The Scheduler

#The Scheduler names the bottleneck

The Scheduler is the Bottleneck Model. Each month it reads the chip supply chain across five stages and names the stage that binds.

The binding constraint of the AI economy sits at different points of the supply chain at different times. The Scheduler moves the Reserve toward the stage that binds by setting five stage weights.

Every read produces three outputs: a public memo, five stage weights on-chain and the next rotation date.

#Five stages cover the chain

Every eligible stock token maps to one stage. The mapping fixes which tokens stand for each stage.

AVGO, AMAT, LRCX and KLAC join the eligible set once Chainlink publishes their feeds. Eligibility is set out in The Reserve.

#The scorecard is published

The Scheduler reads a published scorecard of six inputs.

1. DRAM and HBM contract prices.
2. HBM and CoWoS lead times.
3. Fab and packaging capacity announcements.
4. Equipment order data.
5. Datacenter power constraints.
6. Earnings-call language for each stage.

Each memo explains the read of these inputs.

#A rulebook sets the weights

The rulebook maps the scorecard to five stage weights. It caps every stage at 50%.

The model does not set weights. It writes the memo that explains the read.

The model is a LoRA fine-tune on an open-weight base: Qwen3.8-27B (Apache 2.0) or DeepSeek V4 Flash (MIT).

#The memo precedes every rotation

Rotations run monthly on a fixed date. The memo publishes before every rotation.

The first memo publishes on the day trading opens. The first rotation executes 30 days later.

1. 01
   Read

   The scorecard inputs are collected.
2. 02
   Weigh

   The rulebook sets five stage weights, none above 50%.
3. 03
   Explain

   The model writes the memo.
4. 04
   Publish

   The memo publishes on the Scheduler page with the weights and the next rotation date.
5. 05
   Record

   The weights are written to the on-chain weight registry, which emits a rotation event.
6. 06
   Execute

   From 14:30 UTC on the rotation date, the Reserve's buys follow the new weights.

#Weights are written on-chain

The five stage weights live in the Scheduler weight registry on Robinhood Chain. The owner sets them, and the registry emits an event on each rotation.

Anyone can read the weights from the chain and compare them with the published memo.

#Rotations respect the stage cap

No stage weight exceeds 50%. The Power weight is held in Design, because the stage mapping has no power-stage token.

A rotation changes the target weights, and new purchases follow them. The Reserve does not sell holdings to reach a new target. It sells only to bring a stage back under the 50% cap.

After a rotation, the Reserve's buys follow the new weights. Every buy runs under the buy discipline: slices time-weighted across the trading session, U.S. market hours only, a contract-enforced maximum slice and a Chainlink price bound.

A rotation therefore moves the Reserve toward its weights across trading sessions, not in one transaction. It never buys the weekend premium. The discipline is set out in The Reserve.

#The model picks no stocks

The Scheduler reads constraints. It weights stages, and the stage mapping decides which stock tokens carry each weight.

Memos never claim market-beating performance. Rotations follow the published rulebook and do not aim to outperform any market; see Risk.

Memos follow the same copy rules as this documentation. Every claim in a memo carries a number, a rule or a link.

#Parameter sheet

• ReserveThe ReserveChip equity bought with the tax, grouped by supply-chain stage and priced by Chainlink.
• ReferencePrice feedsHow Chainlink tokenized-equity feeds behave and how each protocol action reads them.
• ReferenceRiskThe risks of holding, staking and redeeming $KYZN, each with its mechanism.
• Get startedGlossaryEvery protocol term, defined once and grouped by the part it belongs to.