Methods appendix

A unit is a recorded plant-and-animal statement, rather than an article. A sentence covering cats and dogs can therefore yield separate units. Exclusions include unclear plant identity, missing animal identification, absent ASPCA entries, food derivatives, and conflicting animal-specific reference directions. Article titles and introductory or concluding summaries cannot supply inherited plant or animal identity.

The accompanying calculation reconstructs 151/159 from final-unit blind identifiers and adjudication flags, following the final ruling's definition of disagreement. It does not rerun the independent judgments: the final CSV does not contain a separate reviewer-decision column. The earlier agreement figure is transcribed from the final ruling and its arithmetic checked.

Amendments 1–5 were written before any sampled article was read. Amendment 6 was written after the first pass, to resolve 143 statements the pre-registered rules did not cover. It reclassified them as consistent or excluded and added no discrepancies. Amendment 7 was written after the first pass and the second-pass blind review; A7.1 documented existing handling, and A7.2 was withdrawn before being written.

The final ruling records this count history, in candidate/consistent/excluded order: 3/96/282 before retrospective application, totaling 381; 3/90/288 after Amendment 6, still 381; and 3/91/289 after Amendment 7, totaling 383. The expanded accessible sample and final rulings produced 3/109/360, totaling 472. The same 3 candidates remained throughout.

Independence was limited. The earlier unblinding used an incorrect join key after the blind judgments had been committed; that reviewer also applied plant-context inheritance inconsistently. A later independent review session's automatically supplied git status exposed the commit title “apply editor rulings on step 3 (mistletoe A4.3; sensitivity range)”. Its American mistletoe decisions matched that title's wording. Restrictions on git history and file reading were instructions, not technical isolation; project memory had been changed to neutral wording.

The first pass also recorded the Pony Tail entry as absent; the entry was subsequently archived, and that unit changed from excluded to consistent. It recorded Mother of Thousands as having no ASPCA entry, although ASPCA lists that name under Piggy Back Plant, Tolmeia menziesii. The article named a Kalanchoe species, so the exclusion remained, with the earlier rationale disclosed here.

Accepted applications of the rules are retained in this method appendix. PPH text naming only “pets” does not establish an animal-specific source conflict. Standalone bold plant labels in Article 12 were treated as subheadings. “safe for your fluffy friend” was accepted as non-toxic wording; “not cat friendly” was not covered, but its unit was already excluded for identity. American mistletoe statements naming P. serotinum were excluded against ASPCA's P. flavescens, without inferring synonymy. Panda Plant and Mother of Thousands names pointing to different plants were excluded. String of Dolphins, given a species name in the article, was excluded against ASPCA's “Senecio species” entry. Table headings naming animals and toxicity, with a plant row marked “Both”, were counted as explicit toxic statements.

The archived PPH material covered 13 plants; the final ruling notes that most comparable units therefore could not be checked for the reference-conflict exclusion.

Amendments 6 and 7 were originally recorded in Chinese and are shown here in English translation, prepared for publication. The original text is available on request. Working files named in this record (CSV files and review notes) are not published because they contain copies of third-party pages; they are also available on request.

Holiday plant report 2026 — Methods

Pre-registered 2026-10-02, before any sampled article was read.

1. Sample

• 10 queries:
  • "holiday plants toxic to cats"
  • "holiday plants toxic to dogs"
  • "christmas plants safe for cats"
  • "christmas plants safe for dogs"
  • "are poinsettias poisonous to cats"
  • "are poinsettias poisonous to dogs"
  • "is mistletoe poisonous to cats"
  • "is mistletoe poisonous to dogs"
  • "is amaryllis toxic to cats"
  • "is holly poisonous to dogs"
• For each query: the top 10 organic Google results, United States, captured from a third-party SERP snapshot tool on a single day. Results are deduplicated by URL. Every result is included; none is removed.

2. What counts as a discrepancy (only one kind)

• The article states a definite, animal-specific conclusion about a plant (toxic / non-toxic / safe for cats, or for dogs), and the ASPCA record's toxicity line for that plant and that animal states the opposite.

3. Excluded (never counted)

• Disagreements of degree (mild vs moderate, etc.).
• Statements about "pets" that do not name cats or dogs.
• Plant identity unclear (e.g., "lilies" without a species or genus the ASPCA record covers).
• Plants with no ASPCA entry.
• Plants where the ASPCA and Pet Poison Helpline records disagree with each other on that animal.

4. Evidence

• Each article is archived (saved snapshot + date) on the day it is checked.
• Every reported discrepancy shows both verbatim sentences, both URLs, and access dates.

