# Tara duration, context and lineage · authoring record · 9 September 2026

## Purpose and audience

This guide continues the Tara selection and handoff sequence for product,
design, engineering, editorial and review readers. Its observable outcome is to
read duration, context and lineage helper results without collapsing them into a
single recommendation score or claiming an application consumer that the source
audit did not locate.

Prerequisites are `tara-selection` and `tara-lilith-handoff`. The guide remains
inside the V1 member-experience route and precedes practice/player depth.

## Source review

Primary documentation is pinned at `1ceae52c7896a59ac74a4429d410f3218f0b91c6`.
The implementation, presentation and narration preservation baseline is
`096fb902ddce8134ec356a93574d581e0ae9b29e`. The complete pinned feature modules
and owning tests are byte-identical across that baseline and the source pin:

- `duration-buckets.ts` and its spec;
- `context-tags.ts` and its spec;
- `lineage-taxonomy.ts` and its spec;
- the mood, theme and modality dependencies loaded by the context module.

Fourteen source cases execute the catalogs, exact boundaries, availability,
quiet-hours and persona gates, issue paths and a bounded direct-symbol search.
The guide cross-references the separate Tara contract implementation but does
not substitute its Zod arrays for the executed feature constants.

## Learning sequence

The 24 pages move through one three-plane case, implementation ownership, all
five duration buckets, exact minute boundaries, the minimum-based availability
edge and validator. They then cover all ten context records, clock and event
selection, wrapping local-hour bands, canonical result order, quiet-hours
filtering and validation. The final section covers all eight lineage records,
record anatomy, cross-lineage and declared-frame decisions, unknown runtime IDs,
validation, the combined worked case and the unresolved application owner.

Every catalog value is visible. Dense inventories are split into four- or
five-row tables; decision semantics use branches; field shapes use record
anatomies. All pages use the existing cream field-guide system, with no new
shared renderer or style change.

## Corrections and boundaries

The feature availability helper compares `availableMinutes` with a bucket's
minimum. Twelve minutes therefore admits Standard even though its stored default
is fifteen. Retreat requires `includeRetreat: true` independently of available
time. Context signal inputs are accumulated into a set and returned in catalog
order, not signal order. Quiet-sensitive tags are filtered after time and event
selection unless the explicit include flag is true.

A lineage mismatch reason can coexist with `requiredPersonaMode: lineage-bound`;
callers must inspect `allowed` and `reasons`. Comparative mode still rejects an
unknown requested lineage. These are source-control-flow facts, not cultural,
clinical or product acceptance.

The helpers do not observe a clock, calendar, sensor, belief or member state.
The direct-symbol search found no application caller. That bounded search could
miss wrappers or alternate algorithms, so the guide records the consumer as
unresolved rather than claiming deployment-wide absence.

## Acceptance plan

- Run the pinned source cases and the repository `@tara/features` target.
- Build the guide, render all 24 narration tracks and verify exact prior speech
  reuse.
- Run the standalone/embedded desktop and mobile guide matrix, enlarged-text
  samples and four negative controls.
- Run the focused Presentation Center and exact PDF text/order checks, inspect
  all rendered PDF pages, and preserve the receipt.
- Assign the duration, context and lineage handbook units only after exact unit,
  heading, slide fingerprint and citation checks pass.

No live member, lineage authority, rights holder, clinician or normal-speed
listener participates in this automated delivery. Those acceptance paths remain
open in the full V1 program.
