# Nisaba · One passage, and what stands behind it · receipt · 12 September 2026

Chapter three of _The other rooms_, delivered in the Eve edition on 12 September
2026. Nine inherited slides become a thirty-six-slide chapter under the `nisaba`
id. The brief is [authoring/nisaba-chapter.md](../../authoring/nisaba-chapter.md);
the track brief is
[authoring/the-other-rooms-track.md](../../authoring/the-other-rooms-track.md).
Machine-readable state is in [state.json](state.json).

## What was delivered

Thirty-six slides in the guide: thirty-four authored here, plus `lilith-v1-0`
and `room-maturity` carried unchanged from Meet Lilith. Twenty-seven are new — a
cover, six section dividers, a close and nineteen teaching slides. All seven of
Nisaba's own inherited slides are rewritten in place; four of them belong to the
reviewed foundation (`foundation-coverage.json`) and keep their identifiers for
that reason. Nothing is merged away.

Two shared manuscripts are retired. `content/04-v1-0-member-rooms.json` and
`content/22-member-deepdives.json` held Nisaba slides and nothing else once
Arete and Nyx had moved out, so both empty and are removed rather than left as
files with no slides in them.

## Evidence

- **The pinned contract, executed here.** A transpile loader compares every file
  with the pin before running it. Thirty-one exported schemas and three helper
  predicates. A published passage with one citation is accepted. It is refused
  for a draft with no citation, for overlapping segments, for a high-confidence
  concept edge with no evidence, for an edge to its own node, for the old
  relation label `derives-from`, for an `unlisted` notebook, for a shared
  notebook with one collaborator, for the four-key body the shipped form posts,
  and for citation-verification authority on an unverified credential.
- **The pinned store, corpus and room builder, executed here.** Nine passages
  from five editions across four traditions, every one estimating a single
  minute. A member with no history holds nine records at zero progress, zero
  annotations, no last-read and unsaved. One passage is flagged daily; one
  carries a manuscript witness. That witness is bound to Dhammapada I.1–2 while
  transcribing an ānāpānasati passage, with folio confidences of 0.92, 0.94 and
  0.93 over a literal base of 0.84. An annotation was written and read back at
  thirteen fields, moving the passage's count from zero to one. A notebook was
  stored at visibility `unlisted`. Both branches of the room builder were
  compared, the member branch assembled exactly as the shipped route assembles
  it.
- **The offset guard, tested here.** An annotation on a ninety-two-character
  passage, claiming offsets 900,000 to 900,100 with an excerpt that occurs
  nowhere in the text, was accepted, stored verbatim and counted.
- **The philology engine, executed here.** The pinned Pali handler supports six
  scripts; `anattā` was rendered into Devanagari, Sinhala, Thai, Myanmar and
  Khmer and transliterated back unchanged; a romanised line of the room's own
  passage was detected at confidence 1 with no validation issues; the geminate
  was identified at offset three.
- **The census, counted at the pin.** Twenty-three directories under
  `libs/nisaba`. Three of them are imported by any code outside that root —
  `study-adapter`, `workspace`, `study-plans`. The backend-for-frontend imports
  none. `@nisaba/languages` and `@nisaba/criticism` appear outside their own
  tree only in CI config, a docs pipeline file, a generated product-graph
  record, one verification script, the root path map and ten documents; no
  application code imports either. `curatedConfidence` has 1,246 call sites.
- **Two captures, rendered from the pin with no server.** The member routes
  cannot be driven on this host, so both use the pinned bundler with declared
  boundary mocks: the shipped reading desk over the shipped room builder with a
  declared passage, and the shipped new-notebook form with its network edge
  mocked and the posted body read back rather than sent. Both are byte-stable
  across runs.
- **An independent source check** of the draft found **twenty-three** wrong,
  overstated or unsupported claims. Every one was re-verified against the pin
  and fixed before the chapter shipped. The largest corrections:
  - the census was **three** packages imported outside the library root, not
    eight — the first count had matched import lines inside a developer-guide
    markdown file;
  - the member's room carries **two** related texts, not none: the shipped route
    passes the record's compare targets, which the first probe omitted;
  - the desk has **ten** depth routes, not nine, and two of them reach the
    service — the new-notebook form is one of the pages;
  - the capture shows a `Sections` heading with **nothing beneath it**, so the
    slide claiming a section list running down the left was describing a page
    that does not exist;
  - the three external environment variables **are** declared in `.env.example`
    with real URLs and checked by one script; what is true is that no TypeScript
    file reads them;
  - the durable store fails closed on **writes**, not on reads;
  - the four browser storage keys are read and written by the shipped surface,
    and two of the four can be downloaded from its own citation export; the
    accurate gap is the account export and erasure, which run on the server;
  - the languages engine covers **twenty-eight** scripts across thirty-two
    directories, four of which are shared machinery;
  - the scholar-mode witnesses are written in an apparatus's form but are not
    the standard sigla, so the chapter no longer vouches for them;
  - the King James Version is public domain in the United States and under
    perpetual Crown copyright in the United Kingdom, so "five public-domain
    editions" became "five named, dated editions" with the exception stated.

## Checks

| Check         | Result                                                                |
| ------------- | --------------------------------------------------------------------- |
| Build         | 1,591 slides, 697 sources; freshness check clean                       |
| Python suite  | 481 tests pass                                                         |
| Node suite    | 489 tests pass                                                         |
| Browser sweep | 36 slides × 1440, 1280 with audio chrome, 390 and print — 0 findings    |
| PDF           | 36 pages, exact authored visual text, the one table complete           |
| Narration     | 1,591 tracks verified, nothing pending; 50.8 minutes in this chapter    |
| Bindings      | no assignment or teaching file binds a Nisaba slide                    |

Two sweep findings were fixed before it passed: `nisaba-opening` overflowed
1440 by 18.6px until its fifth callout was dropped — the passage and its
citation are named in the notes instead — and the `nisaba-gap` divider
overflowed 1280 by 23.8px until its subtitle was shortened.

## Treatment deviations

Two, each deliberate:

- `nisaba-textual-model` ships as a **record anatomy** rather than the card
  grid, stat panel or compare panel the plan lists: its subject is the seven
  fields of one record, and the anatomy is the layout for that.
- `nisaba-visibility` **keeps a table**, which the plan allows when readers need
  exact values side by side. Three layers define the word `visibility` with
  three different vocabularies; the point is only legible in columns.

## What this chapter does not establish

Nothing here touches a deployed service, a real member account, a device or a
network. The captures render pinned components with declared mocks and the
posted notebook body was recorded, not sent. No claim is made that an annotation
reached a database on this host, that a member would be served this passage, or
that the philology engines are reachable from any product surface — they are
not.
