Governed SEAD relational capture replay¶
The retained source capture reconciles 61 tables and 522,270 source rows to 522,270 distinct primary keys, with no duplicated primary keys. All 61 table receipts are complete; eight tables are genuinely empty. The table export retains each table's key field, row/key counts, minimum and maximum key, range count, payload digest, receipt digest and receipt status. Empty-table key endpoints remain empty.
The replay binds the normalized source-key ledger to independently admitted raw acquisition bytes, receipt identities, relational admission and the country boundary authority. It verifies every UUID/country binding through the existing ledger derivation and checks the exact normalized file set and part identities. It does not reacquire a changing provider snapshot or replace missing values.
The summary retains the source run, build, scope, raw admission, normalized manifest, source-key set and UUID-binding identities. The reviewed bounding-box population is 2,195 sites: 2,069 assigned (SE 1,925; DK 59; NO 45; FI 40), 103 requiring review and 23 unassigned. Review and unassigned sites remain outside the assigned country population.
The country discovery and retrieval ledger checks each of those 2,195 identities against its boundary decision, the retrieved source site row and the normalized site consumer. Its country counts are governed boundary assignments because the source supplied no country field.
Scientific boundary¶
The raw acquisition's release and downstream statuses are preserved exactly. Those downstream statuses describe the acquisition-time admission, before later normalized chronology and observation materialization. They are not assertions that the current normalized products are missing. Complete capture does not approve cross-source taxon/parameter mappings, lake events, propagation or independent measured vegetation accuracy. The replay does not convert refused scientific admission into a passing scientific result.
The pollen-named dataset screen exports sample-level source observations and records why the captured rows cannot form an admitted pollen time series or LRA input.
The numerical analysis value review traces 60 typed source values through their parent analysis rows and normalized components. It preserves source precision differences and keeps all rows outside scientific event admission.
The integer analysis value review reconciles all 24,701 typed integer components with their parent observations, source classes and chronology links while preserving qualified display text and event refusal.
The boolean component lineage account checks all 11,603 typed flags against their exact parent values and staged components. Its 17 display/class groups keep true, false and source-null states distinct without treating a flag as an admitted ecological or dated event. This account has its own single-file producer and source-bound checks; the six-family replay below retains its original scope.
The categorical component lineage account reconciles 4,800 linked items, nullable component fields and parent display strings. It preserves 2,821 display/item differences and keeps analysis components distinct from specimens, libraries, sequencing assays and dated events.
The complete analysis-value component census accounts for all 68,705 parent values, including 27,541 without a typed component. It preserves native class/type/unit links and explicit unit gaps; the population of this source table is governed Swedish sites, distinct from the four-country site and chronology populations.
The analysis-value method and type account projects the 27,541 parents without typed component rows into 28 source-declared class, type, method and unit groups. Its 27,451 Dendrochronology and 90 Ancient DNA links are value-class metadata; they do not admit assays, dated events or pollen observations.
The country chronology account joins all 25,109 reviewed claims to the 2,069 assigned site decisions. It separates source-row counts, distinct sites and chronology-display eligibility for each country; every claim remains refused for ecological propagation.
The country retrieval and evidence account reconciles those site decisions with source retrieval and 177,763 refused observation dispositions. It keeps 103 review and 23 unassigned sites outside the country totals, and records zero eligible ecological events.
The country source-unit account follows the same assigned sites through sample groups, physical samples, analysis entities, distinct linked datasets and analysis values. Its source-unit counts are separate from claim rows and ecological events.
Handoff accounting¶
The read-only handoff verifier binds the raw admission and manifest to the normalized materialization, source-key summary, 86-edge public custody, age review, four-country chronology and observation accounts, and built map audit. It checks the owned input digests and shared denominators without copying their row-level evidence into another report. Run:
PYTHONDONTWRITEBYTECODE=1 .venv/bin/python \
docs/report/source-relations/sead/verify_handoff_accounting.py \
--repository .
A passing result means local accounting verified for the pinned source run.
The acquisition release remains refused, all 177,763 observation dispositions
remain refused, and no ecological event or propagation edge is admitted. This
check does not supply qualified classification or chronology authority, a new
source-owner receipt, hosted release evidence, or the handoff's command and
integration history. The admitted capture still records
excluded_country_coordinates_lack_preserved_preflight_payload for the 126
excluded country decisions. Its acquisition-time downstream reasons precede
the later normalized claim and refusal materialization checked here.
Report ownership¶
We replayed the six source-owned report families linked above: the pollen-named dataset screen, numerical values, integer values, country chronology, country coverage and country source units. Their producers reproduced all 11 governed JSON/CSV files byte for byte from the admitted source and normalized inputs. All 15 local links in the six explanatory pages resolved. The owner verifier binds each producer to its exact output names and refuses an extra, missing or differing generated file. Run it from the repository root with a new destination:
PYTHONDONTWRITEBYTECODE=1 .venv/bin/python \
docs/report/source-relations/sead/verify_report_owners.py \
--repository . --output-root artifacts/jobs/sead-report-owner-replay
This covers these six source-owned families. The full mixed-owner docs/report/
tree has other producers and requires its own assembled no-difference check.
Reproduce¶
From an installed development environment at the repository root, use a new, unoccupied destination:
PYTHONDONTWRITEBYTECODE=1 .venv/bin/python \
-m bijux_pollenomics_dev.source_completion.sead_relations \
--data-root "$PWD/data" \
--output "$PWD/artifacts/sead-relations-replay"
The producer uses the existing independent source-key/acquisition validators and pinned identities, not the summary as an input. It refuses an existing destination, a changed admission, inconsistent row/table denominators or a source-key ledger that cannot be rederived from admitted acquisition bytes. A source refresh needs its own governed admission and identities; it cannot silently reuse this replay.