Trev. Research
Independent Research • Economics & Market Structure

A modest overlay. A clearer read.

Project 2 tested whether public carry and market-plumbing signals could sharpen interpretation of an existing Treasury conditions model. The answer was limited but useful: one carry overlay survived, a trimmed context layer survived, and stronger claims did not.

Read what held up See the claim ceiling Technical appendix
Why this project existed

Start with the question the baseline left open.

The earlier framework already separated instability from constraint. Project 2 existed to test whether public carry and plumbing data could make those readings easier to interpret without rewriting the model itself.

The earlier research established a two-axis framework for Treasury market conditions, built around instability and constraint. But it left open several harder questions: whether carry conditions were telling us anything useful, whether the model’s constraint readings reflected real funding strain, and whether public validation signals could make those readings easier to interpret.


Project 2 was built to test those questions without rewriting the original model. In simpler terms, this project asked a practical question: when Treasury market conditions look strained, can public carry and plumbing data help us understand what kind of strain we are looking at?


That is a narrower question than “did dealers lose money?” or “did we find a new stress regime?” The project did not set out to prove either of those, and it did not end up proving them.

What we tested

One main carry candidate, one internal comparator, and a trimmed context layer.

Project 2 tested one main carry candidate, one internal comparator, and a trimmed set of context signals.

Primary carry overlay

2y–BGCR

This became the project’s one surviving carry-style extension. It was retained only as a secondary interpretive overlay, not as a new core input.

Internal comparator

2y–SOFR

This remained useful only as an internal sensitivity check because it overlapped too heavily with the existing funding family.

Context layer

Plumbing and settlement context

Repo volumes, SOMA lending, and surfaced fails were kept as descriptive context for funding, collateral, plumbing, and settlement conditions.

Narrow later additions

Dealer positions and Treasury fails

These were surfaced later as additional context, but they were not strong enough to reopen the project’s limits.

What held up

The project kept a modest overlay and a better descriptive read.

The strongest survivor was 2y–BGCR. It did not become a new stress classifier, but it did show some modest value as a secondary overlay, especially in quieter funding and plumbing episodes where the baseline instability story was not doing all the explanatory work on its own.

Surviving overlay

2y–BGCR stayed, but only in a restrained role.

September 2019 and October 2025 were the clearest examples. In those windows, the overlay added interpretive value without becoming a lead signal, a universal stress meter, or a replacement for the baseline model.

Surviving context

The context layer remained descriptive.

Repo volume families, OFR repo segment volumes, SOMA lending, and surfaced fails help describe conditions as non-calm, funding-tight, collateral-active, or settlement-strained. Their value is texture, not classification.

Project 2 did not produce a new engine. It produced a better dashboard light.
What did not hold up

The demotions are part of the result.

The page is more credible if the failures stay visible. Project 2 did not find a clean leading indicator, did not preserve 2y–SOFR as a public-facing signal, and did not turn context into a classifier.

No clean lead

Timing stayed mixed.

The surviving carry overlay did not behave like a disciplined early-warning signal. In some important windows it added useful context; in others it did not.

No independent confirmation

2y–SOFR collapsed into overlap.

It was almost identical to 2y–BGCR and mainly reflected overlap with the same funding family, which left it useful for internal checking but not for public emphasis.

No context-based classifier

Plumbing data did not solve identification.

The context layer helped describe market plumbing, but it did not separate healthy strain from impaired stress on its own and did not reopen any path toward integration.

Key episodes

Four windows that show what the overlay could and could not do.

Episode 1

September 2019

This was one of the clearest cases where the surviving overlay helped. Carry conditions were weak while instability remained relatively quiet, which made the episode look more like a funding and plumbing disturbance than a pure volatility event.

Episode 2

March 2020

This was the warning case against overreading carry. The overlay deteriorated early, but then flipped while instability kept surging. That blocked any simple story like “negative carry means damage” or “positive carry means things are fine.”

Episode 3

March 2023

Here the surviving context signals helped describe the market environment, but they still did not turn into a clean lead or a new classification rule. The episode was informative, not decisive.

Episode 4

October 2025

This was another useful overlay case. The carry and context complex suggested a non-calm, plumbing-active backdrop even though the instability axis was not loudly signaling a classic dislocation.

Why this still matters

A narrow result can still improve how the broader system is read.

Treasury markets matter far beyond trading desks. They influence borrowing costs, financial stability, and how smoothly the broader system absorbs stress. A model that can better distinguish simple volatility from funding and plumbing strain is useful, even when the improvement is modest.

Plain language

Not every turbulent Treasury episode is the same kind of problem. Some are mostly about prices moving fast. Others involve the funding and settlement machinery underneath the market. This project helped make that distinction easier to read, but not easy enough to declare solved.

Project 2’s value is that it narrowed the story honestly. It tested a stronger carry idea, rejected the parts that did not survive, kept the parts that added real interpretive value, and clarified what stronger evidence would be needed before anyone could make bigger claims. That makes the broader research program more disciplined, not less.

What this project does not prove

The claim ceiling stays visible.

  • It does not prove dealer monetization.
  • It does not prove dealer losses from negative carry.
  • It does not solve Harvest versus Stress.
  • It does not create a new constraint axis.
  • It does not justify integration into the classifier.

That limitation is not a flaw in the page. It is one of the most important findings the project produced.

How the project narrowed

The process got smaller, then stronger.

As the work progressed, the project became more selective, not broader. The final package locked in a baseline-unchanged result: 2y–BGCR stayed as a modest overlay, 2y–SOFR was demoted to QC, the surviving context layer stayed descriptive, and the narrow validator expansion slightly strengthened context without materially changing the final judgment.

Step 01
Question Can carry and plumbing data sharpen interpretation without rewriting the model?
Step 02
Screening Early candidates were reviewed against the claim ceiling and overlap risk.
Step 03
Gate review Integration paths and overclaim routes were blocked before they hardened.
Step 04
Red-team The project was forced into a narrower, more defensible contribution.
Step 05
Narrow validator check Late additions slightly improved context, but did not change the governing judgment.
Step 06
Final lock What remained was smaller than the starting ambition, but more trustworthy.

Each stage reduced overclaim risk. The end result was smaller than the original ambition, but more defensible.

What remains open

The research path continues, but the ceiling does not move.

The project leaves a few narrow future branches open. These include out-of-sample tracking of the retained overlay and context set, archival Treasury fails tracking, and a cleaner public dealer-capacity branch. But all of those remain conditional. None of them reopen integration on the current record.

In other words: the research path continues, but the claim ceiling stays where the evidence put it.
Appendix

Technical support material and source package.

For readers who want the technical version, the underlying research documents contain the formal governance locks, screening results, red-team adjudication, feasibility review, narrow validator build, and final package memo.

Appendix details

Source package referenced for this page

This public page is written as a static explainer, not as a stage log. The supporting document set below governs the claim ceiling, wording discipline, and role assignments behind the page.

  1. Phase 1 Amended Governance Memo
  2. Phase 1 Executive Judgment
  3. Project 2 Canonical Research Brief
  4. Stage 3 Executive Lock Decision
  5. Stage 5 Screening
  6. Stage 6 Integration Eligibility Adjudication Gate
  7. Stage 7 Red-Team Adjudication
  8. Stage 8A Feasibility Judgment Memo
  9. Stage 8B Narrow Validator-Expansion Build
  10. Stage 9 Final Package Memo
  11. Project 2 Website Copy Deck and Engineering Blueprint
Project 2 public page concept for Trev. Research. Replace placeholder links in the header and hero pills with your live research index, framework page, or monitor routes during integration.