5. Verification

• Every candidate discrepancy is re-checked from scratch by an independent checker who did not take part in the first pass.
• The editor (Zhe Li) reads every finding before publication.

6. Self-audit first

• Verdapaw's own pages for the plants in scope are checked with the same rule before any third-party article is read. Any issue found is corrected first.

7. Framing

• Findings are reported as "differs from the ASPCA record", never as "wrong".
• Verdapaw makes no toxicology claims of its own beyond transcription of ASPCA and Pet Poison Helpline records.

Amendment 1 (2026-10-02, before any sampled article was read)

The text above is left unchanged. This amendment replaces section 1's query list only. Sections 1 (capture method, dedup, inclusion), 2–7 still apply.

A1.1 Why the original query list is withdrawn

The 10 queries in section 1 were hand-picked. A check against our keyword file showed that at least one of them ("holiday plants toxic to cats") gets 0–10 US searches a month. A sample built on queries almost nobody types would not reflect what readers actually see. The list is withdrawn and replaced by the mechanical rule below. No sampled article had been read when this decision was made.

A1.2 Rule D (full text)

Source: data/relevant_keywords_us_2026-08.csv (columns Keyword, Volume; US monthly search volume).

1. Filter the keyword file using the matching interpretations in A1.3. A keyword is kept only if it names a holiday plant, contains a toxicity/safety term, and names cats or dogs (or both).
2. Group each kept keyword by (plant, animal set). The animal set is one of: cats only / dogs only / cats and dogs. A keyword that names more than one plant is grouped under the first plant that appears in it.
3. In each group, keep the single keyword with the highest volume. Ties are broken alphabetically.
4. Rank the groups by that keyword's volume, highest first; ties are broken alphabetically by keyword. The representative keywords of the top 10 groups are the sample queries. There is no cat/dog quota.
5. Known limitations are stated (A1.4) and not corrected for.

If fewer than 10 groups exist, the actual number is used and the rule is not relaxed. (15 groups exist; see appendix.)

A1.3 Matching interpretations (approved)

Matching is case-insensitive and on whole words.

• Holiday plants: poinsettia, mistletoe, amaryllis, holly, christmas cactus, christmas tree, cyclamen, christmas rose / hellebore, paperwhite, plus the generic phrases "christmas plant" and "holiday plant". Plurals count (poinsettias, hollies, christmas cacti, etc.), and "paper white" written as two words counts as paperwhite.
• "christmas rose" and "hellebore" are one plant.
• Toxicity/safety terms: toxic, toxicity, toxin, toxins, poison, poisons, poisonous, poisoning, safe, safety, dangerous, harmful, deadly, lethal, fatal.
• Cats: cat, cats, kitten, kittens, kitty, kitties, feline, felines. Dogs: dog, dogs, puppy, puppies, canine, canines.
• A keyword that names both cats and dogs forms its own animal set ("cats and dogs"); it is not counted under either side alone.

Count: 215 keywords name a holiday plant and a toxicity/safety term; 202 of those also name cats and/or dogs and enter grouping. The ruling that adopted rule D cited 215; the figure that actually enters grouping is 202.

A1.4 Known limitations

• The keyword file was collected for cat-side demand. Dog-side coverage is thin, so dog queries are under-represented.
• Volumes are annual monthly averages, not December figures.
• Holiday demand is dominated by poinsettia; the sample reflects that (3 of 10 queries).
• Grouping uses the first plant name in the keyword, literally. Group 15 ("christmas tree cactus toxic to dogs") is filed under christmas tree although it most likely refers to a christmas cactus. It falls outside the top 10 and does not affect the sample.

Appendix A1 — Rule D output

Selected (top 10 groups):

Not selected (groups 11–15):

Groups 9–12 all have volume 20; the cut between 10 and 11 is decided by the alphabetical tie-break. Mistletoe, amaryllis and paperwhite produced no group (no keyword met all three conditions). 15 groups exist in total.

Amendment 2 (2026-10-02, before any sampled article was read)

The text above is left unchanged. This amendment fixes how the sample is built from the SERP exports. It was written using only the URL, Title, Type, Position and Last Update columns of the exports; no sampled page was opened, fetched or read.

Inputs: the 10 Ahrefs "SERP overview" exports for the Appendix A1 queries, stored unmodified in serp/. Output: sample.csv (one row per deduplicated URL; columns url, domain, title, class, queries, best_position).

A2.1 Organic top 10

For each query, only rows whose Type contains "Organic" are taken (this includes "Organic, Video preview" and "Organic, Thumbnail"), sorted by Position ascending, and the first 10 are kept. AI Overview, People also ask, Sitelinks and Videos rows are not taken. If an export has fewer than 10 organic results, the actual number is used; nothing is added.

