The Ledger · Phase 0 · 31 August 2026

Project status

Phase 0, between Weekend 1 and Weekend 2. The gold set and the board exist; the automated pipeline does not. Every row on the board was extracted by hand.

170
Rows
66
Verified
104
In review
95
Subjects
19
Contested
21
Speakers
4/8
Milestones

Where it stands

Weekend 1 — gold set + eval script
66 human-verified rows, quote fidelity 100%. The accuracy gate runs today.
Done
Registries — speakers, subjects, sources
21 speakers, 99 registered subjects (7 awaiting sign-off), 44 episodes.
Partial
Ingestion — Spoken.md list fetched
10 shows curated, 44 episodes registered, 43 transcripts on disk.
Done
Extraction — rows on the board
170 rows: 66 verified, 104 awaiting review. 95 subjects, 19 contested.
Done
Weekend 2 — the board renders
Static site builds from the claim files, by-subject and by-speaker.
Done
Weekend 2 — Stages A–C prompts
0 of 5 stage files written; 0 entries in prompts.lock. No model has run the pipeline — every row so far was extracted by hand.
Not started
Weekend 3 — Stage D + review queue
Attribution check and the approval UI in /review.
Not started
Phase 0 test — consulted in real research
The only assumption that matters. Kill criterion: 8 weeks live without 3 real uses. Clock has not started.
Not started

What needs deciding

3 subject ids not in the registry

Unregistered: CN, JP, software. These are the PROPOSED_SPLIT / PROPOSED_RENAME entries awaiting sign-off. CN is a live defect — one speaker's bearish China-economy view and bullish China-equities view share a page and read as a reversal he never made.

104 rows unverified

Machine-extracted and badged in review. They are visible on the board because this is a personal tool; run build.py --verified-only before anyone else sees it.

Gold set under-covers the new taxonomy

It was labelled before the subject taxonomy expanded, so it misses sector, commodity and macro rows — on one episode it holds 8 of 26. Tuning prompts against it now would score a correct pipeline as imprecise. A recall re-pass comes before Stage A–C work.

58 of 170 rows are stale

Older than 56 days. Most are the Weekend 1 gold set, so the board's most trustworthy rows are also its least current.

Coverage by show

ShowEpisodesRowsVerified
The Master Investor Podcast with Wilfred Frost35023
Monetary Matters with Jack Farley3390
Invest Like the Best (Colossus)41615
Bloomberg Talks21414
Excess Returns1130
BG2Pod with Brad Gerstner and Bill Gurley21010
The Meb Faber Show - Better Investing290
The Real Eisman Playbook170
Exchanges (Goldman Sachs)150
Capital Allocators (Ted Seides)144
Hard Lessons (Morgan Stanley)120
The Acquirers Podcast110

Yield drives whether a show earns its place on the list. A show ingested once produces orphan speakers; a show ingested repeatedly builds a book worth browsing.

Repo

Commits
15
Latest
8455dbb 2026-08-31 extract: Warren Pies on Monetary Matters 2026-08-26 — 11 candidate rows
Rebuild
cd ledger/site && python build.py — regenerates the board and this page from the claim files.

Built 2026-08-31 from the repo. 35 of 170 rows carry a stated recording date; the rest fall back to publication date.