# V1 Sophia depth delivery · 9 September 2026

This receipt records the expanded
[Sophia evidence and answers guide](../../decks/sophia.html). The guide grows
from eight to forty slides while preserving every established slide identifier
and its prerequisite after events, jobs and discovery.

The new sequence begins with the operation and implementation posture. It traces
the live Nisaba-backed answer route, citation mapping, extractive ranking,
count-based state thresholds, answer envelope, claim labels, optional model
cleanup and fallback. It then treats source credibility as a separate route,
including request admission, the 40/20/20/20 scoring model, all six domain tiers,
the 0.5 corroboration floor and the distinction between top-source grounding and
mean confidence.

The second half covers the versioned evidence-adapter boundary, grounding,
review and admin capability matrix, real Nisaba reads, honest empty verification
results, source-set hashing and readiness, trace export, the real-but-unwired
BM25 and verification loops, and the specified ingestion, source-adapter,
multi-store, publication and invalidation architecture. A closing consumer map,
edge-case table and four-lens review prevent library presence, enum membership,
candidate credibility or an invalidation plan from being presented as completed
customer evidence.

Every meaningful unit in `V1/architecture/substrate-sophia.md` and
`V1/features/sophia-grounding.md` is now reviewed: four existing assignments
plus 49 new fingerprint-bound assignments. Global teaching coverage moves from
227 to 276 reviewed units, with 10,018 structural and 103,050 unassigned units
remaining across the broader repository inventory.

The [final browser report](browser-final/report.json) records 280 assertions,
136 captures, zero fit findings, four rejected negative controls and zero page
errors across standalone 1440, 1280, 390 and 320 pixel readers and expanded,
collapsed and mobile embedded states. Visual review found and corrected hidden
recovery conditions and three compact-reader overflows before the final pass.
The [Center report](center-final/result.json) records 14 integration checks, 80
embedded slide captures, two audio boundaries, 86 unchanged-guide continuity
checks and a fresh 40-page tagged PDF. The [PDF text report](reports/pdf-text.json)
binds the exact authored visual text on all forty pages.

Thirty-two new fingerprint-bound narration tracks add 42.6 minutes. The complete
Sophia guide runs 56.5 minutes, and the library has 1,279 current tracks totaling
28.61 hours. All new tracks pass byte, hash, duration, decode and encoded-pause
verification. The integrated library now has 87 guides, 1,279 unique slides,
1,302 guide-slide instances and 618 cited paths.

Presentation verification includes 19 builder and assignment tests, 14
executable pinned-answer cases, generator freshness, coverage freshness, format
and lint checks. A focused current-tree BFF run could not import `fastify` or the
historical `@oshun/evidence-sophia` package and executed zero tests; the
[failure log](logs/bff-sophia.log) is retained. The guide is pinned to
`1ceae52c7896a59ac74a4429d410f3218f0b91c6`, so its source snapshots,
fingerprints and pinned-answer harness remain the evidence for that historical
V1 boundary.

This delivery completes the two pinned V1 Sophia pages. It does not establish a
live model provider, production route, consumer-specific display acceptance,
completed invalidation cascade, native-device behavior, human comprehension or
normal-speed listening. Broader Sophia repository coverage, Iris, Psyche/Lilith
and the remaining V1 program stay active.
