6. The Chinwag Way & Scrimmages

6.1 Academic Foundations of The Chinwag Way

chinwag way audit --repo .

Six peer-reviewed multi-agent, software engineering, and game-theoretic foundations behind The Chinwag Way methodology.

6.1 Academic Foundations of The Chinwag Way

chinwag way audit --repo .

Most AI coding tools optimize a single metric: how fast one developer can generate lines of code in a local editor. Yet empirical studies across enterprise engineering teams show that uncoordinated AI code generation increases PR review bottlenecks, cross-branch merge conflicts, and 7-day post-merge revert rates.

The Chinwag Way is an opt-in team operating methodology grounded in six peer-reviewed research papers spanning multi-agent cooperation, human-AI pair programming, and multi-objective reinforcement learning.


The 6 Peer-Reviewed Foundations

#Research PillarEmpirical FindingArchitectural Realization in Chinwag
1Stanford CooperBench (2025/2026)Uncoordinated multi-agent coding degrades pass rates by 39% when two agents touch shared repository state without explicit boundary contracts.Disjoint --file-scope leases (WP-02), isolated Shadow jj workspaces (WP-01), and <20ms .cwcard.png boundary handoffs.
2Harvard Business School / Jellyfish Field StudyRaw lines-of-code (LOC) and PR count reward noisy churn; true productivity requires measuring downstream review cost and post-merge code survival.7-Day Code Survival (SURV) weighted at 0.30 of every task's composite quality score.
3Pairit Heterogeneous PairingPairing a single model family with itself (GPT reviewing GPT, or Claude reviewing Claude) misses correlated blind spots (+34% shared error rate).Mandatory cross-family Driver + Reviewer pairing (chinwag pair / chinwag triad, WP-03).
4Information-Theoretic EmpowermentAutonomous agents perform best when given high local action freedom inside a strictly bounded causal horizon.CustomerAutonomyFloorV1, p90 <= 150 LOC atomic task sizing, and local H1..H5 pre-handoff hooks.
5MO-GRPO (Multi-Objective Group Relative Policy Optimization)Optimizing a single scalar reward causes reward hacking; balancing orthogonal objectives (Quality, Review Velocity, Collaboration, Survival) prevents gaming.Chinwag Chips (◈) computed over a 4-objective normalized vector (QUAL, REV, COLL, SURV).
6High-Tempo Team Calibration DrillsShort, structured boundary-handoff scrimmages build shared mental models across mixed human + AI squads faster than static documentation.<20ms .cwcard.png sealed handoff drills and 4-minute Friday Captionizm calibration scrimmages.

Simple, Predictable Team Pricing (150 Cr per 4-Minute Calibration Scrimmage)

The Chinwag Way is 100% opt-in per repository or squad:

  • 150 Credits ($0.96 on Growth Pack · $0.75 on Org Volume) per 4-minute squad calibration scrimmage in an isolated Git fork (MO-GRPO Chips ◈ ledger, cryptographic policy enforcement, and 7-day SURV tracking).
  • 0 Cr on idle days, weekends, and holidays · if no scrimmages or handoffs run, zero credits are deducted.
  • Sealed 1:1 Handoffs & Boundary Audits (.cwcard.png): 50 Credits ($0.32 Growth) per two-round sealed 1:1 handoff & tombstone/invariant audit (100 Credits per Cross-model Pair sandbox; 150 Credits per Cross-model Triad sandbox).
# Enable The Chinwag Way in chinwag.workflow.yaml
chinwag_way:
  enabled: true
  chips_ledger: true
  survival_window_days: 7
  friday_calibration_scrimmage: true

Next Steps