# Tara practice catalog and records · delivered increment · 8 September 2026

> Rewritten on 10 September 2026 as _Tara · find a practice_ chapter two. The
> subjects below are unchanged; the treatments, titles and prose are not. See
> [tara-catalog-chapter.md](tara-catalog-chapter.md).

This guide belongs to the [full V1 program](v1-completion.md). Its
implementation baseline is `5532d1513abcccb25a181c9c7a4e187a8b936536`; the
primary source pin remains `1ceae52c7896a59ac74a4429d410f3218f0b91c6`. The
preceding web-player guide is delivered. Recommendation/search/launch adapters,
the complete taxonomy selection policies, native callers, companions, mentors
and workbenches remain separate required subjects in that program.

Visual thesis: a cream field guide follows one concrete recommendation into
linked practice, content and teacher records. Use native object relationships,
record specimens, branch diagrams and worked numerical comparisons to explain
exact fields. One slide has one mechanism; matching names never stand in for an
actual integration.

Content plan: establish the catalog's producer and object family; follow subtype
and identifier selection; inspect practice/meditation/breathing builders;
distinguish supplied from inferred taxonomy; explain the normalization order,
lineage/rights/accessibility defaults and media references; compare the emitted
domain records with their separate validated contract shapes; close with the
actual evidence required to make a practice launchable. Recommendation ranking,
search, saved items and launch routing receive their own subsequent guide, where
their concrete runtime consumers can be taught fully.

Interaction thesis: use the established narration, source and explanation
reveals, with native selectable diagrams that remain complete in PDF. These
record transformations need no decorative animation or generated imagery. Real
surface captures belong where a concrete consuming UI improves recognition.

## Source review

Read the complete 804-line practice-model builder and the relevant taxonomy,
subtype and normalization functions in types.ts. Review its actual
recommendation caller in canonical-adapter.ts, and the independent contract
schemas for practice, teacher, breathing and meditation. Runtime constants from
@tara/features are a separate library from the contract's exported datasets.
Execute complete pinned modules in a declared immutable-source fixture,
retaining source hashes and real outputs for worked cases. A TypeScript
interface, constructor return or metadata availability flag is not a schema
parse, authorization check or media/device delivery.

The [practice discovery](v1-tara-practice-discovery.md) records earlier player,
repository and adapter findings. This guide must deepen the catalog's entire
object-building behavior rather than compressing it into the older overview. No
full feature unit or whole V1 workstream is closed by citation or slide count.

## Reviewed behavior and sequence

The 22-slide manuscript follows the linked invitation, four returned arrays and
actual recommendation-backed producer before separating subtype precedence,
content branches, identifiers and repeated inputs. Direct practice inference and
catalog conversion then explain why the same title can produce a different
modality. Supplied versus inferred taxonomy precedes duration, teacher, rights,
accessibility, media and cadence. Independent contract shapes and invariants
close the path before the concrete consumer-readiness case. This sequence keeps
object construction prerequisites ahead of validation details.

Twenty executed source cases cover linked records; empty and duplicate inputs;
subtype precedence; sleep/transition storage; direct versus catalog modality;
trimmed identity versus the original route input; object reuse and tag
normalization; inferred and explicit content labels; minute/band disagreement;
nonfinite inputs; fractional contract ranges; teacher/lineage/rights
propagation; accessibility defaults; distinct cadence models; cycle arithmetic;
domain versus contract shapes; schema membership and uniqueness; conditional
media references; and breathing label/cycle bounds. The fixture executes
complete pinned modules with local Zod and a declared facade exposing the three
actual feature taxonomy modules. Other modules in that feature barrel are not
simulated as executed. Twelve cited or transitively loaded source paths match
the primary pin and implementation baseline byte for byte.

The source reveals concrete boundaries that the teaching preserves. Catalog
conversion does not deduplicate; sleep and transition content use the meditation
array; all non-breathing recommendations are assigned guided modality. Teacher
identity and lineage are forwarded, while child rights receive independent
defaults. Transcript availability may be true with a null domain asset ID. The
four-phase feature cadence and three-field domain breathing model differ; the
five-minute box example yields thirty twelve-second cycles. Independent Zod
contracts require different shapes and can propagate a RangeError for a
fractional duration gap even through safeParse. These are source findings, not
fixes to Tara runtime behavior or observations of an actual member session.

## Optical review

All 22 initial compact slide captures were inspected. The containment tree,
branch outcomes, record relationships and comparison headers are complete; peer
rows have equal emphasis. The adapter is explicitly selected as the owner of the
catalog read. A space-marker glyph in the normalization example was unclear, so
the displayed input now uses quotation marks. The producer subtitle no longer
implies an upper count ceiling that the helper does not enforce. Narrow-display
review caught one long encoded route. Its encoded ID and route prefix now appear
separately; all 22 narrow slides pass. Final review passes 160 browser checks
and the complete focused center/PDF checks; see the
[receipt](../verification/v1-tara-catalog-2026-09-08/README.md).

## Remaining owning paths

The next recommendation/search/saved/launch explanation must trace its actual
consumers, not infer them from the canonical adapter's names. The canonical
search helper combines recommendation, optional course and favorite sources;
course/favorite reads require its configured searchUserId. The BFF read-adapter
registry currently supplies only apiAdapter, so its canonical search path has a
different pool from a configured caller. The
[web search guide](v1-tara-web-search.md) now explains the local
TaraSearchEngine component and route boundary. Full mood/theme/modality
selection policy, constraints and crisis routing also need their own complete
teaching, as do native callers, companions, mentors and workbenches. Citations
in this guide do not close those subjects.

The [search/launch discovery](v1-tara-search-discovery.md) now records the
complete local web search engine and its route boundary. The
[selection discovery](v1-tara-selection-discovery.md) records complete feature
taxonomy modules and the Tara–Lilith bridge. Neither preparation record is a
delivered guide or a substitute for the remaining consumer review.
