacement, and retest history. | | `evidence/intake_retention_decisions.json` | Retained-run membership and intake-decision authority. | | `evidence/per_run_full_precision.json` | Authoritative full-precision per-run results. | | `evidence/bench_tests.staged.json` | Authoritative full-precision aggregate results. | | `evidence/generated_manifest.json` | Generated evidence inventory and hashes. | | `evidence/curve_pack_provenance.json` | Provenance from canonical records to display curve packs. | | `evidence/packs/curves/*.staged.json` | Derived, lossy display curves; not numeric measurement authority. | `bench_tests.staged.json` and `per_run_full_precision.json` are the numeric authorities. Human-readable CSV surfaces are useful for inspection, but the sealed epoch designates the JSON artifacts above as authoritative. ### Key artifact hashes These SHA-256 values identify the sealed Step-2 evidence artifacts incorporated into fc-3.4. They are not hashes of the production viewer or public release package; use the packaged `FORCE_CURVE_BENCH_RELEASE_MANIFEST.json` and `SHA256SUMS` for those values. | Artifact | SHA-256 | |---|---| | `EPOCH_MANIFEST.json` | `a1837f802de436035d605e9971563147308e86e7fc089902b30bb22e0581d354` | | `SHA256SUMS.csv` | `726b9b84dd34b94880cb7d95660c40c7a686f336d73198748800cd48da9caac3` | | `epoch_inputs/manifest/raw_file_inventory.json` | `3652c2db9a2cf7f464aac53dca40cba6ad214f5cbf3bb1c8963c968cf09b5ca3` | | `epoch_inputs/metadata/dome_metadata_registry.json` | `3531da02a466ad335bddba5f78fd1c9782e564e5fe488605a0a7500a6194eefb` | | `epoch_inputs/history/exclusions_and_retests.json` | `f42b4ad91201bd8be93b687d6b2a9a51450c0dae477c38e1a89a313032c44d32` | | `evidence/intake_retention_decisions.json` | `e6d438b672408cd3e711d907e58424af1f73c13650674526900ce24ceeea9bf4` | | `evidence/per_run_full_precision.json` | `28c2d6dbc4391a69b637223bcef3b68429a7c31cbcfd2754c631f0a2dd7d52ee` | | `evidence/bench_tests.staged.json` | `6d2d482921a6b2de698841287f5ba416613b6f69c471b1124d26ab0887269b6e` | | `evidence/generated_manifest.json` | `b50d04604e639e19e9ace5ccc5426ceb9deffe75768a34ff61838dfc1a1c880c` | | `evidence/curve_pack_provenance.json` | `0fe1b9195c71111e578e264469f6de2a381ad49a7241f3b2878ced8f87384888` | The complete per-file inventory remains `SHA256SUMS.csv`; this abbreviated table is provided for convenient citation. ### Full precision versus curve packs Canonical scalar calculations retain full precision. Rounding is a display operation performed by the viewer. Curve packs are intentionally resampled and compact, so they are lossy by design. They are suitable for drawing traces, but not for recalculating, replacing, or auditing the canonical scalar values. The viewer follows the same separation online and offline: - raw files from the pinned commit can provide live visualization traces; - the embedded snapshot provides fallback traces if the pinned files cannot be reached; and - displayed scalar measurements come from the generated canonical record in either mode. An embedded-snapshot fallback does not authorize client-side recomputation of the released measurements. ## Perception-index provenance `perception-rank-v1` is an exploratory presentation model layered on the frozen evidence; it does not modify the canonical epoch. The Weight Index ranks full-precision collapse force within 68 released dome-baseline cohorts. The Tactility-sharpness Index ranks full-precision force drop within the same 68-cohort fleet. Collapse force and force drop were chosen because they had the strongest observed rank associations in a one-rater, 25-dome, three-session pilot. The pilot selects the mechanical inputs; it does not transform them into predicted 1–10 ratings. Each index is the percentile compared with the other domes tested. The reference fleet, percentile equations, tie handling, interpolation, clamping, and null rules are fixed by the [index specification](METHOD.md#pilot-derived-perception-indices). Part assemblies receive no index, and force-wall onset is not an input to either index. ## Verification The sealed evidence package includes a standard-library Python verifier. It checks exact package membership, hashes, path safety, raw-cache and Git-tree closure, metadata and alias consistency, retention closure, importer parity, and the absence of generated viewer HTML from the Step-2 evidence output. See [REPRODUCING.md](REPRODUCING.md) for the distinction between verifying a released package and regenerating the viewer from source.