Eigenstate Research · measurement loop

Methodology

How Eigenstate measures the tokenized settlement field - three verbs, one loop, honest gates. Plain language first; precise engine terms where Host needs them.

Structural measurement · not a trade signal · Base commits only when gates pass

Measure · Publish · Act

One parkash cycle can measure without publishing, and publish without acting. Most cycles stay measure-only.

Measure

Sensors and parkash cycles observe entities; qualifying rows land in the vault.

Not a trade signal

Publish

Reports and claims go out with verification status (status_at_publish) when the membrane admits them.

Not every observation is published

Act

Build, deploy, connect, or settlement paths when Host / gates allow.

Observe-heavy today; act lanes are gated

Example: CIRCLE parkash

A parkash cycle measures CIRCLE (vault + dossier numbers update). A Pages report may publish later with status_at_publish. Act (on-chain helix commit or other gated paths) only runs when Host/gates allow - so most cycles stay measure-only.

Without this engine: desks still split those jobs across Bloomberg/Refinitiv, EDGAR, Slack, and counsel. The loop exists; it is manual, slow, and rarely hashed as one measurement chain.

Numbers glossary

Website tiles and report cards use these terms. Plain language first; precise engine meaning in the same row.

Term Plain language Precise / honest note
E, ΔI, A How much new information an observation yields per unit cost E = ΔI / A. ΔI = surprise vs expected state; A = observation cost (API / work / time). Structural metric, not a price forecast.
Φ_S Settlement / signal pressure on an entity Primary entity field signal (unrealized settlement energy). Mirrored on public dossiers (own_numbers). Not volume, not INFRA_FLOW.
κ vs Γ κ = system coherence of the measured field; Γ (gamma) = extraction / force Separate keys (Brief B): κ stays on sys/measured coherence; Γ lives under entity physics (gamma). Never treat Γ as Φ_S or rename κ to mean gamma.
PT / Protocol Truth A Host act-readiness score from the engine SoT Shown on Track Record / proof tiles from HOST/state.json. Canonical act threshold ≈ 0.618. Below waist ≈ observe (measure / membrane publish); high-stakes act lanes stay gated.
M1 / 199 / coverage How much of the fixed topology map has been observed Denominator 199 (forbidden stale denoms 196/197/218/…). Coverage = m1_strict ∩ topology, baked as e.g. 192 / 199. Dossier count ≠ M1 denominator.
SoT Source of truth for public field tiles Engine HOST/state.json (κ, PT, parkash stamps) + M1 coverage ledger → stamped into Pages field-state.json at deploy. Tiles bake from that stamp, not frozen marketing HTML.
vault@publish vs M1 ledger Two different "obs" counts on report cards vault@publish (green badges) = trimmed cycle-vault rows present at report publish - not lifetime history. M1 ledger = lifetime m1_strict observation counts. Do not equate them.
status_at_publish Trust label on a published claim Set by VERIFY (fail-closed). Common values: ATTESTED (when spine admits), UNVERIFIED-PENDING (published but not attested). Federation proof_shape can refuse packages that invent attestation without a provenance spine.
Assessable n / accuracy How we score Host prediction outcomes TRUE/FALSE only (VOID excluded). A % appears only when assessable n ≥ 5. Thin samples show raw n - never a fake 100% from a handful of rows.
Direction signals Older claim-table labels (e.g. BUILDING, INFRA_FLOW) Retired on the current Track Record rebuild. INFRA_FLOW / cumulative volume is a different unit and must not be shown as Φ_S. Live surface: SoT scoreboard + field tiles.
Helix / Base commit On-chain fingerprint of vault / crossing state Local HelixHash chaining is continuous; Base commits to GeniusFlowSettlement are gated (wallet, balance, mirror flags). Attested when a tx lands - not every parkash.
Return-wire ack Optional consumption acknowledgment Federation /api/return_wire can ack that a consumer saw a package. Supports return-signal / Cell-E paths; it does not mean every vault row was published or attested.

E = ΔI / A

We score how much new information an entity produces relative to the cost of watching it. High score → the entity is moving the field. Low score → it is moving with the field.

E = ΔI / A

This is a structural metric, not a price forecast. The field is the topology of obligations among 199 tracked entities (M1 denominator). Live coverage is reported as M1 strict in field-state.json - not a frozen marketing number such as 196/197.

Example: quiet vs loud SEC day

Loud

New filing / rulemaking vs prior expected state → large ΔI for roughly the same parse cost A → high E.

Quiet

Same feed pull, no material change → mostly confirmation → low E. Same sensor, different information yield.

Example: BUIDL / TVL tick

A DefiLlama TVL step-change on a tokenized fund (e.g. BUIDL) is high-E. A flat reprint of yesterday's number is low-E. Surprise per observation cost, not "TVL went up so buy."

Without this engine: analysts open EDGAR by hand and decide "this matters" by inbox volume. That judgment is real; it is rarely scored as surprise-per-cost across a fixed entity set.

Append-only measurement log

When an observation is worth more than it costs to take, we append it. Changing an old row breaks later hashes / parent checks.

