Audience
acquisition
22 of 61 mentions (36%) appear in a retrospective top-download list.
Sam AltmanPaul Graham
Peter Thiel
CONVERSATIONS WITH TYLER / 2019-2025
Celebrity opens
the door.
Specificity earns
the room.
A seven-year read of which conversations attracted the largest audience, which became loved, and which Tyler himself found most alive.
- 07
- retrospectives
- 61
- retrospective mentions
- 62
- pre-registered dimensions
THE MAIN FINDING
There are three different ways for an episode to work.
The evidence supports a narrower claim than the first version made. In this selected retrospective panel, fame is strongly enriched among download leaders. History, obscurity, and specialist expertise are enriched among the love signals. Those are descriptive contrasts, not causal effects.
Audience
Paul Graham
Peter Thiel
Audience
affection
24 of 61 mentions (39%) carry a direct listener-favorite or retention signal.
Lazarus LakeStephen Kotkin
Vishy Anand
Tyler’s
enjoyment
24 of 61 mentions (39%) receive the strongest Tyler-praise code. This is a host outcome, not an audience proxy.
KnausgårdDonald Lopez
Any Austin
THE SEVEN-YEAR LEDGER
Top audio downloads and two different forms of love.
| Year | Most listened | Audience-love signal | Tyler’s pick | What matters |
|---|---|---|---|---|
| 2019 | Jordan Peterson | Knausgård | Knausgård | The rare literary episode that Tyler called the CWT highlight of the year. |
| 2020 | Matt Yglesias | John Brennan | John Brennan | The UFO exchange made Brennan Tyler’s personal highlight. |
| 2021 | Amia Srinivasan | Amia Srinivasan | Amia Srinivasan | Polarizing, highly retained, and one of Tyler’s favorites. |
| 2022 | Thomas Piketty | Mary Gaitskill | Vaughn Smith / Roy Foster | Debate pulled downloads. Genuineness created the more durable signal. |
| 2023 | Paul Graham | Lazarus Lake | Lazarus Lake / Katherine Rundell | Lazarus Lake became the clearest audience-love episode in the sample. |
| 2024 | Peter Thiel | Stephen Kotkin | Stephen Kotkin | A history conversation entered the pantheon immediately. |
| 2025 | Sam Altman | David Commins | Any Austin / Donald Lopez | Commins is the year’s most useful outlier. |
WHAT THE PANEL ACTUALLY SAYS
Descriptive enrichment, in percentage points.
Every bar is the observed outcome rate when a feature is present minus its rate when absent. The bracket is an 80% bootstrap interval from 5,000 deterministic resamples. It is not a regression coefficient, a p-value, or a causal estimate.
Acquisition
Top-download-list enrichmentAudience affection
Listener-favorite enrichmentStrong Tyler pick
Highest retrospective praise codeTHE OUTLIERS ARE THE PRODUCT CLUES
High reach and high love rarely arrive by the same route.
David Commins is the cleanest surprise. An obscure Saudi Arabia episode finished seventh in a celebrity-heavy 2025 top ten. Lazarus Lake got there through warmth and recommendation. Amia Srinivasan got there through conflict that listeners stayed to hear.
The axes below are relative within this retrospective-derived sample, not universal podcast scores.
AUDIENCE LOVE
AUDIENCE ACQUISITION
Hover a pointEach dot is an episode with an unusual pattern.
READ THE WEIRD ONES CLOSELY
David Commins
Tyler expected a Saudi Arabia episode to be near the bottom. It finished seventh.
Lazarus Lake
Unusual life, warmth, competition, and a premise that travelled by recommendation.
Amia Srinivasan
Conflict did not create abandonment. It created retention.
Stephen Kotkin
The all-timer pattern: serious prep, difficult material, and obvious host pleasure.
Sam Altman
The clearest acquisition win, with a different mechanism from attachment.
Masaaki Suzuki
A specialist Bach episode that exposes how subject knowledge and accent can shape starts.
THE 62-DIMENSION CODEBOOK
Every dimension, its operational definition, and its data status.
Thirty-five features were coded from the annual retrospectives. Twenty-seven transcript and audio measures are pre-registered here but explicitly unobserved. The table never turns an unmeasured quantity into a zero.
