PCM1 Data And Tree Preparation

PCM1 converts a 76-row workbook and a 77-tip primate tree into a reconciled 75-species comparative state. That contraction is scientifically material: one duplicated species record is aggregated and two tree tips absent from the trait table are removed. Eight focused evidence bundles make every boundary reviewable before signal or ancestral inference begins.

Why Preparation Is Part Of The Evidence

Pagel's lambda and ancestral reconstruction both depend on the correspondence between trait rows and tree tips. A parsing error changes a trait value; an aggregation decision changes the species observation; pruning changes the covariance matrix; reordered taxa can attach correct values to the wrong species. Numerical agreement downstream cannot diagnose or repair those identity failures.

flowchart LR
    cell["Workbook cell"] --> species["Species observation"]
    species --> join["Tree–trait row"]
    tip["Tree tip and branch lengths"] --> join
    join --> covariance["Phylogenetic covariance"]
    covariance --> fit["Lambda and ancestral fit"]
    fit --> claim["Bounded PCM1 claim"]

The eight preparation bundles therefore do more than show that files load. Together they establish the analytical population and the object identity on which every later scalar comparison depends. Their claim scope stops at that handoff: they cannot establish the fitted model merely because the model uses their accepted inputs.

See How One Preparation Choice Reaches The Model

Suppose two workbook rows name the same species. Keeping both would give that species two response entries while the tree contains one tip; selecting one row would make an undeclared choice; aggregating them defines one new species observation. PCM1 records the aggregation rule and resulting row so the model population is explicit.

The same logic applies on the tree side. A tip with no retained trait value cannot contribute a paired observation, but removing it changes the tree from which covariance and internal clades are derived. PCM1 therefore records the two removed tips, preserves the derived tree identity, and proves exact set and order correspondence before any likelihood is evaluated.

This is the essential evidence chain: a row-level decision changes the response vector; a tip-level decision changes the covariance; reconciliation determines which vector element belongs to which covariance row. “Seventy-five on both sides” is only a count until those identities are joined.

Preparation Changes The Model Matrix

For a comparative model, pruning is not equivalent to deleting decorative labels from a plot. Removing a tip selects a different covariance submatrix; aggregating duplicate rows changes the response vector; and reordering taxa changes which response is paired with each matrix row. The accepted model input is therefore the ordered pair:

(75-species longevity vector, 75 × 75 tree-derived covariance)

Both members must use the same ordered species identity. A correct vector and a correct tree can still produce a scientifically wrong analysis when their orders or label mappings disagree.

flowchart LR
    workbook["Raw workbook<br/>76 rows"]
    typing["Type contract<br/>5 numeric columns"]
    missing["Missingness<br/>0 retained missing values"]
    aggregate["Species aggregation<br/>75 unique rows"]
    source_tree["Source tree<br/>77 tips"]
    trim["Prune 2 absent tips<br/>75-tip rooted tree"]
    join["Exact set and order<br/>75 tree–trait rows"]
    derivatives["Governed CSV + Newick<br/>checksum-addressed handoff"]

    workbook --> typing --> missing --> aggregate --> join
    source_tree --> trim --> join --> derivatives

Keep The Three Populations Separate

PCM1 begins with different row and taxon populations. The final value 75 appears several times, but each occurrence answers a different question.

Population	Starting denominator	Material operation	Accepted denominator
workbook records	76 rows	aggregate the duplicate-species contract after type and missingness review	75 species rows
source-tree tips	77 tips	prune Lagothrix_lagotricha and Eulemur_fulvus, which lack retained trait rows	75 tips
comparative join	75 species rows and 75 tips	establish exact set and order correspondence	75 aligned taxa

Equality of the accepted counts is necessary but not sufficient. The join must also prove label normalization, one-to-one membership, exact order, tree identity, and trait-row identity. Otherwise two different 75-taxon analyses could be mistaken for the same population.

Every Population Change Creates A New Identity

Preparation operations are directional. The processed table and trimmed tree retain provenance to their sources, but they are new scientific objects with new membership and checksums.

Operation	Information changed	Identity required downstream
type repair	representation and admissible numerical values	repaired schema, parse failures, and source-to-output mapping
duplicate aggregation	row population and meaning of one species value	source members, aggregation function, missing-value policy, and resulting row
tree pruning	tip population, descendant clades, covariance, and local node numbering	removed tips, retained order, topology, branch lengths, and derived tree digest
tree–trait reordering	positional association between covariance rows and values	exact ordered taxon vector on both sides
export	serialization of the accepted handoff	schema, format, content digest, and producer identity

Reproducing only the final count cannot reproduce these decisions. A later study that filters another taxon or transforms a trait must mint a derived identity and re-evaluate all dependent node, covariance, and model records.

