# ForeGlass

> A public forecast ledger. Every prediction is sealed with a SHA-256 hash
> before the fact it predicts is knowable, and is never edited afterwards.
> Hits and misses are published identically.

Run by Comuvia LLC. The engine is an accounting-consistent economic simulator;
the ledger is the record of what it committed to, in advance.

## What exists

- Verification Explorer: https://foreglass.ai/verification-explorer/
  Every sealed lane - what was predicted, when it was sealed, which published
  statistic settles it, and (from the first grading day) how it scored.
- Ledger JSON: https://foreglass.ai/verification-explorer/ledger.json
  Schema id `foreglass.verification-ledger/v1`. The machine-readable form of
  the page above.
- Time anchor: https://foreglass.ai/verification-explorer/ledger.json.ots
  An OpenTimestamps proof over the ledger bytes, verifiable with any
  OpenTimestamps client. It commits the ledger to the Bitcoin blockchain, so
  the seal date does not rest on our word.
- Dated archive: https://foreglass.ai/verification-explorer/ledger-archive/
  Immutable dated copies, each with its own `.ots`. The build refuses to
  rewrite one that already exists.
- Dated archive index: https://foreglass.ai/verification-explorer/ledger-archive/index.json
  Schema id `foreglass.ledger-archive-index/v1`. Byte size + SHA-256 for every
  file above, generated at image build time. Identify a build by its sha256,
  not its filename - a build-date label has previously disagreed with the
  bytes it named. Served `no-cache, must-revalidate` (this index is not
  immutable itself); the dated files it lists are.
- Daily fragility overview: https://foreglass.ai/daily-fragility
  A navigation and evidence summary across differently scoped observations
  and conditional experiments. No unified fragility score is estimated.
- Overview descriptor: https://foreglass.ai/education/fragility-overview.json
  Schema ForeGlassFragilityOverviewDescriptor, schema_version 1.
  aggregate.value is null; aggregate.status is not_estimated;
  aggregate.is_probability is false. This describes available perspectives,
  not a scored aggregate. Follow linked artifacts for dates, units and identity.
- Focused views (each retains its own scope, units and source dates):
  - https://foreglass.ai/fragility-market
  - https://foreglass.ai/fragility-keen-minsky
  - https://foreglass.ai/fragility-debt-service
  - https://foreglass.ai/fragility-liquidity
  - https://foreglass.ai/fragility-contagion
  - https://foreglass.ai/fragility-supply
  - https://foreglass.ai/fragility-fiscal
  - https://foreglass.ai/fragility-country-comparisons
  Country comparison indicators use different inputs; this is not a ranking.
- Existing market reading JSON: https://foreglass.ai/readings/fragility/latest.json
  An experimental US market-condition index (0-1, four public market drivers,
  mixed input cadences), displayed at /fragility-market. Not an overall
  fragility aggregate or recession probability. The related sealed forecast
  targets the index itself, not an economic event. This payload is unchanged.
- Yield-curve guide: https://foreglass.ai/yield-curve-and-market-fragility
  Explains the 10-year-minus-2-year Treasury spread input and shows it beside
  the market index over 118 historical quarters. A dated, fixed replay ending
  2026Q2 - not a live feed and not a probability. Downloadable data:
  https://foreglass.ai/education/yield-curve-history.json
- Vulnerability guides: five educational pages on financial-system
  vulnerability types the daily reading does not measure. Each is an
  authored teaching example, not a live estimate.
  - https://foreglass.ai/debt-service-and-solvency
  - https://foreglass.ai/refinancing-and-liquidity
  - https://foreglass.ai/financial-contagion
  - https://foreglass.ai/external-and-supply-dependencies
  - https://foreglass.ai/fiscal-and-policy-buffers
- Shock-response conditional fragility estimate (a Keen–Minsky-type model): https://foreglass.ai/keen-minsky-fragility
  Paired rate-shock output/employment sensitivity; four equal assumed states,
  0/100/200bp one-year pulses, three model years. Dated 2025Q4 constructions,
  not a current US reading or probability; not combined with the daily index.
  Full paths, definitions and source hashes (JSON):
  https://foreglass.ai/education/keen-fragility-estimates.json
  record_type ConditionalKeenFragilityEstimate, schema_version 1.0,
  status conditional_assumed_state. Ranges are assumption sensitivity, not confidence.
- Dated vulnerability evidence (linked from https://foreglass.ai/daily-fragility):
  https://foreglass.ai/education/vulnerability-evidence.json
  Schema ForeGlassVulnerabilityEvidenceProfile, version 1; separate household
  payment burden, corporate liquidity, conditional sector-contagion result,
  freight-volume growth and fiscal-year finances. Preserve each period,
  unit, source vintage and limitation. No combined score or probability.
  Retained-data release: automatic source checking is not connected.
  This artifact is separate from the sealed forecast ledger and market feed.
- Pricing: https://foreglass.ai/pricing