Reading a cell: Δ is feature-present rate minus feature-absent rate, in percentage points. Brackets are 80% bootstrap intervals. “Sparse” means fewer than four observations in a comparison group. “Not observed” means the available source corpus cannot identify the feature.
| Family | Dimension and definition | Observed n | Acquisition Δ | Affection Δ | Tyler Δ | Status |
|---|---|---|---|---|---|---|
| Guest and distribution | Guest fameGuest is coded high-fame in the retrospective. | 21/61 | +82pp[73, 93] | -52pp[-65, -40] | -39pp[-53, -24] | Coded |
| Guest and distribution | Repeat guestGuest had appeared on CWT before this episode-year. | 3/61 | +67ppsparse | -41ppsparse | -41ppsparse | Sparse |
| Topic | AIPrimary subject includes artificial intelligence. | 4/61 | +42pp[11, 70] | -42pp[-51, -33] | -42pp[-51, -33] | Coded |
| Topic | Politics / policyPrimary subject includes politics, policy, intelligence, defense, race, or geopolitics. | 21/61 | +24pp[8, 42] | -16pp[-33, 0] | -2pp[-19, 15] | Coded |
| Topic | Economics / financePrimary subject includes economics, finance, labor, or investing. | 8/61 | +2pp[-23, 25] | -2pp[-26, 23] | +12pp[-13, 37] | Coded |
| Topic | HistoryPrimary subject includes history, archaeology, or a historical geography. | 11/61 | -33pp[-46, -18] | +30pp[8, 51] | +7pp[-15, 29] | Coded |
| Topic | Science / psychologyPrimary subject includes science, psychology, or statistics. | 9/61 | +23pp[-1, 46] | -20pp[-39, 0] | -20pp[-40, 0] | Coded |
| Topic | Visual arts / poetryPrimary subject includes visual art, painting, or poetry. | 2/61 | -37ppsparse | -41ppsparse | -41ppsparse | Sparse |
| Topic | LiteraturePrimary subject includes literature, translation, or fiction. | 5/61 | -39pp[-48, -32] | +44pp[19, 68] | +22pp[-6, 51] | Coded |
| Topic | MusicPrimary subject includes music, Bach, the Beatles, or classical music. | 5/61 | -39pp[-48, -30] | +1pp[-28, 30] | +1pp[-26, 30] | Coded |
| Topic | ReligionPrimary subject includes religion or Buddhism. | 1/61 | -37ppsparse | +62ppsparse | +62ppsparse | Sparse |
| Topic | Sports / competitionPrimary subject includes sports, chess, ultramarathon, or competition. | 3/61 | -38ppsparse | +29ppsparse | +29ppsparse | Sparse |
| Topic | TechnologyPrimary subject includes technology, AI, YouTube, or aviation. | 12/61 | +38pp[19, 57] | -28pp[-45, -12] | -18pp[-36, 1] | Coded |
| Topic | GeographyPrimary subject includes a named region or country. | 5/61 | -18pp[-41, 8] | +66pp[57, 73] | +22pp[-6, 51] | Coded |
| Conversation design | Single subjectThe episode has one explicit intellectual object or question. | 2/61 | +14ppsparse | +63ppsparse | +63ppsparse | Sparse |
| Conversation design | Debate / pushbackRetrospective tags show serious disagreement or a combative exchange. | 6/61 | +15pp[-12, 43] | +12pp[-16, 40] | +31pp[3, 56] | Coded |
| Conversation design | ControversyThe episode was described as controversial or polarizing. | 0/61 | +nullppsparse | +nullppsparse | +nullppsparse | Sparse |
| Human texture | AuthenticityThe retrospective describes unguarded, unpolished, or non-performative answers. | 9/61 | -29pp[-44, -12] | +58pp[42, 73] | +58pp[41, 73] | Coded |
| Human texture | Emotional disclosureThe retrospective records personally consequential disclosure. | 1/61 | -37ppsparse | +62ppsparse | +62ppsparse | Sparse |
| Human texture | HumorThe retrospective records humor as a meaningful feature. | 0/61 | +nullppsparse | +nullppsparse | +nullppsparse | Sparse |
| Human texture | NoveltyThe retrospective signals a surprising premise, person, or subject. | 3/61 | -3ppsparse | -7ppsparse | +29ppsparse | Sparse |