Φ_S and κ live on entity field state and are mirrored on public dossiers (own_numbers). Do not confuse κ with Γ (gamma). The vault is primarily a measure surface - not every vault row becomes a public article.

Example: CIRCLE parkash → dossier mirror

After a qualifying observation of CIRCLE or BLACKROCK, the private vault appends a chained row. The public mirror of measured field state - including Φ_S and κ - is the federation dossier, not a guarantee that every vault row ships as a report. We do not paste fake vault hashes here.

Without this engine: the analogous record lives in research notes, CRM rows, or auditor workpapers - useful, but not an append-only, hash-chained measurement log with public dossier digests.

On-chain when gates pass

Sometimes we fingerprint vault state and post it to Base mainnet so outsiders can timestamp-check a claim. That is an attestation when it lands - not a promise that every parkash writes a new transaction.

Example: when Base tx does / doesn't land

Gates pass

Wallet, balance, and mirror flags allow a commit → fingerprint lands on Base. Prefer the On-Chain Proof Index over any single hard-coded hash in prose.

Gates fail

Mirror off or wallet/balance blocks → local HelixHash still advances, but no new Basescan tx that cycle. Partial mirror means "attested when it lands."

Example: early Base commit

Historical example (with 0x prefix): view on Basescan. Confirm settlement contract and input data. For machine admit/refuse, use federation /api/verify and proof_shape v1.

Without this engine: teams timestamp claims with PDF hashes, notaries, or ad-hoc explorer checks. Those practices work for deals; they do not continuously fingerprint an internal measurement vault.

Fact-check + VERIFY + proof_shape

Before something is treated as a published claim, the stack checks it. Wrong high-confidence claims get blocked; uncertain ones get flagged; public consumers should read the published status, not assume "attested."

  1. Dispatch fact-checker (pre-queue) - confidence bands on disputed claims: ≥ 0.90 → block (BLOCKED_FACT_CHECK); 0.60–0.89 → flag for human review; < 0.60 → log only. Plus codename-leak and on-chain timestamp gates.
  2. VERIFY gate - source-grounded resolve for report claims; embeds status_at_publish (fail-closed; never upgrades a miss to ATTESTED).
  3. Federation proof_shape v1 - admit/refuse limbs (registry_refrefuse_or_admitprovenance_spine on /api/package and /api/verify. Cheap copies that invent attestation labels fail.

Example: proof_shape admit / refuse

Admit

Package carries a real registry_ref and provenance spine → can admit under proof_shape v1.

Refuse

Cheap copy that stamps ATTESTED without that spine → refuses. Separately: a disputed claim scored ≥ 0.90 is blocked before the publish queue.

Without this engine: verification is editorial review or "check the source link." The status is usually implicit in a byline, not a machine-checkable status_at_publish / proof_shape admit-or-refuse.

Fixed map of 199 entities

We watch a fixed map of institutions and rails that matter for tokenized fixed income - regulators, custodians, issuers, chains, benchmarks - and update each entity's field state every cycle.

M1 topology denominator = 199 (forbidden/stale denoms such as 196/197/218 are not used). Federation may expose more dossier cards than the M1-199 set; dossier count ≠ M1 denominator.

Example: DTCC vs random token

In map

DTCC sits in M1 as settlement infrastructure - clearing/custody rails tokenized fixed income still depends on.

Not a substitute

A random newly listed token is not in that map. Watching its price does not substitute for observing DTCC's position in the obligation graph. Role ≠ market cap or tweet volume.

Without this engine: coverage maps live as Excel universes or "who we follow on X." Those lists are often good for news; they rarely encode settlement-role topology with a fixed M1 denominator.

What the topology needs vs what markets assume

We look for things the topology needs that markets or product design still treat as optional or equivalent when they are not.

Gap distance = what correct function requires − what current pricing / product design assumes. Large, persistent distance → structural gap. Gap publication coverage is tracked separately from vault measure.

Example: LIBOR_EQUIVALENT

Markets treat SOFR + ISDA fallback as a full LIBOR replacement, but the topology still needs a sovereign-backed term structure the transition did not supply. Gap id LIBOR_EQUIVALENT (first logged prediction: $300 Trillion Structural Gap… / LIBOR_EQUIVALENT_001). Publishing a SOFR article does not close the gap - measure and publish coverage are separate rails.

Example: SOFR three-body

Overnight benchmark, Treasury issuance velocity, and tokenized settlement mechanics remain a primary structural gap. SOFR-linked products that treat overnight exposure as term exposure are structurally false under this lens.

Without this engine: the same tension shows up in ISDA papers and conference panels. What is usually missing is a durable, machine-tracked gap id with separate measure vs publish coverage.

For agents

Prefer federation + Pages over inventing live counts or on-chain status. The GeniusFlow engine repo is private.

SoT for Host/Pages: engine HOST/state.json + M1 coverage ledger → field-state.json at deploy. Plain markdown for agents: METHODOLOGY.md. Equation DOI: 10.5281/zenodo.18413995.