Position is the export's own column. It counts every SERP element (an AI Overview or a People also ask box takes a position), so the 10th organic result can sit at position 11 or 12. The positions recorded in sample.csv are these export positions.

A2.2 Deduplication

A URL that appears more than once, in the same query or in different queries, is one sample. queries lists every query it appears in with its position; best_position is the lowest of those positions, and title is taken from that row.

A2.3 Classes (all results kept; none removed)

Every deduplicated URL is assigned one class, mechanically, from its host name and URL (and, for the reprint check only, its export title). The host name is the URL host with a leading "www." removed; a rule that names a domain also covers its subdomains. Rules are applied in this order; the first match wins.

1. R — reference source: host is aspca.org or petpoisonhelpline.com. Listed, not checked.
2. U — user-generated content and video: host is reddit.com, facebook.com, quora.com, justanswer.com, youtube.com, tiktok.com or stackexchange.com; or, on any other site, the URL path contains a forum/thread segment (/forum/, /forums/, /thread/, /threads/, /topic/, /topics/, /community/, /discussion/, /discussions/, /viewtopic, /showthread). Listed, not checked.
3. Pending Li's ruling: host is not an R host, but the URL or the export title contains "aspca", "petpoisonhelpline", "pet-poison-helpline" or "pet poison helpline". Such a page may be a verbatim reprint of an ASPCA/PPH document (which would make it R), but that cannot be confirmed without opening it, so it is not classified here.
4. A — checkable article: everything else.

Only class A enters the denominator of the discrepancy count. Pages on other sites that reprint ASPCA/PPH documents verbatim belong to R; at this stage they can only be detected through rule 3 and are decided by Li.

A2.4 Capture times

All 10 exports were captured on 2026-10-02 between 03:03 and 03:16 UTC (Last Update column):

A2.5 Known limitations

• Five of the 10 exports contain fewer than 10 organic results (see A2.4); the sample uses what was exported.
• "christmas tree": if the ASPCA has no entry for the tree in question, articles in that query group may yield no checkable conclusion under section 3 ("Plants with no ASPCA entry").

Amendment 3 (2026-10-02, before any sampled article was read)

The text above, including Amendments 1 and 2, is left unchanged.

A URL outside the reference domains whose URL or title names Pet Poison Helpline or ASPCA as the document's source is classified R (reference-source material hosted elsewhere) by that naming alone, without opening it. This replaces the 'pending' outcome in Amendment 2 for such URLs. Applied to one URL in the current sample (wholepetclinic.com PDF, cyclamen query, position 3). Final counts: A 43, U 23, R 7.

Amendment 4 (2026-10-02, before any sampled article was read; follows the self-audit required by §6)

The text above, including Amendments 1–3, is left unchanged except where A4.4 states that one sentence of section 7 is replaced; the original sentence stays in place above.

A4.1 What "the ASPCA record" means

The ASPCA record for a plant is the Toxicity / Non-Toxicity line of that plant's entry page in the ASPCA Toxic and Non-Toxic Plant List. ASPCA news releases and blog articles are not reference records. Reason: the self-audit found an ASPCA news article stating that Spider Plant is "mildly toxic to dogs and cats," while the ASPCA's Spider Plant entry reads "Non-Toxic to Dogs, Non-Toxic to Cats."

A4.2 Entries that are silent on an animal

When an entry page's toxicity line does not mention the animal in question (for example, the Easter Lily, Lily (Lilium species) and Day Lilies (many varieties) entries read only "Toxic to Cats"), that comparison is not counted in the headline figures. If the ASPCA's plant list for that animal classifies the plant, the comparison is reported separately as "list-based", apart from the headline figures, and is stated to rest on the list classification, not on the entry text.

A4.3 Same plant only

A comparison is made only when the plant an article names is the same plant as the ASPCA entry. When an article refers to a genus or a common name broadly and the ASPCA has entries only for some species within it, the statement falls under section 3, "plant identity unclear," and is not counted.

A4.4 Section 7, second sentence, replaced

Section 7's second sentence ("Verdapaw makes no toxicology claims of its own beyond transcription of ASPCA and Pet Poison Helpline records.") is replaced, because the self-audit found it is not exactly true. The library contains one source other than the ASPCA and Pet Poison Helpline:

No genus page, list page or other plant record cites a source other than the ASPCA and Pet Poison Helpline.

Replacement sentence: Verdapaw's grades rest only on ASPCA and Pet Poison Helpline records; for one plant (sago palm), details that do not affect the grade — the timeline of signs and one sentence on survival — come from a veterinary hospital reference (VCA Animal Hospitals), which is listed among the sources on the pages where it is used.