| Conversation design | Practical utilityThe retrospective tags a concrete, usable implication. | 3/61 | +33ppsparse | -41ppsparse | -7ppsparse | Sparse |
| Conversation design | Expertise depthThe guest is tagged for specialist knowledge or a singular subject. | 19/61 | -37pp[-50, -24] | +27pp[9, 44] | -3pp[-20, 14] | Coded |
| Guest and distribution | ObscurityThe retrospective frames the guest as underrated or outside the obvious fame pool. | 22/61 | -49pp[-62, -37] | +45pp[30, 60] | +17pp[0, 34] | Coded |
| Guest and distribution | Institutional accessThe guest offers unusual access to an institution or state actor. | 6/61 | +15pp[-12, 42] | -44pp[-53, -35] | -25pp[-45, -3] | Coded |
| Guest and distribution | Female guestGuest gender, hand-coded only for descriptive balance checks. | 8/61 | -13pp[-34, 9] | -2pp[-26, 21] | -2pp[-25, 22] | Coded |
| Production | Accent / audio frictionThe retrospective calls out accent comprehension or rough audio. | 2/61 | +14ppsparse | +63ppsparse | +63ppsparse | Sparse |
| Human texture | RapportThe retrospective identifies host-guest rapport as salient. | 3/61 | -38ppsparse | +29ppsparse | +64ppsparse | Sparse |
| Human texture | Host enthusiasmTyler's retrospective language scores the exchange as a strong personal success. | 26/61 | -36pp[-50, -22] | +66pp[53, 78] | n/aoverlaps outcome | Coded |
| Conversation design | Preparation intensityRequires episode transcript and prep materials, not coded in this panel. | n/a | n/a | n/a | n/a | Not observed |
| Conversation design | Narrative arcRequires full transcript or listening notes, not coded in this panel. | n/a | n/a | n/a | n/a | Not observed |
| Conversation design | Unexpected endingThe retrospective identifies a closing turn that readers or listeners repeat. | 1/61 | -37ppsparse | +62ppsparse | +62ppsparse | Sparse |
| Conversation design | Fast answersThe retrospective describes unusually quick, concrete answers. | 3/61 | -38ppsparse | +64ppsparse | +64ppsparse | Sparse |
| Production | Long runtimeThe episode is tagged long relative to the CWT format. | 1/61 | -37ppsparse | +62ppsparse | +62ppsparse | Sparse |
| Production | Live / panel formatThe episode was recorded live or as a panel. | 1/61 | +65ppsparse | -40ppsparse | -40ppsparse | Sparse |
| Guest and distribution | Social distributionRetrospective tags document a notable social or YouTube distribution channel. | 2/61 | +14ppsparse | +11ppsparse | +11ppsparse | Sparse |
| Production | Production qualityThe retrospective flags a material audio-quality issue or its absence. | 60/61 | -65ppsparse | -62ppsparse | -62ppsparse | Sparse |
| Transcript measurement | Question densityTyler questions per 1,000 transcript words. | n/a | n/a | n/a | n/a | Not observed |
| Transcript measurement | Follow-up densityFollow-up questions per 1,000 Tyler words. | n/a | n/a | n/a | n/a | Not observed |
| Transcript measurement | Why-question rateShare of Tyler questions beginning with or functioning as why questions. | n/a | n/a | n/a | n/a | Not observed |
| Transcript measurement | Pressure-question rateRate of direct belief-probing questions, including “what do you really think?”. | n/a | n/a | n/a | n/a | Not observed |
| Transcript measurement | Production-function promptsCount of prompts that ask how an outcome is produced. | n/a | n/a | n/a | n/a | Not observed |
| Transcript measurement | First-person rateFirst-person pronouns per 1,000 speaker words. | n/a | n/a | n/a | n/a | Not observed |
| Transcript measurement | Hedge rateHedges per 1,000 speaker words. | n/a | n/a | n/a | n/a | Not observed |
| Transcript measurement | Certainty rateHigh-certainty terms per 1,000 speaker words. | n/a | n/a | n/a | n/a | Not observed |