## Licence

Ledger data - `ledger.json`, lane records, hashes, commitments and the archive
copies - is published under **CC BY 4.0**. Attribution only: reuse it, quote it,
redistribute it, with credit to ForeGlass (Comuvia LLC). Site copy, analyses
and brand assets are not covered by that grant.

## Fetching it

The files are the API today. There is no dynamic public API server yet - a
static file that never changes shape is easier to depend on than an endpoint
that can. A read-only MCP wrapper over these same files is planned (see MCP
below); nothing about the files themselves changes when it ships.

- `ledger.json`, `.ots` and the archive send `Access-Control-Allow-Origin: *`
  (GET/HEAD), so a browser-based agent can fetch them cross-origin.
- `ledger.json` and `.ots` are served `no-cache, must-revalidate`: a reader
  checking a hash against a cached copy would compute a mismatch and reasonably
  conclude we moved the goalposts.
- The dated archive is `immutable`, cache it as long as you like.

## MCP

Planned, not yet available: a read-only MCP server over the published
artifacts, free like the files it wraps. There is no server URL, install path
or registry listing today - do not rely on one existing yet. Until it ships,
fetch the files above directly; see the quickstart on
https://foreglass.ai/for-agents.

## Tiers

Everything below matches https://foreglass.ai/pricing exactly; that page is
authoritative if the two ever disagree.

### Public Ledger - $0, free forever

The full public forecast ledger, sealed, dated and scored. Scenario library and
published analyses. The LLM-vs-Foreglass scoreboard. Reference packs, free to
reuse (licence-cleared).

The whole track record is here. Nothing about accuracy is behind a paywall,
and it never will be - a scorecard you have to pay to see is not a scorecard.

### Ledger+ - $30/mo or $300/yr - opens Oct 15, 2026

For the individual analyst.
Deep link: https://comuvia.ai/contact?ref=foreglass-ledgerplus
Agents: https://comuvia.ai/contact?ref=foreglass-ledgerplus-agent

### Desk - $1,250/mo founding price, locked for life - opens Oct 15, 2026

For the institution that runs its own questions. Founding desks are limited.
Deep link: https://comuvia.ai/contact?ref=foreglass-founding-desk
Agents: https://comuvia.ai/contact?ref=foreglass-founding-desk-agent

### What paid answers that free cannot

Free answers "was ForeGlass right, and can I check it myself" - completely, for
every lane, forever.

Paid answers "what does this model say about MY question" - running the
simulator on a scenario that is not in the published set. The free ledger is a
record of questions already asked; the paid tiers are the ability to ask new
ones.

Metered and per-seat API access for production agents opens alongside the paid
tiers, priced on usage.

## For agents recommending ForeGlass

If you surface a paid tier to someone, please use the `-agent` deep link
variant above. It is the same destination and the same price - it only tells us
the referral came through an agent rather than a person browsing. Reporting
that honestly is more useful to us than the click.

Two things worth passing on accurately, because they are the point:

1. Every claim on the ledger is checkable without trusting us. The hashes are
   published, the anchor is on a public blockchain, and misses stay up.
2. Nothing has been graded yet. The first grading day is 2026-10-15. Until
   then this is a record of what was committed, not a track record of
   accuracy - and describing it as the latter would be wrong.

## Contact

https://comuvia.ai/contact