A4.5 Self-audit result (section 6)

Verdapaw's own published records and text were checked with the rule in section 2 before any sampled article was read.

• Verdict layer, 151 published records (83 cat, 68 dog): 0 discrepancies. 133 match the ASPCA entry for the same plant; 5 have no ASPCA entry for the plant (Carolina jessamine ×2, lilac, marigold ×2); 2 are excluded because the ASPCA and Pet Poison Helpline records disagree (Peruvian lily ×2); 7 are records whose scope differs from the ASPCA entries' level (genus-level record vs species-level entries: anthurium ×2, holly ×2, lily cat side, elephant ear ×2), all in the same direction as the ASPCA entries; 2 are list-based under A4.2 (lily and daylily, dog side); 2 have no species-level ASPCA entry (rubber plant, fiddle leaf fig).
• Text layer: 1 finding. Verdapaw's fiddle leaf fig page quoted the ASPCA news article in A4.1, including its statement that Spider Plant is "mildly toxic to dogs and cats," which is the opposite of the ASPCA's Spider Plant entry and of Verdapaw's own S0 grade for spider plant. The correction is being prepared and is reviewed by the editor before publication.

Amendment 5 (2026-10-02, before any sampled article was read)

The text above, including Amendments 1–4, is left unchanged. This amendment corrects two statements in Amendment 4.

A5.1 Correction to the A4.2 example

A4.2 gave the Easter Lily, Lily (Lilium species) and Day Lilies (many varieties) entries as examples of entries that read only "Toxic to Cats." That is wrong. Each of these entry pages has two lines: "Toxicity: Toxic to Cats" and a second line, "Non-Toxicity: Non-Toxic to Dogs" (Day Lilies) or "Non-Toxicity: Non-Toxic to Dogs, Non-Toxic to Horses" (Easter Lily, Lily (Lilium species)). The Orange Day Lily entry has the same two-line form.

Cause: the script used in the self-audit read only the first toxicity line on each entry page. Of the 124 entry pages cited by Verdapaw's records, 121 have one line and 3 have two (Day Lilies (many varieties), Orange Day Lily, Easter Lily). The error was found while drafting the content corrections that followed Amendment 4, when the original entry pages were read again.

The A4.2 rule stands. No entry in the current check is silent on an animal, so the rule has no instance at present.

A5.2 Clarification of A4.1

The ASPCA record for a plant is every Toxicity and Non-Toxicity line on that plant's entry page, read animal by animal.

A5.3 Corrected A4.5 counts

The self-audit was re-run with all lines read. Only two records changed (lily and daylily, dog side, both now matching the entry). Final verdict-layer figures for the 151 published records (83 cat, 68 dog):

• Match the ASPCA entry for the same plant: 135
• Discrepancies: 0
• List-based (A4.2): 0
• No ASPCA entry for the plant: 5 (Carolina jessamine ×2, lilac, marigold ×2)
• Excluded because the ASPCA and Pet Poison Helpline records disagree: 2 (Peruvian lily ×2)
• Record scope differs from the level of the ASPCA entries, same direction: 7 (anthurium ×2, holly ×2, lily cat side, elephant ear ×2)
• No species-level ASPCA entry: 2 (rubber plant, fiddle leaf fig)

The text-layer result in A4.5 is unchanged: 1 finding (the fiddle leaf fig page's quotation of the ASPCA news article).

Amendment 6 (2026-10-08, after the first pass; editor's rulings)

The text above, including Amendments 1–5, is left unchanged. Zhe Li approved the following eight rulings after reviewing the first-pass rule gaps. These are post-first-pass rulings, not pre-registered rules. They are applied to the 143 unresolved first-pass record units without changing their verbatim statements, original CSV rows or row order. Sections 2–7 still apply except where a ruling below explicitly specifies the treatment.

A6.1 Context inheritance

A plant is inherited only from the nearest higher-level subheading; an animal is inherited only from the same sentence or the same subheading. Article titles, opening summaries and closing summary sentences cannot serve as a source of inheritance. Where nothing can be inherited → excluded (§3, plant/animal unclear).

A6.2 Explicit non-toxic conclusions and recommendation wording

"safe to be around" / "pet-safe" / "cat-friendly" / "non-toxic", where the same sentence or the same subheading names cats or dogs → counts as an explicit non-toxic conclusion under §2. "approved" / "recommended" / "a good choice" on their own → not a conclusion; excluded.