| Transcript measurement | Counterfactual rateCounterfactual constructions per 1,000 speaker words. | n/a | n/a | n/a | n/a | Not observed |
| Transcript measurement | Numbers rateNumbers per 1,000 speaker words. | n/a | n/a | n/a | n/a | Not observed |
| Transcript measurement | Proper-noun densityNamed entities per 1,000 speaker words. | n/a | n/a | n/a | n/a | Not observed |
| Transcript measurement | Named-example rateConcrete named examples per 1,000 speaker words. | n/a | n/a | n/a | n/a | Not observed |
| Transcript measurement | Technical-jargon rateDomain terms per 1,000 speaker words. | n/a | n/a | n/a | n/a | Not observed |
| Transcript measurement | Plain-language rateShare of words below a pre-registered reading-complexity threshold. | n/a | n/a | n/a | n/a | Not observed |
| Transcript measurement | Metaphor rateHand-checked metaphoric constructions per 1,000 speaker words. | n/a | n/a | n/a | n/a | Not observed |
| Transcript measurement | Analogy rateExplicit analogies per 1,000 speaker words. | n/a | n/a | n/a | n/a | Not observed |
| Transcript measurement | Negative polarityNegative-sentiment terms per 1,000 speaker words. | n/a | n/a | n/a | n/a | Not observed |
| Transcript measurement | Positive polarityPositive-sentiment terms per 1,000 speaker words. | n/a | n/a | n/a | n/a | Not observed |
| Transcript measurement | Surprise markersSurprise expressions per 1,000 speaker words. | n/a | n/a | n/a | n/a | Not observed |
| Transcript measurement | Laughter markersTranscript laughter markers per 1,000 speaker words. | n/a | n/a | n/a | n/a | Not observed |
| Transcript measurement | Interruption rateOverlaps or interruptions per hour from speaker diarization. | n/a | n/a | n/a | n/a | Not observed |
| Transcript measurement | Short-answer rateGuest answers under a pre-registered word threshold. | n/a | n/a | n/a | n/a | Not observed |
| Transcript measurement | Long-sentence rateSentences above a pre-registered word threshold. | n/a | n/a | n/a | n/a | Not observed |
| Transcript measurement | Food / travel asidesFood or travel digressions per hour. | n/a | n/a | n/a | n/a | Not observed |
| Transcript measurement | Guest-specific lexiconDistinctive vocabulary after matching on topic and guest background. | n/a | n/a | n/a | n/a | Not observed |
A PROGRAMMING RULE, NOT A FORMULA
Use recognizable guests to acquire an audience. Use strange specialists to give it a reason to stay.
01
Keep a portfolio
Use a 70/30 mix of recognisable anchors and discovery bets. Fame is a legitimate distribution channel.
02
Package the question
Never market a specialist as obscure. Market why Saudi Arabia mattered, what Buddhism is actually like, or what a conductor hears.
03
Measure the chain
Track starts, ten-minute retention, completion, referral source, and unsolicited recommendation separately.
THE 2026 GUEST SLATE
Who should come next?
The first outreach should go to Michael Levin, John Jumper, Max Martin, Sarah Paine, Lisa Su, and Kevin Esvelt. They combine a clear premise with unusual expertise, strong preparation payoff, and enough 2026 relevance to travel.
793primary-source items
25audience-love picks
25Tyler-enjoyment picks
22names in both sets
Michael Levin
developmental biology
- Audience
- 85
- Tyler
- 96
John Jumper
AI biology
- Audience
- 95
- Tyler
- 94
Max Martin
music
- Audience
- 92
- Tyler
- 92
Sarah Paine
military history
- Audience
- 90
- Tyler
- 98
Lisa Su
semiconductors
- Audience
- 90
- Tyler
- 91
Kevin Esvelt
biosecurity
- Audience
- 90
- Tyler
- 98
THE FORECAST MOSTLY AGREES
22 guests clear both thresholds.
The overlap is the finding. Audience affection and Tyler’s enjoyment both reward depth, a narrow but generative subject, original material, and the visible payoff from preparation.
