Institutional data room

Stress-test the instrument without us in the room.

What ships in production today — sealed historical proof, live credibility, and a machine-verifiable attestation — indexed for CIO, quant, and counsel.

Production density

What ships on the sealed engine today.

Engine

V5-locked (1.19.0)

Crisis: 80–90 on the 0–100 NII scale; Collapse ≥90

Sources

30

30 live pipelines · 29 configured on the spine

Coverage

1,000

Active constructs of 1,856 registered — live data on the sealed engine

Observations

1,137,922

Point-in-time readings scored under sealed physics in production

API endpoints

417

Production HTTP handlers shipping on one deploy

Live credibility gate

Monthly trail. Pass or warn in public.

Live monthly layer meets internal Wave-1 gate — safe for allocator diligence alongside sealed backtests.

Live gate · Open

Oct 2026

39.2/100

NII · 0–100 NII scale

Sep 2026

43.3/100

NII · 0–100 NII scale

Aug 2026

42.1/100

NII · 0–100 NII scale

Jul 2026

37.4/100

NII · 0–100 NII scale

  • Single canonical monthly pathpass

    Published NII is computeRII / V5-locked monthly_scores. Legacy dual-run engine_delta retired 2026-07-12.

  • No unexplained flat quarterpass

    No three-month identical NII plateau in recent trail.

  • Clean monthly cyclespass

    7 month-over-month moves since last flat streak (need 2 for Wave-1 live gate).

  • Published trail depthpass

    8 months in trail (latest Oct 2026).

Full live credibility gate

Sealed proof

Clocks on known breaks.

Deterministic replay under frozen physics. Watch = earliest Crisis. Act = dual Crisis. Team diligence PDF per episode.

Process as product

Validation is a product surface, not a back-office secret.

Sophisticated buyers stress-test process before they stress-test thermodynamics. This is how we score, freeze, attest, and bound claims.

  • 01

    Sealed V5-locked physics

    Buyer value · Same engine object in backtest and live watch — no retro-fit after the event.

    Frozen amp / κ / C under V5-locked (1.19.0). Watch and Act clocks are product objects, not post-hoc labels.

  • 02

    Point-in-time source discipline

    Buyer value · No future information in the month being scored.

    Multi-source, deterministic, source-traceable · vintage-honest ledgers per episode · lag policy documented per construct.

  • 03

    Triple-blind domain extraction

    Buyer value · LLM-assisted domains reconciled before the engine sees a single score.

    3 blind draws per period when models disagree; early exit when range < 0.5 on 0–5 scale. Opt-out via INDEX_TRIPLE_BLIND=0 for audits.

  • 04

    Immutable prediction ledger

    Buyer value · Dated divergence claims are facts, not screenshots.

    Published rows under immutability policy · daily chain anchor · Pulse calls timestamp third-party attestation.

  • 05

    Watch vs Act dual clocks

    Buyer value · Early signal and capital posture are different product clocks on purpose.

    Watch = earliest Crisis of any fashion. Act = dual Crisis (I_sys Crisis ∧ gap threshold). Dotcom explicitly discloses Watch ≠ Act.

  • 06

    Claim ceiling enforcement

    Buyer value · Public language cannot outrun Travis-signed packaging doctrine.

    Sealed case studies, mandate chips, and GTM copy pull from single canonical sources — no invented live clocks.

Triple-blind reconciliation ships in production extracts. Ledger policy documented in research records.

Machine probe

Same room. One curl.

Unauthenticated attestation of the same production density and live gate on this page — claim ceiling plus a SHA-256 pin. No key. No recipe. Entity scores stay on the keyed commercial API.

Schema · data-room.v1

GET /api/public/data-room
curl -sS "https://www.entropyindex.com/api/public/data-room"

Boundary

Validation ships. Recipe stays protected.

What this room discloses for IC review — and what stays under contract with the commercial API.

Disclosed hereUnder contract
Validation process, PIT policy, sealed backtest outcomes, live gate checksFull NII formula, coefficient tuning, rubric weights, scoring prompts
Production density — sources, coverage, observations, platform scale — plus evidence tiersConstruct inventory, observation rows, ingest recipes, panel fusion, proprietary normalizations
Published NII/FAI, Watch/Act clocks, sealed case PDFs, monthly trailRaw domain scores without contract, internal diagnostic runs
Divergence qualification rules, ledger immutability policy, blank slot honestyUnpublished forward calls until ledger row exists

Claim ceiling

What we refuse to sell.

  • We do not claim live monthly scores retroactively prove pre-2008 positioning buckets.

  • We do not claim tick or real-time CTI — Seal is monthly; Acute and Delivery are named clocks on every paid receipt.

  • We do not treat Watch (gap-alone) as Act / allocate.

  • We do not sell prediction — structured early-warning and decision-support only.

  • We do not hide validation process — it ships as product documentation.

  • We do not publish methodology recipe — validation yes, trade secret no.

Straight answers

Is this a prediction engine?
No. Structured early-warning and decision-support — Watch surfaces gap, Act requires dual Crisis. We sell measurement discipline, not forecast theater.
Does AI write the score?
One sealed formula on the live Index and every backtest — formulaic result, not narrative. Scoring is deterministic under sealed physics. Zero AI touching the engine: AI may structure evidence and read the face; it does not write the number.
Does live monthly data retroactively prove the GFC backtest?
No. Sealed episodes use frozen V5-locked physics and point-in-time sources. Live monthly NII is a separate production layer with its own credibility gate.
When is Watch enough vs Act?
Watch = earliest Crisis of any fashion (brief the room). Act = dual Crisis (I_sys Crisis ∧ gap threshold). Dotcom explicitly documents Watch ≠ Act.
What happens if monthly NII flat-lines?
Three identical consecutive prints trigger review — analytical note or RCA required before allocator diligence. Visible on the live credibility gate.

Fort Knox

Security posture — machine-audited, publicly verifiable.

Every deploy runs a Fort Knox audit that fails closed. Private keys never touch the production server. The audit result is a public endpoint — verify it yourself.

Loading audit…

Security doctrine

  • Private keys never on production server

    Payer, minter, anchor, and ownership keys are local-only. Fort Knox machine audit fails closed if any appear on Vercel.

  • Split key semantics

    Edge proxy, monitoring, and attestation signing each use a separate secret. Compromise of one does not grant payment bypass.

  • Cash-settlement integrity

    Only on-chain tx_hash with settlement_status=success counts as revenue. Empty-tx facilitator calls are rejected.

  • SSRF-hardened webhook registration

    Webhook subscribers must prove wallet ownership via a recent cash-confirmed settle. URLs are validated against IP, metadata, and RFC1918 patterns.

  • Durable rate limiting

    Console login and public form routes use Supabase-backed rate limits that survive Vercel isolate rotation.

  • Row-level security on all operator tables

    Supabase anon and authenticated roles are revoked from every operator table. Service role only.

  • Timing-safe auth on every internal header

    All secret comparisons use constant-time equality. Session tokens are time-bucketed with HMAC-SHA256.

  • No trade-secret leakage to client bundle

    TVG cutoffs, engine thresholds, and internal key names never ship to the browser.

Live machine audit

Loading…

Verify it yourself

No login. Pass/fail per check — no secrets exposed.

curl -s https://www.entropyindex.com/api/v1/x402/fort-knox \ | python3 -m json.tool

Next step

Review the room. Then scope the desk.

Desk Pilot for bounded side-by-side evaluation — firm license when the IC packet clears.