# Clay & Modeling Materials: research notes

Date: 2026-10-25. Scope: the clay or clay-like material itself — polymer (oven-bake) clay, air-dry clay, kiln-fired ceramic/pottery clay, oil-based sculptor's modeling clay (plastilina/plasteline, the studio material sold to sculptors and mold-makers, per Chavant's own "Fine Arts, Transportation Design, Special Effects" framing), and two-part epoxy modeling clay/putty that cures permanently hard through a chemical reaction. In the Zamana taxonomy this is `supplies > crafts > clay`. Not in scope, and left to siblings in the same catalog: the kiln itself (a machine, not a supply); sculpting/cutting tools, texture stamps and rollers (`supplies > crafts > cutting-tools`, `brush-stamp-tools`); glazes, underglazes, liquid clay mediums and finishing varnish (`paint` — "Paints, Inks & Dyes"); bundled multi-item clay project kits (`craft-kits`); and children's play dough and toy putty — Play-Doh-style dough, Silly Putty, Thinking Putty and similar toys that stay pliable and are never meant to cure (`supplies > play > toys > play_dough`, "Play Dough, Putty & Drawing"). That last exclusion needs its own reasoning since it sits in a different taxonomy branch (`supplies > play`, not `supplies > crafts`) and the two groups can look alike: Wikipedia's own putty article draws the putty half of the line for us — toy putty "such as Plasticine and Silly Putty are common toys" that stay pliable, while "polyester putty and epoxy putty are thermosetting polymers... [that] become permanently rigid after curing." A two-part epoxy clay or putty belongs in this register because it cures; a stretchy toy putty belongs to the sibling because it never does. The dough half of the line follows from use, not chemistry, since this register's oil-based "stays soft forever" family also never cures: it is the sculptor's studio material (Chavant, Castilene), not the flour-and-water or cornstarch dough sold as a children's toy, even though both stay soft indefinitely. Where a listing bundles clay with tools or a kit, this register covers only the clay itself; the bundle's tools and packaging, and any true play-dough or toy-putty product, are the sibling categories' job.

Quote labels: **verbatim** = the text in quotation marks is what `curl` plus HTML-tag-stripping, or `pdftotext`, returned when I read the page or PDF myself. **fetched excerpt** = the WebFetch tool's summary of a page I did not read raw. **search excerpt** = a search-engine summary only, because WebSearch was exhausted for this session before any category research began (it reported its 200-call budget used on the first two queries) and the page itself would not load. No search-excerpt-only number is in category.json; every number below was read in raw text or a rendered PDF/image I viewed myself.

Source types I could not reach despite trying: FDA (`fda.gov/food/...` lead/ceramicware pages all returned empty 404-style bodies), CPSC (`cpsc.gov` returned an Akamai "Access Denied"), and Internet Archive snapshots of either (no snapshot existed for the specific CPSC ceramicware URL I tried, and a later Wayback request was itself rate-limited as "suspected abusive bot traffic"). I could not reach a retail listing (Amazon, Michaels, Joann) at all; every attempt either wasn't tried because the house note in `checks.md` already documents they block fetches, or would have needed WebSearch to find the right URL, which was unavailable. `listing_field_hint` below is therefore inference from brand pages and general retail convention, not a confirmed look at a live listing, consistent with the second-commonest failure `checks.md` flags.

## Seed

Seed run (recorded verbatim from the script's own output):

```
=== Example entry (from an unrelated category) ===
category: olive-oil
what this shows: a facet that experts rank first, that almost no listing carries,
and that is expensive to backfill. ... Also shows a derived metric hanging off a
researched facet.
<example: harvest_date, from olive oil>

=== Persona ===
A sommelier, who thinks in origin, process and how a thing changes over time
```

