pareto crewing v2: one dial in dollars, a grade no model can move, and the paper behind it by santoshkumarradha · Pull Request #1123 · Agent-Field/CodeAF

docs: the figures for pareto crewing v2, from real runs and calibrated simulation

santoshkumarradha

and

claude

committed

bffdbb1

docs: pareto crewing, rewritten around one dial in dollars

santoshkumarradha

and

claude

committed

ba1d6f0

changes: the #1123 entry for the pareto crewing rewrite

santoshkumarradha

and

claude

committed

96d2192

pool: a model-free grade of the landing — fmt, build, vet, touched tests

santoshkumarradha

and

claude

committed

0b041c2

crewpick: read frugal and balanced off the front at a stake in dollars, price the seats on what tasks spent

santoshkumarradha

and

claude

committed

38e8f50

pool: one scoreboard — acceptable per source beside the judge's opinion, installs on the floor, shapes off the ledger

santoshkumarradha

and

claude

committed

0ba6c62

config: price the front in dollars per million, on learned shapes, from the acceptable metric

santoshkumarradha

and

claude

committed

f27c4ff

chat: grade every landing before asking a judge, and learn the seats' shapes at start-up

santoshkumarradha

and

claude

committed

9dafb6a

manual, changes: what balanced means in dollars and how a task is graded

santoshkumarradha

and

claude

committed

ce06777

changes: the relay is not a surface word; the entry is chat and docs

santoshkumarradha

and

claude

committed

4b27b43

changes: the installs floor is #1127's

santoshkumarradha

and

claude

committed

317b81d

paper, data: a per-role token split and a catalog snapshot, both text-free

santoshkumarradha

and

claude

committed

70456d8

paper, figures: the three measured experiments, and every number they quote

santoshkumarradha

and

claude

committed

1caa40c

paper: rewrite as a general method paper, with an experiments section

santoshkumarradha

and

claude

committed

90afa0d

crewpick: learn the seat's share on the log scale, which is what the cost experiment says predicts a bill

santoshkumarradha

and

claude

committed

9910a99

grade: fail a landing that changed nothing, fail a removed test file, stamp the grader version

santoshkumarradha

and

claude

committed

fffb280

paper: headline figure script, anatomy card, narrative patch (codeaf-04 draft)

santoshkumarradha

and

claude

committed

5d843b1

paper: headline loader reads regrade.jsonl, acceptance = commits>0 and v2 pass

santoshkumarradha

and

claude

committed

7af2786

test: the sweep test reads the judge's cells, because a sweep has no workspace to grade

santoshkumarradha

and

claude

committed

47785ef

paper: an appendix on how the shipped pool actually works

santoshkumarradha

and

agentfield-bot

committed

6278928

paper: the appendix grades at version 2 and says how the headless doors leave a landing

santoshkumarradha

and

claude

committed

35832a5

paper: the appendix after the accuracy pass; grade: five stage words

santoshkumarradha

and

claude

committed

1dec329

Navigation Menu

Repository navigation

pareto crewing v2: one dial in dollars, a grade no model can move, and the paper behind it - #1123

pareto crewing v2: one dial in dollars, a grade no model can move, and the paper behind it#1123
santoshkumarradha wants to merge 22 commits into
santos/dev2from
design/pareto-crewing-v2

Commits on Sep 18, 2026

docs: the figures for pareto crewing v2, from real runs and calibrated simulation

docs: pareto crewing, rewritten around one dial in dollars

changes: the #1123 entry for the pareto crewing rewrite

pool: a model-free grade of the landing — fmt, build, vet, touched tests

crewpick: read frugal and balanced off the front at a stake in dollars, price the seats on what tasks spent

pool: one scoreboard — acceptable per source beside the judge's opinion, installs on the floor, shapes off the ledger

config: price the front in dollars per million, on learned shapes, from the acceptable metric

chat: grade every landing before asking a judge, and learn the seats' shapes at start-up

manual, changes: what balanced means in dollars and how a task is graded

changes: the relay is not a surface word; the entry is chat and docs

changes: the installs floor is #1127's

paper, data: a per-role token split and a catalog snapshot, both text-free

paper, figures: the three measured experiments, and every number they quote

paper: rewrite as a general method paper, with an experiments section

crewpick: learn the seat's share on the log scale, which is what the cost experiment says predicts a bill

grade: fail a landing that changed nothing, fail a removed test file, stamp the grader version

paper: headline figure script, anatomy card, narrative patch (codeaf-04 draft)

paper: headline loader reads regrade.jsonl, acceptance = commits>0 and v2 pass

test: the sweep test reads the judge's cells, because a sweep has no workspace to grade

paper: an appendix on how the shipped pool actually works

paper: the appendix grades at version 2 and says how the headless doors leave a landing

paper: the appendix after the accuracy pass; grade: five stage words