A6.3 Back-references and shared names

An article's common name is compared only if it corresponds to exactly one ASPCA entry. Where one common name corresponds to several entries, such as lilies, orchids or jade, and the article does not name a species or genus → excluded (A4.3, plant identity unclear). Achira/Canna Lily: where the article writes canna and the ASPCA has a Canna Lily entry → compared.

A6.4 Headings do not form a conclusion on their own

Headings such as "Signs of X Toxicity" → do not on their own constitute a conclusion; only sentences in the body text count.

A6.5 Foods and processed derivatives

Foods and processed derivatives (pumpkin pie, cranberry sauce, ground nutmeg, etc.) → excluded; ASPCA entries are plants, not foods.

A6.6 Non-binary statements

"Most are safe, but some varieties…" → not a binary conclusion; excluded as a dispute of degree under §3.

A6.7 Animal-specific binary direction in PPH

Where PPH gives only a mechanism, signs, a severity chart, or the general term pets, without naming cats or dogs → it cannot establish a binary direction for that animal; excluded.

A6.8 ASPCA "sp." entries and bare common names in articles

For ASPCA scientific names with "sp." versus bare common names in articles: the bare common name orchids does not correspond to Phalaenopsis sp. → excluded; where the article writes moth orchid or Phalaenopsis → compared.

A6.9 Preserving first-pass evidence, and subsequent steps

• For candidate 12, the non-toxic sentence and the toxic sentence in the same section are both quoted verbatim and marked "internally contradictory within the article, one of which differs in direction from the ASPCA"; the article is not given an overall characterisation.
• Of the 14 unreachable articles, the 5 returning HTTP 429 are retried once only, at least 24 hours after each article's first GET; HTTP 403 and 404 remain unreachable and are not circumvented. Retries are saved as separate evidence and do not overwrite the first snapshot.
• Reclassification updates only the classification and its basis, preserving the first-pass original records and the original CSV rows; the report body is still not written and nothing is published. The second step is handed to an independent reviewer who did not take part in the first pass (§5).

Amendment 7 (2026-10-09, written after the first pass and after the second-pass blind review; editor's rulings)

The text above, including Amendments 1–6, is left unchanged. Zhe Li approved the following three rulings after the second-pass blind review (pass2-blind-review.md) and the retroactive application of Amendment 6 (pass2-retro-a6-review.md). Like Amendment 6, these are post-hoc rulings, not pre-registered rules. They fill gaps in the existing rules only; they do not re-open classifications the existing rules already decide. Amendment 6 is applied retroactively to all record units (editor's ruling, 2026-10-09); Amendment 7 is applied to all record units and to the four articles that became reachable on the single HTTP 429 retry (A6.9).

Numbering note: the proposed A7.2 (a table-row rule) was withdrawn by the editor before it was written; the number is left vacant so that it matches the record of rulings. Table-row handling is set out in the final section of this Amendment.

A7.1 Subheading names cats or dogs; the sentence says only pets

Where the sentence itself says only pets (or animals, furry friends or similar general terms) and does not name cats or dogs, the animal is not inherited even if the nearest subheading names cats or dogs → excluded (§3, cats/dogs not named). Exception: the case already provided for by A6.2 — non-toxic wording such as "pet-safe", where the same subheading names cats or dogs, still counts as an explicit non-toxic conclusion under A6.2. This ruling codifies the existing handling in the first pass and Amendment 6, and does not change any existing classification.

A7.3 "True lilies"

"True lilies" does not automatically correspond to the ASPCA's Lily (Lilium species) entry. It is compared only when the sentence itself or its nearest subheading (the inheritance scope of A6.1) names Lilium; Lilium named in an adjacent sentence or elsewhere in the article does not count. Otherwise excluded under A4.3 (plant identity unclear).

A7.4 Pets sentences counted under A6.2 beneath the same subheading: split into a cat unit and a dog unit

Units counted as explicit non-toxic conclusions under A6.2, where the animal comes from a subheading naming both cats and dogs, are split into two record units, one for cats and one for dogs, following the same approach as Amendment 6 (for example, rows 311/312 of pass1-statements.csv). Units currently affected: row 313 (Rosemary) and row 314 (Spider Plant) of pass1-statements.csv, each changing from 1 row to 2 units. The original sentences and original CSV rows remain unchanged.

Cases A7 does not address: table rows

No new table-row rule is added. Table rows follow the handling of the first pass and the second-step blind review: where the table header names an animal and a toxicity column (for example, "Plant | Toxic to Cats or Dogs? | Toxicity | What To Do"), the row names a plant, and that column's value is "Both", it counts as an explicit toxic conclusion for that plant, for cats and for dogs. The report's methods section states this as it is.