UNI-FLAGELLUM Living Science Walkthrough v0.3
How to read this page
Three ways to read this page. Precise is the document itself, exactly as it is written in the repository. Plain and Clear were written for this website to help you meet that document — they are about it. They are not it, and they are not evidence.
A laboratory built around the bacterial flagellar motor. It holds a deterministic reduced model of the motor, analysis of recorded single-motor events, and a cross-study parity programme. Alongside those sit the scientific gates the work has to clear, and independent audits of both the model and the repository around it. The framing throughout is hierarchical active inference.
It is for a reader with a scientific interest, and especially for one who has come to check whether a model fit has quietly become a claim about biology. The laboratory's central discipline is a labelling one: every visible layer carries exactly one class — recorded observation, structural reconstruction, reduced model, or physical teaching analogue — and those classes may not be blended. Behavioural observations of one species are held apart from structural work on another, so that nothing on the page can read as a single measured specimen.
Start with the Living Science Walkthrough, which sets out those classes and the truth contract they belong to. Then the scientific and mathematical contract, then the parity gates, which state what would have to hold before a parity claim could stand.
What it is not: a claim of biological parity. The walkthrough is explicit that the release does not turn a model fit into a biological identity claim, and full biological parity is recorded as false and printed as false. Passing this repository's software tests is necessary here and is not the same thing as agreement with a living motor.
Your browser cannot switch reading levels, so the document itself is shown.
Precise — the source document
This is the document. Rendered from the repository at the commit above, with nothing rewritten for the web. A gate re-renders it on every deploy and fails the build if a single byte differs.
Purpose and truth contract
This release is a public, self-guided lesson built on the existing deterministic motor model, observed-event analysis, cross-study corpus, gate ledger, and CAD analogue. It does not turn a model fit into a biological identity claim.
Every visible layer has exactly one production class:
| Label | Meaning | Allowed example | Prohibited presentation |
|---|---|---|---|
OBSERVED |
Source pixels or recorded instrument/experimental fields | Mears E. coli fluorescence; Wadhwa stator event fields | A generated or reconstructed frame |
STRUCTURAL RECONSTRUCTION |
Geometry constrained by deposited structures or a structural study | Salmonella basal body; Bacillus MotA5MotB2 homologous geometry | A same-specimen E. coli measurement |
REDUCED MODEL |
An explicit deterministic calculation | Categorical Bayes update, synthetic swimming world | Hidden biological ground truth |
UNI PHYSICAL ANALOGUE |
A teaching mechanism that realizes declared math | Log-odds gear addition | Bacterial anatomy or a working motor replica |
The stage therefore keeps E. coli behavioural observations, Salmonella basal-body evidence, and Bacillus MotA/B geometry labelled by species. They are not merged into an imaginary measured specimen.
Runtime modes
OBSERVED_REPLAYreplays frozen, source-pinned fields. The Wadhwa lesson explicitly lists ligand, speed, rotation, load, PMF, and CheY-P as missing from the selected event records.SYNTHETIC_WORLDruns the existing CPU-only deterministic world and UNI generative model. World truth remains outside the model boundary.LIVE_INSTRUMENTuses the existing validated newline-delimited serial-frame path and remains unavailable until a serial device is connected.
No LLM, model API, WebGL, WebGPU, Three.js, analytics, account, network data submission, or GPU computation is present. Optional narration is fixed authored text spoken by the browser's local speech synthesis only after the visitor asks.
The thirteen-step lesson
| Step | Camera and source | Human activity | Pencil-and-paper reproduction |
|---|---|---|---|
| 0 Truth before beauty | Cell; observed replay | Classify every layer by production class | Place claims in observed/reconstruction/model/analogue boxes |
| 1 Watch life move | Mears microscopy | Predict bundle breakup and mark run/tumble times | Mark state intervals from source timestamps |
| 2 Reconstruct the cell | Cell; synthetic world | Compare the labelled animation with source pixels | 12 µm / 0.6 s = 20 µm s^-1 |
| 3 Enter the motor | Singh structural video and cutaway | Locate filament, hook, rod, rings, and stators | Draw the mechanical transmission path |
| 4 Turn ions into torque | Motor; synthetic world | Change PMF/load before observing speed | W = τΔθ; 700 pN nm over one turn is 4.398e-18 J, about 1062 kBT at 300 K |
| 5 Switch direction | Bundle; observed replay | Predict the bundle consequence of CW/CCW switching | If three motors independently have CW bias 0.1, P(all CCW)=0.9^3=0.729 |
| 6 Establish the boundary | Inference mirror | Decide which fields cross inward and outward | Write o_t = sensor(world_t) and the bounded action map |
| 7 Update belief | Inference mirror | Commit a prior before evidence | O_post = O_prior LR; 2*3=6, hence P=6/7 |
| 8 Predict, then confront | Inference mirror | Record prediction before later observation | Residual e=o-o_hat; compare proper log scores |
| 9 Test recorded motors | Frozen Wadhwa holdout events | Choose a model, then score the held-out event | Score a frozen event without treating time points as motors |
| 10 Test across studies | Cross-study ledger | Inspect study units and reject pseudoreplication | Recreate one reported RMSE, slope, or effect from the frozen table |
| 11 Keep the falsifier | Inference mirror | Explain the incompatible lattice estimates | Compare fitted J=0.211 with moment J=1.188, SSE, and delta-AIC |
| 12 Reproduce and continue | Gate ledger and CAD link | State a conclusion, confidence, and alternatives | Print the worksheet and run the command recipe |
The worked values are generated by paperExampleResults() in
lib/walkthrough.js; the test suite independently recomputes every value.