AUDIENCE ONLY / 3Aidan GomezMustafa SuleymanMira Murati
AUDIENCE LOVE + TYLER ENJOYMENT / 22
Lisa SuBrian CheskyJohn JumperTobi LütkeMax MartinMichael LevinKevin EsveltTed ChiangChris OlahPatrick Radden KeefeA. R. RahmanNgozi Okonjo-IwealaSarah PaineCaitlin TalmadgeAndrei SoldatovKori SchakeDawn SongRegina BarzilayPercy LiangHernan DiazKiana HayeriMisha Glenny
TYLER ONLY / 3Amor TowlesVenki RamakrishnanKatherine De Kleer
TWO RANKINGS, TWO JOBS
One slate is built to be loved. The other is built around Tyler’s curiosity.
| Rank | Guest | Score | Why it should travel |
|---|---|---|---|
| 01 | John JumperAI biology | 95 | Nobel-level breakthrough with a clean premise |
| 02 | A. R. Rahmanmusic | 93 | global audience and unusually broad cultural reach |
| 03 | Max Martinmusic | 92 | a universally known output with an unfamiliar maker |
| 04 | Kevin Esveltbiosecurity | 90 | high-stakes subject with a real moral voice |
| 05 | Misha Glennygeopolitics | 90 | broad relevance without pundit monotony |
| 06 | Sarah Painemilitary history | 90 | specialist intensity and recommendation value |
| 07 | Andrei SoldatovRussia and surveillance | 90 | rare on-the-ground expertise |
| 08 | Kiana Hayeriphotojournalism | 90 | first-hand reporting gives the episode narrative momentum |
| 09 | Ted Chiangfiction and technology | 90 | loyal, unusually discussion-prone audience |
| 10 | Lisa Susemiconductors | 90 | AI infrastructure with broad recognition |
| 11 | Chris OlahAI interpretability | 88 | one of the few technical AI conversations listeners would share |
| 12 | Regina BarzilayAI medicine | 88 | clear human payoff from frontier AI |
| 13 | Patrick Radden Keefenarrative nonfiction | 88 | an established long-form audience and vivid material |
| 14 | Ngozi Okonjo-Iwealatrade and development | 88 | global trade with an immediately legible institutional role |
| 15 | Tobi Lütkecommerce and software | 88 | credible builder with a distinct public voice |
| 16 | Brian Cheskyhospitality and product | 88 | consumer recognition plus a real operating story |
| 17 | Aidan Gomezenterprise AI | 88 | founder status plus the transformer origin story |
| 18 | Dawn Songcomputer security | 86 | AI and security without generic founder talk |
| 19 | Percy LiangAI research | 86 | a credible, less overexposed AI expert |
| 20 | Caitlin Talmadgesecurity studies | 86 | war and deterrence have immediate relevance |
| 21 | Mustafa SuleymanAI policy | 86 | large current AI audience and clear stakes |
| 22 | Hernan Diazfiction | 85 | a distinctive author likely to generate strong word of mouth |
| 23 | Michael Levindevelopmental biology | 85 | high recommendation value for curious listeners |
| 24 | Kori Schakesecurity studies | 85 | clear and timely foreign-policy conversation |
| 25 | Mira MuratiAI leadership | 84 | rare access to a central AI actor |
| Rank | Guest | Score | Why it suits Tyler |
|---|---|---|---|
| 01 | Kevin Esveltbiosecurity | 98 | technical depth and candid institutional criticism |
| 02 | Misha Glennygeopolitics | 98 | crime, technology, Central Europe, and history |
| 03 | Sarah Painemilitary history | 98 | precisely the kind of deep, surprising historical explanation Tyler prizes |
| 04 | Andrei SoldatovRussia and surveillance | 97 | the combination of regime knowledge and direct reporting is unusually CWT-compatible |
| 05 | Kiana Hayeriphotojournalism | 97 | Afghanistan, migration, aesthetics, and lived knowledge |
| 06 | Chris OlahAI interpretability | 97 | a rare chance to make a difficult frontier intelligible |
| 07 | Regina BarzilayAI medicine | 97 | medicine, algorithms, and personal scientific motivation |
| 08 | Dawn Songcomputer security | 97 | technical originality and a focused subject |
| 09 | Percy LiangAI research | 97 | excellent fit for questions on evaluation and the production function of science |
| 10 | Ted Chiangfiction and technology | 96 | fiction, computation, and moral philosophy reward slow questions |
| 11 | Patrick Radden Keefenarrative nonfiction | 96 | research method, state capacity, and narrative craft |
| 12 | Hernan Diazfiction | 96 | literature, capitalism, and form create unusually rich preparation payoff |
| 13 | Michael Levindevelopmental biology | 96 | the strongest pure Tyler-fit: unusual theory, experiment, and vocabulary |
| 14 | Amor Towlesfiction | 96 | historical imagination, Russia, and social observation |
| 15 | Venki Ramakrishnanbiology | 96 | science, institutions, India, and autobiography |
| 16 | Katherine De Kleerplanetary science | 95 | pure expertise and an alien world, with a clear single subject |
| 17 | John JumperAI biology | 94 | deep scientific explanation with unusually high preparation payoff |
| 18 | A. R. Rahmanmusic | 94 | composition, India, technology, and religious aesthetics |
| 19 | Ngozi Okonjo-Iwealatrade and development | 94 | development, Africa, trade rules, and state capacity create a high-preparation conversation |
| 20 | Caitlin Talmadgesecurity studies | 94 | technical strategy plus an accessible explanatory style |
| 21 | Kori Schakesecurity studies | 94 | institutional experience, history, and frank strategic judgment |
| 22 | Max Martinmusic | 92 | song construction, Sweden, and artistic production are classic CWT terrain |
| 23 | Lisa Susemiconductors | 91 | engineering, Taiwan, supply chains, and management reward preparation |
| 24 | Tobi Lütkecommerce and software | 91 | management, entrepreneurship, culture, and technology all travel well |
| 25 | Brian Cheskyhospitality and product | 82 | travel, cities, regulation, and design create many angles |
HOW THE FORECAST IS BUILT
Transparent weights, deliberately separate outcomes.