1. The clay equivalent of `harvest_date` is the ACMI seal (`acmi_seal`): experts and safety-conscious buyers (parents, teachers) rank it first, it is almost never a structured field on a listing, and confirming it means reading the brand's own packaging language or searching ACMI's certified-product tool per SKU, not reading a spec table. Survived the data check: ACMI's own site states the rule in one clear sentence I read myself (see Sources), and it is a genuine safety fact, not a marketing one.
2. A sommelier thinks about process and change over time. What that persona would notice first here is that clay inverts the wine model: a sommelier judges a process the maker already finished before the bottle reached them, but for clay the irreversible "aging" step (baking, firing, or simply letting it dry) happens in the buyer's own hands, after purchase, using heat or time the buyer controls. The one family that explicitly refuses this logic — oil-based "never hardens" clay — is formulated so that step never happens at all, which the trade treats as a feature, not a flaw (Chavant: clays that are "permanently pliable" and "reusable," "requires no heating"). Survived: this became the backbone of the `cure_method` split and the oil-clay trade-off. Dropped: a literal batch or "freshness" date for clay. Air-dry clay can stiffen in an unopened bag over long storage, but no brand I reached publishes a shelf-life figure, so there is no sourced number to hang a facet on; I did not invent one.

## Sources and what each supports

- **ACMI, "What do the Seals mean?"** (`acmiart.org/materials-safety`, verbatim, read via curl + tag-strip). Supports: the AP seal is for products safe for children in sixth grade or younger; "Never give children in sixth grade or younger a product with the CL Seal"; both seals certify to ASTM D4236, LHAMA and the FHSA; toxicological evaluation is done by board-certified toxicologists. This is the source for the `acmi_seal` facet and judgement J4.
- **ACMI, "Governing Regulations"** (`acmiart.org/regulations`, verbatim, curl + tag-strip). Supports: LHAMA (1990) amends FHSA to require chronic-hazard labeling under ASTM D4236; CPSIA (2008) lowered lead limits in children's products and required third-party testing; CPSC enforces LHAMA. This is the source for judgement J2's safety framing and for `food_safe_after_fire`'s general caution, used because I could not reach FDA/CPSC directly (see above).
- **Sculpey, "What to Know About Polymer Clay"** (`sculpey.com/pages/new-to-polymer-clay`, verbatim, curl + tag-strip, Polyform Products Co.). Supports: "Polymer clay stays soft until you bake it in your home oven"; "NEVER put oven-bake clay in the microwave"; store leftover clay in a polypropylene (#5) container; "if you are making pieces thicker than 1/4 inch it is best to create an armature... out of crumpled aluminum foil"; condition (knead) before use; "leach" an over-soft batch by sandwiching it in paper under a weight overnight. Supports the `cure_method` mechanism text, the armature mention in `why_experts_care`, and judgement J5 (microwave is explicitly warned against, not a real cure path).
- **Sculpey Premo Bead Making Kit instructions** (PDF at `cdn.shopify.com/.../PE_4048_Premo_Bead_Making_Kit.pdf`, verbatim, `pdftotext`). Supports: "Bake on oven-proof glass or metal surface at 275°F (130°C) for 30 minutes per 1/4" (6 mm) thickness. DO NOT MICROWAVE. DO NOT OVERBAKE." Source for `cure_temp_f`'s upper bound (275°F), `cure_time_min_per_quarter_in` (30 min), and the derived metric `bake_time_total_min`.