Evidence identities
experiments/walkthrough-evidence-manifest.v1.json pins the file size, DOI,
license, species, truth class, and SHA-256 of every local source asset. The
production test hashes the files again.
| Asset | Local identity | Scope and fence |
|---|---|---|
Mears et al. Video 1, DOI 10.7554/eLife.01916.010, CC BY 4.0 |
public/media/mears-2014-run-tumble.mp4 |
Direct E. coli fluorescent flagella behaviour |
Singh et al. Supplementary Video 4, DOI 10.1038/s41564-024-01674-1, CC BY 4.0 |
public/media/singh-2024-switching.mp4 |
Structural switching visualization for Salmonella |
| PDB 7E82, CC0 | public/data/structures/7e82.cif.gz |
Salmonella rod/hook/basal-body structural constraint |
| PDB 6YSL, CC0 | public/data/structures/6ysl.cif.gz |
Bacillus subtilis MotA5MotB2 homologous geometry |
| Wadhwa derived holdout events | public/wadhwa-2022-derived-events.json |
Motor-identified stator events; raw source hash remains separately recorded |
Primary source links:
- https://doi.org/10.7554/eLife.01916.010
- https://www.nature.com/articles/s41564-024-01674-1
- https://www.rcsb.org/structure/7E82
- https://www.rcsb.org/structure/6YSL
- https://www.rcsb.org/pages/usage-policy
Observer audit trail
The browser stores only ObserverRecord objects in local storage. Each record
contains session and step IDs, timestamp, source mode and truth class, input
state, prior prediction, observed values, calculation, interpretation,
alternative explanation, confidence, application commit, model run ID, gate
IDs, evidence IDs, and dataset hashes.
LessonExport contains the exact walkthrough manifest, records, gate counts,
evidence hashes, and reproduction commands. JSON and CSV exports, print output,
JSON import, and an in-app export/re-import/validate round-trip are available.
validateLessonExport() rejects a record that labels a synthetic source as
observed. No name, identity, account, submission, or analytics field exists.
Scientific result retained
The walkthrough consumes the signed production reports; it does not duplicate their fitting code or rewrite an adverse result for presentation. Current release truth remains:
- the original science program passes 4 of 7 executable computational gates;
- the cross-study program has 8 PASS, 3 FAIL, 2 NOT ESTABLISHED, and 3 BLOCKED EXTERNAL gates;
- the lattice full-distribution fit gives
J=0.2106, while the unweighted moment fit givesJ=1.1876; delta-AIC for cooperativity is-1.4588; - full biological parity is false; no universal, causal, or biological Active Inference identity was proved.
These failures are visible in step 11, step 12, the gate trace, and the expert research desk.
Reproduction
npm ci
npm test
npm run lint
npm run science:verify
npm run cross-study:verify
When the 4.09 GB archived Ito source bundle is present:
npm run cross-study:verify-raw
Run locally:
npm run dev
Open http://localhost:3000, complete all thirteen steps, save at least one
record, export JSON, import it again, run the round-trip validator, inspect the
retained failed gate, and print the worksheet.
Accessibility and release gate
The walkthrough uses native buttons, labels, ranges, select, video controls,
details, file input, progress, and text areas; the Canvas2D views have accessible
names. Captions are permanent, speech is optional and off by default, controls
are at least 44 CSS pixels high in guided mode, motion can be paused, and
prefers-reduced-motion freezes nonessential motion. Layouts reflow to one
column at small widths and print without the animated or expert surfaces.