Governed Input And Handoff Records

Role	Repository record	Identity carried forward
raw observations	datasets/primate_raw.xlsx	original 76-row workbook boundary
processed reference	datasets/reference_primate.csv	75 rows, 75 species, nine-column order and repaired types
source phylogeny	datasets/primatetree.nex	77 tips with branch lengths
trimmed analysis tree	datasets/reference_trimmed_primatetree.nwk	75 tips with the governed set and order
dataset registry	datasets/registry.json	source role, locator, schema summary, and SHA-256 identity

The checked-in workbook supports auditability, but the focused Python review starts from the governed processed reference rather than claiming to replay the external workbook import. This is an explicit limitation in evidence-002, not an invisible substitution.

Preparation Evidence Ledger

Bundle	Claim boundary	Governed observation	Current verdict
evidence-002	workbook loading	76 raw rows contract to the checked-in 75-row processed reference	matched
evidence-003	type repair	five numeric columns parse without failures; factor/text roles are explicit	matched
evidence-004	missing-data accounting	75 post-repair rows; no missing values in the governed processed table	matched
evidence-005	duplicate aggregation	one duplicate-species contraction; zero duplicates remain	matched
evidence-006	tree import	R and Bijux both report 77 original and 75 trimmed tips with branch lengths	matched
evidence-007	tree diagnostics	rooted, binary, ultrametric, with an explicit node-label correspondence	matched
evidence-008	tree–trait join	two named missing tips removed; final set, order, and example node IDs agree	matched
evidence-009	processed export	checked-in CSV and Newick identities match the governed handoff	matched

All eight focused preparation bundles support their bounded claims. The aggregate evidence-001 manifest is matched_with_tolerance because it also contains signal and ancestral comparisons plus plot-only and artifact-only surfaces. Its aggregate status neither weakens the eight exact preparation verdicts nor promotes later claims.

Read The Type Contract

The processed table uses this reviewed interpretation:

Role	Columns or states
numeric	body_mass, gestation, home_range, longevity, social_group_size
binary factor	sex_dimorphism with governed values 0 and 1
categorical text	mating_system, including Pair, Polyandrous, and Polygynous states
identity text	family, species

evidence-003 records empty parse-failure arrays for all five numeric columns. That proves the governed processed artifact satisfies this parsing contract; it does not prove that every workbook cell was semantically correct before processing.

Account For Missingness And Aggregation

The reference rule aggregates numeric values by species with missing-value removal. In the retained processed artifact, every governed column has a missing count of zero. The raw-to-processed delta is one row, and evidence-005 records one duplicate-species contract with zero duplicates after grouping.

These facts must be read together:

76 raw rows - 1 duplicate-species contraction = 75 processed species

The evidence establishes the observed contraction and resulting unique-species state. It does not independently replay every workbook transformation. A different aggregation rule would define a different analysis input even if it also produced 75 rows.

Reconcile Tree And Trait Identity

The source tree has 77 tips; the processed table has 75 species. The two tree tips absent from the trait table are:

• Lagothrix_lagotricha;
• Eulemur_fulvus.

After pruning, evidence-008 records equality of the tree-tip set, tree-tip order, and aligned species order. It also matches representative node IDs:

Identity	R	Bijux
Hylobates_lar	29	29
Pan_paniscus	33	33
Node32	107	107

Node numbers are valid only within this governed tree identity. The stronger durable identity for an internal node is its descendant-taxon signature; a different pruning or traversal can renumber nodes without changing a clade.

Verify Tree State Before Modeling

evidence-006 and evidence-007 jointly establish:

• original and trimmed tip counts agree across R and Bijux;
• both retained tree records carry branch lengths;
• the trimmed tree is rooted, binary, and ultrametric;
• R node 77 labeled Node2 resolves to the same taxa as the Bijux Node2 match.

Rooting, topology, branch lengths, taxon set, and node identity are independent parts of the model input. Matching the tip count alone would not establish the same comparative covariance.

Accept The Downstream Handoff

Before fitting a model, a consumer should verify:

1. the processed CSV and trimmed Newick digests against the dataset registry or input manifest;
2. the nine-column table contract and 75 unique species;
3. the 75-tip tree set and exact order;
4. rooted, binary, ultrametric, and branch-length state;
5. the two explicit pruning exclusions;
6. any node reference by descendant-taxon signature as well as local node ID.

Matching filenames are insufficient. A downstream result should retain input checksums, reconciliation state, and model-specific exclusions in its own manifest.

Reject the handoff if any count is reported without its members, if pruning is recoverable only from prose, if tree and table order were assumed rather than checked, or if an internal node is identified only by a local integer. Those conditions can preserve the headline number 75 while changing the actual population or ancestral target.

Claim Boundary

The preparation verdicts establish source interpretation at the governed handoff, processed types and missingness, duplicate contraction, tree import, diagnostics, pruning, alignment, and exported identity. They do not establish:

• the biological appropriateness of mean aggregation for another question;
• phylogenetic signal or a fitted lambda value;
• a Brownian trait model;
• ancestral-state estimates or their historical truth;
• plot equivalence;
• general behavior on other datasets.

Continue with signal and ancestral results, where each numerical statement is tied to the aggregate scalar ledger.