- **Sculpey Clay Comparison Chart** (PDF at `cdn.shopify.com/.../ClayComparisonChart_1.2026_OL.pdf`; text is outlined/vector in the PDF so `pdftotext` returned nothing — rendered to a PNG at 150 dpi with `pdftoppm` and read visually, which I am treating as verbatim since it is the brand's own typeset chart, not a summary). Supports the `firmness_raw` facet and the `firmness_raw` trade-off: Sculpey III / Bake Shop are "Soft and easy to condition," matte, aimed at "KIDS," "BEGINNERS"; Premo is for "FREQUENT CLAY USERS" and "PROFESSIONAL ARTISTS," "Premium strength and flexibility... Perfect for advanced techniques like caning"; Soufflé is "Lightweight... Suede-like"; Liquid Sculpey "Starts out as a liquid but bakes into solid"; Original/Super Sculpey are "Ceramic-like," made for carving; a wax-based "Non-Drying" line exists for "Claymation videos, play time," confirming a never-hardens line exists even within a mainstream oven-bake brand's catalog (I kept this as evidence that `never_hardens` is a real, marketed category, not that this specific product belongs to the `clay` register's oven-bake family).
- **STAEDTLER, FIMO soft** (`staedtler.com/intl/en/products/fimo-modelling-clay-accessories/fimo-soft/`, verbatim, curl + tag-strip). Supports: "oven-hardening modelling clay... Preheat the oven to 110°C. ... Leave the modelled item to harden in the oven for 30 minutes... FIMO models are only completely hardened once they have cooled down." Source for `cure_temp_f`'s lower bound (110°C / 230°F) and judgement J1 (brand variance).
- **STAEDTLER, FIMOair** (`staedtler.com/intl/en/products/fimo-modelling-clay-accessories/fimoair/`, verbatim, curl + tag-strip). Supports: "you don't need to use an oven or kiln to harden the clay... simply be left to dry naturally in the air"; "The drying time depends on the thickness... a few hours to a few days"; "Do not attempt to accelerate the drying process using a hairdryer or by placing in sunlight. This may result in models cracking." Source for the `air_dry` branch and its `makes_irrelevant` list.
- **Chavant Inc., homepage** (`chavant.com`, verbatim, curl + tag-strip). Supports: oil-based clay ("Plasteline") is "The Finest Name in Modeling Clay... Serving sculptors in Fine Arts, Transportation Design, Special Effects"; "since 1892." Source for the `never_hardens` family description.
- **Chavant, Fine Art Sculpting Clays** (`chavant.com/fine-art-sculpting-clays`, verbatim, curl + tag-strip). Supports: "Sculptex... is an ideal alternative to sulfur-based clays, which are incompatible with some mold rubbers"; Prima Plastilina "never hardens or changes with air exposure"; Castilene comes in "medium grade... great for detailing, and the hard grade is best for sharp detail"; Renaissance (sulfur-based) "has been used by fine art sculptors since the 1950's." Source for `sulfur_content`, judgement J3, and the oil-clay trade-off. This is the only source behind J3: the page says sulfur-based and mold rubber are "incompatible," and the register's "can stop... from curing" names the specific mechanism (standard trade knowledge, not stated in these words on this page). J3's confidence is now `medium` and its note says single-sourced, rather than reading as confidently as the two-brand-confirmed oven-bake numbers above.
- **AMACO, shop homepage navigation** (`amaco.com`, fetched link listing, curl). Supports the family split itself: AMACO's own navigation separates "ceramic-clay," "air-dry-clay," "modeling-clay" (oil-based) and "self-hardening-clay" as distinct product lines, corroborating `cure_method` as the leading brand-level division, independent of the Sculpey/FIMO evidence above.