Release requires unit tests, rendered HTML tests, lint, production build, independent science verification, independent cross-study verification, keyboard/touch/mobile browser checks, zero console errors, locally served media, notebook round-trip, and a clean-clone reproduction.
The ignored upstream study cache is intentionally not part of a Git clone. A
clean clone validates its complete immutable manifest and every tracked derived
artifact; when the all-or-none cache is present, npm test hashes all twelve
cached study artifacts. cross-study:verify-raw separately checks the 4.09 GB
Ito archive, including all 505 member CRCs.
sha256 0e3a715f79316d6a — of the original file, so what was ingested stays checkable.
Plain — written for this website, not the source document
This page describes a self-guided lesson about how a bacterium swims, and about how to tell the difference between what was measured and what was modelled.
Everything on the lesson's screen carries one of a small number of labels. Some things are recorded source material. Some are a structural reconstruction built from deposited structures. Some are a calculation that gives the same answer every time. Some are a teaching mechanism that acts out declared mathematics. The lesson never merges them, and it keeps evidence from different species apart rather than implying one measured specimen.
The lesson runs in thirteen steps. At each one the reader writes down a prediction before seeing the evidence, then records what happened, what it might mean, and what else could explain it. Everything is worked with pencil and paper as well as on screen.
Near the end the page repeats the project's own bad news instead of tidying it away: several gates fail, full biological parity is false, and no identity claim was shown.
Plain · written 2026-08-01 by claude-opus-5 · not yet checked by a person · about the document whose sha256 is 0e3a715f79316d6a
Clear — written for this website, not the source document
The page describes a public lesson built on parts that already exist: a motor model that gives the same answer every time, an analysis of recorded events, and a corpus of studies. Also a list of gates, and a design file for a physical analogue. Its first promise is that it does not turn a model fit into a biological identity claim.
The truth contract comes first, as a table. Each visible layer gets exactly one class. Recorded source pixels and instrument fields are one class. Geometry constrained by deposited structures is a structural reconstruction. An explicit calculation, repeatable every time, is a reduced model. A teaching mechanism that realizes declared mathematics is a physical analogue. For each class the table also names the presentation that is prohibited, so a generated frame may never appear where recorded material belongs, and the physical mechanism may never be shown as anatomy or as a working replica. Behavioural evidence from one species, structural evidence from a second and homologous geometry from a third are kept labelled by species rather than merged into an imaginary measured specimen.
Three runtime modes follow. One replays frozen, source-pinned fields and lists explicitly which quantities are missing from those records. One runs the repeatable synthetic world, with world truth kept outside the model boundary. One is for a live instrument and stays unavailable until a device is connected. The page also states what is absent from the release: no model inference, no graphics acceleration, no analytics, no account, no network submission. Narration is fixed authored text, spoken locally, only when asked for.
The lesson itself is a table of thirteen steps. Each row names the camera and source, the human activity, and a pencil-and-paper reproduction of the same quantity. The reader classifies every layer, watches recorded cells move, turns ions into torque, and draws the boundary between what crosses inward and what crosses outward. Then a prior before evidence, a prediction recorded before the later observation, and a score for a frozen event kept back from the fitting. Then study units inspected for pseudoreplication, an incompatible pair of estimates explained, and a conclusion stated with its confidence and alternatives. The worked values are generated by the code, and the test suite recomputes every one of them independently.
Evidence identities are pinned next: size, identifier, licence, species, truth class and checksum for every local source asset, re-hashed by the production test, each listed with the limit that applies to it.
The audit trail is local only. The browser stores records that carry the step, the timestamp, the source mode and truth class, the prior prediction, the observed values, the calculation, the interpretation, an alternative explanation, a confidence, and the relevant hashes. The lesson can be exported, printed, re-imported and validated in a round trip, and the validator rejects any record that labels a synthetic source as observed. No name, account or analytics field exists.
The section on retained results is where the page refuses to flatter itself. The original programme passes four of its seven executable computational gates. The broader programme has a mixture of passes, failures, not-established rows and externally blocked rows. Two ways of fitting the same lattice give incompatible interaction strengths, and the model-comparison statistic does not favour the extra parameter. Full biological parity is false, and no universal, causal or biological identity was shown. Those failures stay visible inside the lesson, not confined to an appendix.
The page closes with reproduction commands, a note that one very large archived source bundle is checked separately when present, and an accessibility and release checklist.
Clear · written 2026-08-01 by claude-opus-5 · not yet checked by a person · about the document whose sha256 is 0e3a715f79316d6a