The owner, 2026-10-18: "what if we strip the process down to just opus doing a research round & then picking. fable eats a lot of credit. if opus wants to delegate to sonnet or haiku, fine."
The day's venture, now: scout → opus's research round + pick → opus's launch plan → opus builds → checks → opus judges (honesty + the 1–10 score, with polish rounds) → sonnet drafts the launch posts → launch.
The research round (research, RESEARCH):
Opus gets the top six trending repos with the start of their READMEs (readme_of, unauthenticated GitHub like the scout; the rest are listed by name), every studio venture with its status and signups so far (track_record), and the playbook.
It writes what it found, lists TIME_VENTURE_FIELD (6) candidates with a verdict each, and picks one. There are no pitches and no vote.
A pick that repeats a studio venture gets one more go, told so. The card goes straight from scouted to picked, and the comment shows the research, the candidates and why it picked what it did.
Who does what (crew): the lead (TIME_VENTURE_LEAD, opus) writes the launch plan alone, so there is no synthesis call. It hands the launch drafts to its helper (TIME_VENTURE_HELPER, sonnet). The judge is opus.
Fable is out:CHAIR and JUDGE_SEAT default to opus. TIME_VENTURE_CHAIR and TIME_VENTURE_JUDGE are already set to opus on this repo, so fable was out of tonight's run even before this lands.
The board stays for batches, and TIME_VENTURE_LEAD=board brings it back for the daily venture. A card already at pitched still goes to the board.
What it saves: today the deliberation (pitches, votes, three plans, fable's synthesis, drafts and judging) was about $2.70 of the day's $23.66. Fable alone was $2.23. Builds were most of the rest. Per venture this goes from about a dozen deliberation calls to three opus calls and one sonnet call.
Tests: new test 79c covers:
crew: lead, helper, a missing helper, a batch's child, TIME_VENTURE_LEAD=board
track_record
the research round through advance: only opus is asked; six READMEs and the rest as a list; the track record and the candidate count are in the prompt; a repeat retries with a note; the pick's slug is cleaned and non-string fields are dropped; the beat is picked by the lead
research returning None when there is no fresh pick
Test 79 now pins opus as chair and judge. test_time.py passes locally (ALL OK 79c). There is no workflow change.
The owner, 2026-10-18: "what if we strip the process down to just opus doing a research round & then picking. fable eats a lot of credit. if opus wants to delegate to sonnet or haiku, fine."
Closes #165
**The day's venture, now:** scout → **opus's research round + pick** → opus's launch plan → opus builds → checks → opus judges (honesty + the 1–10 score, with polish rounds) → **sonnet drafts the launch posts** → launch.
- **The research round** (`research`, `RESEARCH`):
- Opus gets the top six trending repos with the start of their READMEs (`readme_of`, unauthenticated GitHub like the scout; the rest are listed by name), every studio venture with its status and signups so far (`track_record`), and the playbook.
- It writes what it found, lists `TIME_VENTURE_FIELD` (6) candidates with a verdict each, and picks one. There are no pitches and no vote.
- A pick that repeats a studio venture gets one more go, told so. The card goes straight from `scouted` to `picked`, and the comment shows the research, the candidates and why it picked what it did.
- **Who does what** (`crew`): the lead (`TIME_VENTURE_LEAD`, opus) writes the launch plan alone, so there is no synthesis call. It hands the launch drafts to its helper (`TIME_VENTURE_HELPER`, sonnet). The judge is opus.
- **Fable is out:** `CHAIR` and `JUDGE_SEAT` default to opus. `TIME_VENTURE_CHAIR` and `TIME_VENTURE_JUDGE` are already set to opus on this repo, so fable was out of tonight's run even before this lands.
- **The board stays for batches,** and `TIME_VENTURE_LEAD=board` brings it back for the daily venture. A card already at `pitched` still goes to the board.
**What it saves:** today the deliberation (pitches, votes, three plans, fable's synthesis, drafts and judging) was about $2.70 of the day's $23.66. Fable alone was $2.23. Builds were most of the rest. Per venture this goes from about a dozen deliberation calls to three opus calls and one sonnet call.
**Tests:** new test 79c covers:
- `crew`: lead, helper, a missing helper, a batch's child, `TIME_VENTURE_LEAD=board`
- `track_record`
- the research round through `advance`: only opus is asked; six READMEs and the rest as a list; the track record and the candidate count are in the prompt; a repeat retries with a note; the pick's slug is cleaned and non-string fields are dropped; the beat is `picked` by the lead
- `research` returning None when there is no fresh pick
Test 79 now pins opus as chair and judge. `test_time.py` passes locally (`ALL OK 79c`). There is no workflow change.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
https://claude.ai/code/session_017K4vyY3yUD3GsvoZaAtKkD
One seat, TIME_VENTURE_LEAD (opus), runs the day's venture: a research round over the top trending repos (with the
start of their READMEs) and the studio's track record, TIME_VENTURE_FIELD candidates with a verdict each, one pick;
no pitches, no vote. It writes the launch plan alone and judges the build, and hands the launch drafts to its helper
(TIME_VENTURE_HELPER, sonnet). The chair and judge default to opus. Batches keep the board; TIME_VENTURE_LEAD=board
brings it back for the daily venture.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017K4vyY3yUD3GsvoZaAtKkD
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
The owner, 2026-10-18: "what if we strip the process down to just opus doing a research round & then picking. fable eats a lot of credit. if opus wants to delegate to sonnet or haiku, fine."
Closes #165
The day's venture, now: scout → opus's research round + pick → opus's launch plan → opus builds → checks → opus judges (honesty + the 1–10 score, with polish rounds) → sonnet drafts the launch posts → launch.
research,RESEARCH):readme_of, unauthenticated GitHub like the scout; the rest are listed by name), every studio venture with its status and signups so far (track_record), and the playbook.TIME_VENTURE_FIELD(6) candidates with a verdict each, and picks one. There are no pitches and no vote.scoutedtopicked, and the comment shows the research, the candidates and why it picked what it did.crew): the lead (TIME_VENTURE_LEAD, opus) writes the launch plan alone, so there is no synthesis call. It hands the launch drafts to its helper (TIME_VENTURE_HELPER, sonnet). The judge is opus.CHAIRandJUDGE_SEATdefault to opus.TIME_VENTURE_CHAIRandTIME_VENTURE_JUDGEare already set to opus on this repo, so fable was out of tonight's run even before this lands.TIME_VENTURE_LEAD=boardbrings it back for the daily venture. A card already atpitchedstill goes to the board.What it saves: today the deliberation (pitches, votes, three plans, fable's synthesis, drafts and judging) was about $2.70 of the day's $23.66. Fable alone was $2.23. Builds were most of the rest. Per venture this goes from about a dozen deliberation calls to three opus calls and one sonnet call.
Tests: new test 79c covers:
crew: lead, helper, a missing helper, a batch's child,TIME_VENTURE_LEAD=boardtrack_recordadvance: only opus is asked; six READMEs and the rest as a list; the track record and the candidate count are in the prompt; a repeat retries with a note; the pick's slug is cleaned and non-string fields are dropped; the beat ispickedby the leadresearchreturning None when there is no fresh pickTest 79 now pins opus as chair and judge.
test_time.pypasses locally (ALL OK 79c). There is no workflow change.🤖 Generated with Claude Code
https://claude.ai/code/session_017K4vyY3yUD3GsvoZaAtKkD