- **AMACO, Low Fire Gloss glazes** (`shop.amaco.com/glazes-underglazes/low-fire-glazes/lg-low-fire-gloss/`, verbatim, curl + tag-strip). Supports: "Fire To: Cone 05 = 1031°C / 1888°F ... Apply to bisqueware fired to Cone 04 = 1063°C / 1945°F... Always slow fire lead free glazes for best results." Source for the cone/temperature mention in `firing_range`'s definition and for judgement J2's "lead-free" framing.
- **AMACO, Potter's Choice glazes** (`shop.amaco.com/glazes-underglazes/high-fire-glazes/pc-potters-choice/`, verbatim, curl + tag-strip). Supports: "Cone 5 = 1186°C / 2167°F - Cone 6 = 1222°C / 2232°F ... bisqueware fired to Cone 04 = 1063°C / 1945°F." The Cone 04 = 1945°F figure matches the Low Fire Gloss page exactly, so I am treating it as corroborated within AMACO's own catalog — but both figures still come from one brand. I did not find a second, independent brand or an Orton-hosted numeric chart (the Orton pages I could reach were product-navigation, not the chart itself; see below). I have therefore kept precise cone-to-degree numbers out of category.json's structured `range` fields and used them only narratively in `definition`/`why_experts_care`, attributed to AMACO by name.
- **The Edward Orton Jr. Ceramic Foundation, "Pyrometric Cones"** (`ortonceramic.com/pyrometric-cones`, verbatim, curl + tag-strip). Supports: cones "measure heat-work, the effect of time, and temperature"; used "worldwide to monitor ceramic firings in industrial kilns, pottery kilns, and small hobby kilns." This is the standards-body source for the cone system as the trade's own visual/measurement grammar (`visual_grammars[0]`). I tried to reach Orton's actual cone-to-temperature chart (`/temptab`, `/large-cones`) to get a second, independent numeric source for the AMACO figures above; both returned only product-ordering pages, not the chart, so the precise degree numbers stayed single-brand-sourced as noted.
- **Apoxie Sculpt (two-part epoxy clay)**: still could not fetch `avesstudio.com` on the 2026-10-25 re-check (403 Forbidden) or a Wayback snapshot of it (tool cannot reach web.archive.org at all). No brand-specific number for Apoxie Sculpt or any other two-part epoxy product is in category.json.
- **Wikipedia, "Epoxy putty"** (`en.wikipedia.org/wiki/Epoxy_putty`, fetched 2026-10-25). Supports: the two parts are "mixed thoroughly in equal quantity," kneading "triggers an exothermic chemical reaction" (epoxide polymerization), and the cured product "will retain its shape rather than melting when heated." This is a generic, brand-independent source, not a substitute for Apoxie Sculpt's own page, but it is a real citation for the `mechanism.plain`/`families` description of two-part epoxy clay (added to `mechanism.evidence`) and for why `cure_temp_f`/`cure_time_min_per_quarter_in`/`clay_body_type`/`firing_range` are irrelevant to it (no oven, no kiln, cures at room temperature by chemistry, not heat). It says nothing about sulfur content or knead-firmness grades.
- **Wikipedia, "Putty"** (`en.wikipedia.org/wiki/Putty`, fetched 2026-10-25). Supports: "Modeling clay and play putty, such as Plasticine and Silly Putty are common toys" (stay pliable); "Polyester putty and epoxy putty are thermosetting polymers that can be molded by hand, but become permanently rigid after curing." Source for the scope note above and for a new judgement distinguishing this register's two-part epoxy family from the `play_dough` sibling.