These 0 to 100 scores are structured editorial forecasts. The weights inherit the retrospective’s directional findings, while each candidate receives a 1 to 5 judgment on the underlying dimensions. They rank outreach. They do not estimate causal effects.
Audience love
Reach24%
Expertise depth22%
Single-subject coherence18%
Current relevance16%
Candor12%
Novelty8%
Tyler enjoyment
Expertise depth29%
Idiosyncrasy22%
Preparation payoff18%
Intellectual range13%
Current relevance10%
Anticipated rapport8%
Screening rule. Every shortlisted person had no prior CWT archive match and no material exact-name result in Marginal Revolution’s internal search at the time of the audit. That is a documented search result, not proof that Tyler has never encountered the person under a variant spelling, inside a link list, or offline.
METHOD AND SOURCES
The number is useful only when you can reconstruct it.
Unit and corpus. The panel has 61 retrospective episode-year mentions covering 58 distinct guests from the seven official 2019 to 2025 CWT retrospectives. It is a selected sample of episodes the show chose to discuss. It is not the full episode universe, so neither its rates nor its contrasts generalize mechanically to the show.
Outcomes. Acquisition means an episode was named in a retrospective top-download list. Audience affection means the retrospective supplied a listener-favorite, unusually strong retention, or equivalent direct listener-response signal. Strong Tyler pick means the hand-coded Tyler-enjoyment score equals 3, reserved for the strongest personal-praise language. These are three binary outcomes, reported separately because a download leader and a beloved conversation are not the same object.
Estimator. For each coded feature d and outcome Y, the reported statistic is Δd = mean(Y | d = 1) − mean(Y | d = 0). It is an unadjusted rate difference in percentage points. The interval is an 80% percentile bootstrap with 5,000 seeded resamples (seed 20260717). Intervals are withheld when either comparison group has fewer than four rows. The code and exact input files below reproduce every displayed number. Host enthusiasm is withheld from the Strong Tyler Pick comparison because the two codes overlap.
Prospective audit. The guest slate adds 793 primary-source items: 500 recent Marginal Revolution entries from its public sitemap and 293 CWT archive entries. Candidate names were screened against the CWT catalog and Marginal Revolution’s internal exact-name search. Current relevance was checked separately. The resulting shortlist was coded on twelve editorial dimensions before either ranking was produced.
What it does not identify. Features and outcomes are both hand-coded from the same retrospective documents. Selection into the panel is outcome-related. The historical table describes co-occurrence in a curated retrospective sample. The prospective scores are structured forecasts, not causal estimates. A defensible next stage would use the full episode universe, downloads at fixed ages, starts, retention, completion, promotion exposure, guest fame measured before release, release timing, and transcript or diarized-audio measures. Then it could estimate preregistered models with year and format controls.
Primary sources: 2019 · 2020 · 2021 · 2022 · 2023 · 2024 · 2025 · Marginal Revolution · CWT archive
Reproducibility: coded panel CSV · analysis output JSON · guest forecast CSV · 793-item source ledger · candidate exclusion screen · historical analysis script