# British Columbia broadcast media pack v2 — January–June 2026

This is the revised version of the original broadcast media pack, not a May-only
replacement dataset. It retains the same two cohort folders, 93 selected BC MLAs,
43 selected federal MPs representing BC, six-month reporting window and familiar
filenames. Every selected member remains, including those with no observed activity.

Open index.html locally or publish the whole directory. It provides the original
download layout plus a month selector and member/source verification pages; no
server or API key is needed. Click a Concept count to inspect supporting stored
contribution text and source links. The graphical JSON files contain no nested
fine-topic arrays or contribution evidence.

## Same files, revised presentation

Each of bc-mlas/ and bc-mps/ contains:

| File | Root | Contents |
| --- | --- | --- |
| members.json | members | Full selected roster, current database identity, graph-ID mapping, six-month totals |
| member-name-graphics.json | lower_third | Flat May name-and-one-stat graphics; also CSV/XML |
| member-topic-bars.json | lbar | Six May topic slots, ties, activity visibility, topic colours and comparable bars; also CSV/XML |
| monthly-topic-counts.json | monthly_topic_counts | Every member × January–June month × product category, with exact supporting record IDs |
| contribution-records.json | records | All six-month contribution records, full stored text/summary, excerpts, original source references and classifications |
| data-quality.json | members, records | Missing fields, provenance concerns and editorial checks |

As in the original pack, root graphics use May, a completed sitting month. This
pack also provides the same graphics files for **every month** under months/YYYY-MM/.
Member-specific summaries and optional paginated evidence are under
evidence/MEMBER_ID/YYYY-MM/. No date after June 30 contributes to activity metrics.

Root topics.json contains the 14-category product catalogue. Thirteen categories retain product category IDs. Social Services uses the explicit product ID product-SOC with no database_topic_id; this demo projection does not create a Neo4j Concept. Fine-topic IDs remain database element IDs. IDs are the existing MVP contract,
not approved durable public identifiers. Graphics id remains the original roster
member key, joined to members.json; graph_id there maps to the live API Person ID.

Presentation fields use name_display, role_title, party_short and district_short.
These are a revised lean contract: stations should rebind these fields rather than
assuming old name/role/line_2 and nested fine-topic columns still exist. Period,
metric and sample/review context appear in every graphic row, including CSV/XML.
Select by id or member_key, never by array position.

## Evidence and interpretation

Every count is reconstructed from the exact captured records. A contribution counts
once per Concept even with several fine Topic tags; different Concepts may overlap.
Share is the percentage of all member contributions in that month tagged with the
Concept, including unclassified contributions in the denominator. It is not floor
time or effectiveness. Cohort bars use the highest monthly single-topic count in
the selected cohort, with the denominator published in metadata and bar rows.

Evidence pages pin an immutable analysis_id to captured text and classifications.
Static files do not execute include_evidence flags; lean summaries and separate
evidence pages demonstrate the API's two modes. The original source text may cover
one contribution rather than a whole sitting. Open the official transcript link
for the full document. BC archive references are resolved consistently with the API;
original stored URLs are preserved. Source contents/availability are not reverified
or certified by matching counts.

Activity records and classifications were freshly captured with bounded read-only
January–June database queries. The selected roster basis is retained from the
original export. Names, party, basic role and district are current stored database
values, explicitly separate from the historical activity window. Historical party
or office affiliation is not reconstructed. Cabinet/critic titles absent from the
database are not invented. Display palettes/aliases remain proposals; Ross/font
testing, approved branding, portrait redistribution and PWA/QR destinations still
require external inputs. activity_eligible is not editorial approval.

## Verify and publish

Run python validate.py in this directory. It checks file hashes, complete six-month
coverage, every evidence-linked count, all monthly/root graphics and JSON/CSV/XML.
verification-report.json records differences from the original record set.
Copy this entire folder into the webapp's static/downloads area or share its ZIP.
No credentials or database connection are required to view the published package.
Original historical and on-air review packs remain unchanged. Nothing is deployed
automatically. See API_CONTRACT.md for the live API and REMAINING_GAPS.md for limits.

## Website defaults and operator handoff

The integrated website defaults to May and cohort-scaled topic bars. Member scaling
remains available for individual profiles; it must not be used to compare bar widths
between members. Rank labels compare topics within one member, not peers.
Every positive topic count is displayed. Counts below 3 are labelled Small sample. Eligibility is not editorial approval.

Use BINDING_GUIDE.md for the shipped filenames, roster-to-graph ID mapping, visibility
bindings, evidence pagination and Ross acceptance checklist. REMAINING_GAPS.md lists
open title, portrait, font, taxonomy and live-integration work. The demo uses all 14 PWA product categories, including the reviewed Social Services mapping. The production database taxonomy is unchanged.

## Interactive API request demo

The website also provides a separate live summary/evidence request panel for the
selected member and month. It shows the actual request URL and response JSON.
Default API requests omit evidence; selecting Include evidence sends
include_evidence=true and pins the summary's analysis_id across topics and pages.
API credentials remain on the backend. API errors are displayed without substituting
these captured files. Configure the web-app backend's BROADCAST_API_BASE_URL and
optional server-side credentials; the new broadcast API must be running.