## Data points used for derived metrics

`bake_time_total_min` (thickness-scaled oven-bake time) depends on `cure_time_min_per_quarter_in`, whose only sourced value is Sculpey Premo's own instruction: "30 minutes per 1/4" (6 mm) thickness" (verbatim, Sculpey Premo kit PDF). FIMO soft's instructions give a flat "30 minutes" with no explicit thickness scaling in the text I read, so the metric is attributed to Sculpey's stated method specifically, not asserted as a universal brand-independent rule. The metric's own `why` field now says this explicitly (Sculpey-specific scaling, confirm the package before relying on it), matching the hedge already in `cure_time_min_per_quarter_in`'s `backfill_note` instead of contradicting it. `cure_temp_f`'s range (230–275°F) is the one number in this register genuinely confirmed from two independent brands: FIMO soft = 110°C/230°F, Sculpey Premo = 130°C/275°F.

## Against the existing add-on schema

AGENTS.md field-kind vocabulary used below: *vertical anchor*, *category quality currency*, *hard filter*, *high-signal markdown* (per `zamana-expertise/AGENTS.md` rule 4), plus the note that visually-derived fields (rule 1) do not belong in the schema.

| Add-on field (`clay.jsonc`) | This register's view | Proposed kind |
|---|---|---|
| `cure_method` enum `[air_dry, oven_bake, kiln_fire, never_hardens, chemical_two_part, microwave, not_applicable, unspecified]` | Keep as the vertical anchor — every source confirms this is the lead division (AMACO's own navigation splits exactly this way). **Drop `microwave`**: I found no mainstream product that cures in a microwave, and Sculpey's own instructions explicitly say "NEVER put oven-bake clay in the microwave" (judgement J5). `not_applicable`/`unspecified` are pipeline null-handling sentinels, not real category values, so they correctly sit outside this register's `values` list for the facet. | vertical anchor (keep enum, minus `microwave`) |
| `food_safe_after_fire` boolean | Keep, and it should match this register's `food_safe_after_fire` facet exactly. Add a pipeline rule to cut backfill cost: default `false` for every `cure_method` other than `kiln_fire` (polymer, air-dry, oil-based and two-part clays are not food-contact-safe regardless of claims — judgement J2); only `kiln_fire` rows need a real look-up, and only when the maker states a lead-free glaze. | hard filter (keep boolean) |
| *(not present)* `acmi_seal` | Propose adding. This is the facet that actually governs whether a clay is appropriate for a young child, which is a real safety question this category carries and the add-on currently has no field for. Enum `[ap_seal, cl_seal, none_stated]`. | hard filter (new) |
| *(not present)* firmness, clay body type, sulfur content, cure temperature/time | Considered and **not** proposed as structured fields. These are real expert facts (see facets below) but they are brand- and line-specific description, not a clean per-row filterable value every row can carry — AGENTS.md rule 3 says this is exactly what `high_signal_markdown` is for. Recommend a line in `crafts-standard.jsonc`'s or `supplies-standard.jsonc`'s `high_signal_markdown` comment calling out "cure temperature/time, firmness, sulfur content for mold compatibility" as high-signal when present. | high-signal markdown, not a field |
| *(not present)* color/translucency/finish (matte, satin, translucent) | Considered and dropped. This is visually derivable from product photos (AGENTS.md rule 1: "If a field can be better determined by looking at the product photo than by reading the page, it doesn't belong here"), so it is image-pipeline territory, not a schema field. Kept out of this register's facets for the same reason; see "Trimmed detail." | none (visual embeddings) |

## Branches

Every leg below is traced to a specific sentence in this file or in `category.json`.

**`cure_method` = `oven_bake`** → `then_ask: [firmness_raw, cure_temp_f, cure_time_min_per_quarter_in]`, `makes_irrelevant: [clay_body_type, firing_range, sulfur_content]`.
- `firmness_raw`, `cure_temp_f`, `cure_time_min_per_quarter_in` only exist because of the Sculpey/FIMO sources above, both of which describe oven-bake (polymer) clay specifically.
- `clay_body_type` and `firing_range` are sourced only from AMACO's ceramic-clay and glaze pages, which describe kiln-fired clay; they do not apply to a clay that never sees a kiln.
- `sulfur_content` is sourced only from Chavant's oil-based clay pages; sulfur in an oven-bake polymer clay is not something any source here discusses.

**`cure_method` = `air_dry`** → `makes_irrelevant: [clay_body_type, firing_range, sulfur_content, cure_temp_f, cure_time_min_per_quarter_in]`.
- FIMOair states directly: "you don't have to harden your modelled items in the oven" and needs "No kiln required," which removes both the oven-bake facets and the kiln-only facets in one stated fact.
- `sulfur_content` again traces only to the oil-clay (Chavant) sources, not to air-dry clay.

**`cure_method` = `kiln_fire`** → `then_ask: [clay_body_type, firing_range, food_safe_after_fire]`, `makes_irrelevant: [sulfur_content, cure_temp_f, cure_time_min_per_quarter_in, firmness_raw]`.
- `clay_body_type` and `firing_range` trace to AMACO's ceramic-clay catalog split (low-fire / mid-high-fire / premium mid-range) and to the Orton/AMACO cone sources.
- `food_safe_after_fire` becomes the live question here specifically because AMACO's glaze instructions say "Always slow fire lead free glazes" — i.e., food safety is a glaze-and-firing decision the buyer makes only once they're in the kiln-fire branch.
- `cure_temp_f`/`cure_time_min_per_quarter_in` trace only to Sculpey/FIMO (oven-bake); a kiln firing schedule is a different, unsourced-here process, so I removed these rather than guess a kiln number.
- `firmness_raw` (soft/medium/firm, as marketed on polymer and oil clay lines) has no equivalent distinction in any ceramic clay source I read; AMACO's own split for ceramic clay is by firing range and audience (beginner/classroom vs professional), not by knead-firmness, so I made it irrelevant here rather than force the mapping.

