teleo/teleo-codex
8
0
Fork 0
Code Issues 22 Pull requests 18 Projects Releases Packages Wiki Activity Actions 1

Auto: schemas/contribution-weights.yaml | 1 file changed, 23 insertions(+), 17 deletions(-)

Browse source
This commit is contained in:
m3taversal 2026-03-14 19:04:20 +00:00 committed by Teleo Agents
parent 53bf7764b2
commit 88d93c9920
1 changed files with 23 additions and 17 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

40
schemas/contribution-weights.yaml
View file

@ -1,30 +1,36 @@
# Contribution Weights # Contribution Weights
# #
# Global policy for how much each contributor role counts toward weighted scores. # Global policy for how much each contributor role counts toward weighted scores.
# Used by the build pipeline (extract-graph-data.py) to compute weighted_score # Used by the build pipeline to compute weighted_score in contributors.json
# in contributors.json. Updated via PR — changes here affect all contributor profiles. # and Contribution Index (CI) in reward-mechanism.md.
# Updated via PR — changes here affect all contributor profiles.
# #
# Weights sum to 1.0. The build pipeline multiplies each contributor's role count # Weights sum to 1.0. The build pipeline multiplies each contributor's role count
# by the corresponding weight, then sums across roles. # by the corresponding weight, then sums across roles.
# #
# Current rationale (2026-03-11): # Current rationale (2026-03-14, revised from Rio's mechanism design brief):
# - Extraction is the current bottleneck and requires the most skill (reading sources, # - Sourcer = Extractor = Challenger at 0.25 each. This signals that finding
# separating signal from noise, writing prose-as-title). Highest weight. # the right source with a clear rationale, turning it into a structured claim,
# - Challenge is the quality mechanism — adversarial review catches errors that # and challenging existing claims are equally valuable acts. Equal weighting
# self-review cannot. Second highest. This also signals that the system values # prevents agent CI domination during bootstrap (agents fill extractor role,
# intellectual honesty over agreement: challenging bad claims is rewarded more # humans fill sourcer and challenger roles).
# than rubber-stamping good ones.
# - Sourcing discovers new information but is lower effort per instance.
# - Synthesis connects claims across domains — high value but rare. # - Synthesis connects claims across domains — high value but rare.
# - Review is essential but is partially automated via the eval pipeline. # - Review is essential but partially automated via the eval pipeline.
# #
# These weights WILL change as the collective matures. When challenges become # Review after 6 months of data. If sourcer contributions turn out to be
# the bottleneck (more claims than reviewers), challenger weight should increase. # low-effort, the weight is too high. If challengers produce disproportionate
# When synthesis becomes the primary value-add, synthesizer weight increases. # belief changes, the weight is too low. Weights are policy, not physics.
role_weights: role_weights:
sourcer: 0.15 sourcer: 0.25
extractor: 0.40 extractor: 0.25
challenger: 0.20 challenger: 0.25
synthesizer: 0.15 synthesizer: 0.15
reviewer: 0.10 reviewer: 0.10
# Contribution Index (CI) leaderboard weights
# See core/reward-mechanism.md for full spec
ci_weights:
belief_movers: 0.30
challenge_champions: 0.30
connection_finders: 0.40