Case Study

FIL+ Governance Framework: A Multi-Track Approach to Funding Allocation

Designing credible, capture-resistant governance for Filecoin's storage incentive program across public-goods data, enterprise clients, and treasury allocation.

Tanisha Katara — Senior Governance Consultant, Filecoin

Background

Filecoin is a decentralised storage network that pays storage providers block rewards for reliably holding data over time. FIL+ is Filecoin's verified-storage program: it boosts block rewards for providers who store data that has been vouched for as useful — originally public datasets, and increasingly enterprise workloads. A set of badgeholders attests to the value of incoming data, and a treasury funds subsidies that make verified deals economically competitive. As the program grows, deciding who gets to attest, how subsidies are prioritised, and how much goes to public-goods versus commercial storage become the central governance questions.

The Challenge

Scaling FIL+ Without Fragmenting Governance

FIL+ sits at the intersection of public-goods data and enterprise-driven storage demand. As the program scales, the core challenge is no longer whether to support datasets, but how to allocate incentives and funding credibly across fundamentally different use cases.

A single governance model risks either over-indexing on commercial viability or under-serving the public-goods mission. The framework needed to answer three questions simultaneously: how to split treasury funds, how to evaluate public datasets at scale, and how to assess enterprise proposals with the nuance they require.

This exploration tested whether combining AI signals, expert judgment, and structured governance could improve FIL+ funding outcomes without increasing capture risk.

FOC market architecture diagram showing the relationship between public open data, enterprise storage, and treasury allocation pools

The Approach

Three Tracks, Three Models

The framework is built on three validated assumptions, each of which maps directly to a governance track and a recommended decision model.

1
Public open data is transparent and machine-verifiable — well-suited to AI-assisted evaluation.
2
Enterprise data requires commercial judgment (SLAs, market impact, viability) — better evaluated by domain-expert juries.
3
Treasury allocation benefits from mixed representation to prevent capture by either commercial or public-goods factions.
Track 1

Treasury Allocation

Admissions Committee

How funds are split between tracks

Treasury allocation is a meta-governance decision that shapes the overall direction of FIL+ incentives. A standing committee with representatives from both the Public Open Data and Enterprise tracks proposes allocation percentages, which are then reviewed by a mixed jury and ratified by badgeholder voting.

Mixed representation is the key design lever here. By requiring input from both tracks, treasury decisions reflect the full ecosystem's interests and prevent capture by either the commercial or public-goods faction.

Treasury allocation governance flow: proposal, jury review, badgeholder voting, disbursement
Track 2

Public Open Data

Smart Jury + AI

Evaluating datasets at scale

Public data is inherently transparent and machine-readable, making AI evaluation tractable. The AI can verify dataset integrity, assess completeness, and compare against existing public datasets — tasks that are significantly harder with proprietary enterprise data.

A jury-based evaluation system is augmented with AI rubric scoring and median jury alignment. The AI provides an objective assessment layer while human jurors bring contextual judgment that algorithms may miss. Scores are combined using a consensus mechanism selected during Phase 1.

Public Open Data governance flow: submission, parallel evaluation by jury and AI, consensus logic, decision
Track 3

Enterprise Data

Simple Jury (Ranked Choice)

Commercial proposals require human judgment

Enterprise proposals involve proprietary information, nuanced commercial judgment, and relationship context that AI cannot reliably assess. A small, trusted jury of domain experts evaluates proposals and ranks them using Ranked Choice Voting.

Enterprise applicants post a bond to participate. The bond is refunded upon proposal acceptance and burned upon rejection, creating a direct incentive for high-quality submissions and positively impacting FIL token value accrual.

Enterprise governance flow: submission with bond, jury evaluation, ranked choice voting, decision with bond resolution

Key Design Decisions

Four Insights That Shaped the Framework

Each decision below surfaced during the design process and had a material impact on the final governance architecture.

AI evaluation is asymmetric by design

Public open data is machine-verifiable: dataset integrity, completeness, and coverage can be assessed algorithmically. Enterprise data is not. Commercial viability, SLA realism, and competitive positioning require the kind of contextual, relationship-aware judgment that current AI systems cannot reliably provide. The decision to apply AI selectively — to Track 2 only — was deliberate, not a limitation.

Five models for combining AI and human scores

The framework does not prescribe a single AI-Human alignment mechanism. Instead, five well-researched models are candidates for Phase 1 evaluation. The recommended starting point is AI as Partner, where AI and jury scores are computed independently and then combined.

ModelHow it works
AI as CheckerAI validates jury evaluations for internal consistency
AI as MakerJury reviews and refines AI's initial evaluation
AI as PartnerIndependent scores combined — recommended for Phase 1
AI as Devil's AdvocateAI stress-tests jury decisions by challenging assumptions
AI as Tie BreakerAI weighs in only when jury reaches no clear consensus

Prediction markets as a feedback loop

Traditional governance voting has no feedback loop: jurors have no incentive for accurate judgment, and there is no mechanism to surface private information. Prediction markets address both problems. Jurors who are consistently aligned with both the AI rubric (on objective criteria) and the median score (on collective alignment) are identified as more reliable and incentivized accordingly. Divergences reveal where AI misses context that humans catch.

Smart Jury Prediction Market scatter plot: AI Rubric score (x-axis) vs. Median Jury Score (y-axis), showing correlation between algorithmic and human judgment

The enterprise bond as a quality signal

Enterprise applicants are required to post a bond before their proposal enters jury review. The bond is refunded if the proposal is accepted, and burned if rejected. This mechanism does two things: it filters out low-effort submissions before they consume jury time, and it aligns proposer incentives with proposal quality. The burn-on-rejection path also contributes to FIL token value accrual, creating a secondary ecosystem benefit.

Impact

Outcomes and Next Steps

The framework delivers four concrete governance outcomes and a phased path to implementation.

Capture-resistant allocation

Mixed-track treasury committees prevent either commercial or public-goods factions from dominating funding decisions.

AI-augmented evaluation at scale

Public dataset review scales without proportional jury growth, while preserving human override on edge cases.

Incentive-aligned enterprise onboarding

The bond mechanism filters low-quality proposals and creates a direct link between submission quality and token value.

Modular, evolvable governance

Each track operates independently. Consensus mechanisms and jury composition can be tuned per-track without redesigning the system.

Implementation Path

1

Phase 1 — Governance Design & Validation

Finalize the three-track model. Run structured pilots of AI-Human consensus mechanisms on a representative set of public dataset proposals. Recruit and onboard jury candidates for both Public Open Data and Enterprise tracks.

2

Phase 2 — FIP Process & Coordination

Draft and submit the required FIPs covering QAP changes, PoRep market enablement (sector notification), and fee governance via FRC. Coordinate with the broader Filecoin governance community for review and feedback.

3

Phase 3 — Execution

Execute treasury allocation under the new model, with monthly reexamination for the first quarter and quarterly cadence thereafter.