**`cure_method` = `never_hardens`** → `then_ask: [firmness_raw, sulfur_content]`, `makes_irrelevant: [clay_body_type, firing_range, cure_temp_f, cure_time_min_per_quarter_in, food_safe_after_fire]`.
- `firmness_raw` and `sulfur_content` trace directly to the Chavant Fine Art Sculpting Clays page (Castilene's medium/hard grades; Sculptex vs sulfur-based alternatives).
- Kiln and oven facets are irrelevant by the same chain as above (no source ties them to oil-based clay).
- `food_safe_after_fire` is made irrelevant because oil-based clay never fires or cures at all (Chavant: "never hardens or changes with air exposure"), so there is no fired, glazed surface for a food-safety claim to attach to; this is a one-link chain from that single stated fact, not a chain I had to assemble from two sources.

**`cure_method` = `two-part epoxy`** → `makes_irrelevant: [clay_body_type, firing_range, sulfur_content, cure_temp_f, cure_time_min_per_quarter_in]`.
- `clay_body_type`, `firing_range`, `cure_temp_f`, `cure_time_min_per_quarter_in` are irrelevant because epoxy clay cures by a room-temperature chemical reaction, not an oven or a kiln — now backed by the Wikipedia epoxy-putty source above, not just `mechanism.plain`'s own text.
- `sulfur_content` is irrelevant for the same reason it is irrelevant on every other branch except `stays soft forever`: the facet's own `definition` scopes it to oil-based clay specifically ("Whether an oil-based modeling clay is made with..."), and two-part epoxy is a different chemistry (resin, not oil-suspended mineral). This is the same definitional chain already used for the `oven_bake`/`air_dry`/`kiln_fire` branches, not a new inference.
- `firmness_raw` was dropped from this branch's `makes_irrelevant` (it was in the first draft). No source here or anywhere in this register says whether knead-firmness varies across two-part epoxy brands, and unlike `sulfur_content` the facet's own definition is not scoped away from epoxy clay — Wikipedia's "clay-like consistency" line, if anything, suggests firmness could vary by brand. Rather than assert either that it matters or that it doesn't, the branch now simply says nothing about it, which is honest given the evidence.
- The cold-start question for `cure_method` now lists all five values, including `two-part epoxy`, so this branch is reachable without the assistant having to recognize a named product first.

I considered a branch on `food_safe_after_fire` itself (e.g., "when yes, then_ask clay_body_type") but dropped it: by the time this split is reached `cure_method` is already known, so the only new fact a "yes" answer adds is a constraint on which value of `acmi_seal`/`clay_body_type` is acceptable, which I state in the split's `rationale` text instead of inventing a second branch object for the same underlying fact.

## Trimmed detail

- **Color range / mixability**: real expert knowledge (polymer clay caning and color-mixing is a whole sub-craft; translucent lines like Premo's and Pardo's are specifically prized for faux-gem jewelry work), but dropped as a facet because it is visually derivable from product photos — the image pipeline's job per `AGENTS.md` rule 1, not a text field.
- **Shrinkage percentage**: I looked for a sourced shrinkage number for ceramic and air-dry clay (a real expert concern — "the piece you fire isn't the size you made") on AMACO's product pages but could not reach a live product page with the figure (404s on the specific SKUs I tried); rather than invent a percentage, I left shrinkage as qualitative language inside `why_experts_care` and the kiln trade-off, with no number in the JSON.
- **Apoxie Sculpt mix ratio / cure time**: wanted for the `chemical_two_part` family description; `avesstudio.com` returned 403 Forbidden and the one Wayback snapshot I found was itself rate-limited. No number for this family appears in category.json.
- **Ballistic/industrial use of Chavant's Roma Plastilina** (body-armor testing clay since 1977, per Chavant's own homepage) — true and sourced, but off-scope for a consumer crafts category; mentioned here only in case it resurfaces as a confusing search result for someone else working this category later.

