Documentation
A comprehensive guide to the Meridian coaching algorithm, hindsight protocols, and growth verification engine.
Platform Overview
Meridian is an AI-powered developer coaching platform purpose-built for persistent growth. Traditional tools score you once and forget you exist. Meridian builds an evolving memory of your development session history across GitHub, LeetCode, and Codeforces to identify silent growth patterns and predict future blockers.
At the core of the platform is the Multi-Signal Triangulation Score (MSTS), a composite metric that evaluates six independent behavioral layers. These signals are synced via the Hindsight SDK and analyzed by Groq LLMs to produce a high-fidelity map of your technical trajectory.
MSTS Composite
Six-layer triangulation ensures every growth signal is verified and weighted properly.
Hindsight Memory
Persistent storage of dev sessions allows the coach to remember your progress over years.
Trajectory Delta
Real-time mapping of your growth velocity relative to previous memory snapshots.
Multi-Signal Triangulation Score (MSTS)
Each contributor is scored from 0 to 100 based on the weighted sum of five independent signal layers. The triangulation of multiple uncorrelated signals ensures manipulation of any single metric has negligible effect on the final score.
Temporal Decay Weighting
Commits and PRs are decayed exponentially. A commit from last week retains 100% weight while a commit from 3 years ago retains approximately 10%. This rewards active maintainers keeping projects alive.
Downstream Blast Radius
A breadth-first search of the ecosystem dependency graph. If a high-traffic project depends on your utility library, you receive a weighted portion of its user base. Depth 1 = 1.0x, Depth 2 = 0.6x, Depth 3 = 0.3x.
Code Survival Rate
Measures what percentage of original lines written by a contributor remain unmodified after one year. An 85% survival rate proves structurally sound code that does not require immediate rewriting.
Issue Resolution Velocity
Tracks how quickly critical bugs are resolved, weighted by severity label. A P0 bug closed in 2 hours yields a high score. A P3 feature request closed in 30 days yields minimal impact.
Cross-Ecosystem Portability
Detects if libraries are ported or referenced across multiple package managers including npm, PyPI, Maven, and crates.io. Cross-ecosystem adoption proves fundamental architectural value.
Proof-of-Build Verification
The Proof-of-Build protocol serves as the primary Sybil prevention mechanism. No contribution receives a score unless it is accompanied by a verified CI/CD execution trace. The system dynamically queries the GitHub Actions API to confirm that a specific workflow run physically compiled and executed with a success status.
If no verified execution trace exists, the multiplier is exactly 0 — completely zeroing out any accumulated score. This makes it impossible to inflate metrics through auto-generated code or bot-driven commits without actually burning compute and producing functional software.
Hindsight-Weighted Trajectory Scaling
Standard developer metrics are fundamentally broken because they focus on output volume. Meridian addresses this with a Trajectory Scaling mechanism that replaces raw commit counts with the exponential growth of a developer's MSTS impact score over time.
Volume Metrics (Flawed)
Meridian Trajectory Scaling
Funded Project Case Studies
Real organizations leveraging Meridian to sustainably track and grow their engineering talent.
Vercel
Next.jsMeridian coached 42 infrastructure developers who maintained the core webpack integration and streaming SSR pipeline. Developers saw an average 3.4x increase in hindsight recall accuracy compared to traditional documentation models.
Protocol Labs
libp2pHindsight-Weighted Trajectory Scaling allocated 96% of the coaching compute to foundational protocol developers, effectively surfacing silented growth patterns in P2P discovery logic.
Supabase
PostgreSQLRecall Sync verification caught 14 behavioral drifts in async query handling. Only developers with verified activity traces received advanced challenges, saving the cohort an estimated 8,200 hours in redundant learning.