Each focused lab has an extensionless URL, an AJAX workflow, bounded examples, explicit human and machine views, a defensive transform, and a clear limitations section.
The 40 local laboratories compare human-visible and machine-visible representations across Unicode, encodings, linguistic signals, model-facing patterns, HTML, documents, media, messaging, supply-chain artifacts, transformations, and controlled AI boundaries.
Inspect zero-width characters and format controls, encode a small benign payload, decode it, and compare security-oriented normalization views.
Compare visually similar characters from different scripts, construct a small educational skeleton, and flag mixed-script identifiers.
Expose the logical sequence, rendered sequence, escaped sequence, and code-point sequence for text containing bidi controls.
Reveal ordinary spaces, tabs, line endings, non-breaking spaces, thin spaces, trailing whitespace, and repeated-space patterns.
Compare raw UTF-8, code points, an explicitly approximate grapheme view, four bounded normalization copies, scripts, bidi controls, selected default ignorables, and an educational confusable skeleton without overwriting the original.
Encode and decode Base64, Base64URL, hexadecimal, percent encoding, HTML entities, JSON escapes, Unicode escapes, binary, and UTF-8 byte views.
Extract first or last letters from lines, sentences, and paragraphs, or inspect an nth-word positional rule.
Map word lengths to binary classes, inspect periodicity, and see how small lexical changes corrupt decoding.
Inspect serial commas, contractions, quote and dash variants, and selected US/UK spelling alternatives.
Use a fixed curated dictionary to encode or decode a small benign bit string and observe how paraphrase-like substitution changes the result.
Compare fixed active/passive, clause-order, and optional-conjunction realizations with simple tree-like structural views.
Map fixed harmless categories such as animal, fruit, vehicle, and weather to two-bit symbols and inspect category sequences.
Inspect prepared benign text for acrostic, fixed-position, word-length, punctuation, lexical, and transformation indicators using only the fixed marker MACHINE_TRADECRAFT_DEMO_OK.
Compare characters, UTF-8 bytes, whitespace-aware pieces, and a documented local toy subword segmentation.
Apply a synthetic preference to a fixed candidate vocabulary and visualize how individual normal-looking choices create an aggregate signal.
Partition a synthetic vocabulary with a local seed, generate a harmless token sequence, and calculate a simple green-token z-score.
Analyze repetition, word-length periodicity, unusual whitespace, character classes, runs, and simple n-gram regularity.
Compare deterministic miniature character, byte, BPE, WordPiece, and Unigram-style profiles, then inspect local round trips and bounded chunks. The bundled vocabularies do not reproduce a vendor tokenizer.
Compare raw source, structural text, likely rendered text, comments, hidden nodes, attributes, and a sanitized representation without executing the HTML.
Extract alt text, aria-label, aria-labelledby, titles, hidden labels, and visible text into separate provenance-labelled views.
Parse JSON or XML metadata, inventory field provenance, and compare strict, balanced, and research allowlist profiles.
Generate downloadable JSON, XML, HTML, or Markdown fixtures containing only a fixed benign marker and descriptive metadata.
Upload a PDF for bounded, non-executing inspection of the header, object count, metadata strings, XMP, annotations, embedded-file indicators, and active-content indicators.
Treat Office files as ZIP containers, list bounded package entries, and extract common core, app, and custom properties without macros or active execution.
Inspect PNG textual chunks and common JPEG EXIF fields, compare file properties, and identify metadata-only strings.
Parse a bounded inert HTML fragment and compare raw source, structural text, likely rendered text, accessibility fields, JSON-LD blocks, hidden mechanisms, and sanitized output.
Inspect a PDF, OOXML, PNG, or JPEG as inert bytes, identify its format, compute a digest, map structural features, and report active-content and ambiguity indicators.
Compare deterministic fixture DOM, accessibility snapshot, focus order, geometry, hidden state, locator strategy, and confirmation requirements without external navigation or real side effects.
Compare raw, human-visible, structural, escaped, metadata-like, and canonical views of the same text or HTML artifact.
Apply trim, whitespace collapse, normalization, HTML-to-text, punctuation, case, and paraphrase simulations to fixed benign carriers.
Record explicit maturity selections for nine independent dimensions, preserve one bounded evidence note, and generate a transparent local action plan without an overall score or certification claim.
Run one deterministic fictional representation-layer exercise, reveal bounded injects, record categorized decisions, preserve evidence requirements, identify missing authority, and export a non-scoring runbook.
Map one system profile across thirteen source-bound controls, nine maturity dimensions, six independent evidence states, existing local fixtures and laboratories, organizational evidence, and exact external-verification authorities.
Create fixed-marker carriers, inspect extraction separately from uptake, and copy an analysis-first prompt for an authorized AI test environment.
Rank a small local corpus, show retrieved documents and metadata, assemble model context, and locate a fixed benign marker.
Compare deterministic application-layer outcomes for source labelling, instruction/data separation, tool allowlisting, human confirmation, egress filtering, and bounded memory using a fixed harmless marker.
The recurring workflow is to preserve one artifact and compare its human, rendered, structural, metadata, Unicode, token, and decoder views before deciding what an AI should receive.