## Script notes

`validate.py`, `render.py` and `viewer.py` ran as documented; no misbehavior to report.

## Corrections

### Corrections after the blind check (2026-10-25)

1. `cure_method`'s split question and the facet's `cold_start_question` now list all five values ("Air-dry, oven-bake, kiln-fired, stays soft forever, or two-part epoxy?"), so the `two-part epoxy` branch is reachable from the question itself, not only by recognizing a named product. (Check item 1 / Part B(c).)
2. Added `two-part epoxy clay/putty` vs. the `play_dough` sibling ("Play Dough, Putty & Drawing", `supplies > play > toys > play_dough`) to the scope paragraph above, with a sourced reason (Wikipedia's "Putty" article: epoxy putty cures permanently, toy putty like Plasticine/Silly Putty stays pliable) for the epoxy half of the overlap, and a stated use-based reason (studio/professional vs. children's toy, per Chavant's own framing) for the oil-based "stays soft forever" half. Also removed "including non-hardening 'non-drying' kids' clay" from the oil-based clay scope line, since nothing in `category.json`'s oil-based family description is aimed at children and that phrase was the direct cause of the overlap. (Check item 2 / Part B(e).)
3. Fetched two generic sources for two-part epoxy clay (`avesstudio.com` is still unreachable — 403, and this tool cannot reach web.archive.org at all): Wikipedia's "Epoxy putty" article, added to `mechanism.evidence` and to a new judgement; Wikipedia's "Putty" article, added to a new judgement (see #2). Removed `firmness_raw` from the `two-part epoxy` branch's `makes_irrelevant` since nothing supports it either way; kept `sulfur_content` there, but on a definitional basis now stated in the Branches section (the facet's own `definition` is scoped to oil-based clay, the same reasoning already used for the oven-bake/air-dry/kiln-fired branches), not on "general trade knowledge." Added a new, explicitly `low`-confidence judgement stating plainly that the family's mix ratio and no-heat cure are sourced, but its sulfur content and firmness are not. (Check item 3 / Part A #8 / Part C's two `two-part epoxy` legs.)
4. Rewrote `bake_time_total_min`'s `why` field to state the Sculpey-specific scaling and the brand-variance caveat, matching `cure_time_min_per_quarter_in`'s own `backfill_note` instead of asserting it as a universal rule. (Check item 4.)
5. Lowered judgement J3 (sulfur/mold-rubber) from `high` to `medium` confidence and added "single-sourced" plus the "incompatible" vs. "can stop... from curing" distinction to its note, matching how the AMACO cone figures are already flagged as single-brand. (Check item 5.)
6. Added a sentence to the `food_safe_after_fire` facet's `why_experts_care` stating that no dedicated food-contact regulator backs the claim directly, so a reader of the facet itself (not only `notes.md` or the judgement) sees the gap. (Check item 6.)
7. Changed `firmness_raw.source` from `listing` to `derived` (it is a line-name-to-value lookup, the same inference pattern as `cure_method`, not a labeled field) and reworded its `listing_field_hint` to match. Changed `sulfur_content.backfill_cost` from `high` to `medium`, since the facet's own `backfill_note` already describes a default-plus-exception-check rule, not a true per-SKU lookup. Neither finding is rejected; both are applied as the check suggested. (Check item 7.)

No finding from the check was rejected. Item 3's fix is partial by design: the branch's weakest two legs are now either definitionally justified (`sulfur_content`) or dropped (`firmness_raw`) rather than asserted on "general trade knowledge," and the family description is cited to one generic encyclopedia source rather than the two independent sources this register holds every numeric claim to — there is still no brand-specific number for any two-part epoxy product, and that gap is now stated in a `low`-confidence judgement instead of being invisible outside `notes.md`.