Industry intelligence · 12 sectors

Ask a question about an industry. Get an answer where every claim is traced to its source.

Forum Atlas answers in one to two minutes from a corpus of public documents across twelve sectors. Each claim carries a verbatim citation to a named document. Claims that cannot be matched to a source are withheld and shown as withheld.

1,693,979
documents
17,155
sources tracked
100.0%
of citations resolve

corpus and sources counted 2026-10-31 01:47 UTC · citations counted 2026-10-31 02:03 UTC

Forum Atlas / ask · ForumEnergy · ForumTechdeliberate

Question

How do PJM capacity prices affect Virginia data-center buildouts?

answer · 6 citations · 6 verified · 1 withheld

PJM's Base Residual Auction sets the price Virginia load-serving entities pay for generation availability. The 2025/26 auction cleared at $269.92/MW-day RTO-wide, roughly nine times the prior year 1; the 2026/27 auction cleared at the $329.17 collar agreed in the FERC settlement 23.

Dominion Energy Virginia is the largest load zone affected. Dominion reported approximately 40 GW of contracted data-center capacity in its pipeline at year-end 2024 4, and its 2024 Integrated Resource Plan filed with the Virginia SCC projects zone load roughly doubling by 2039 5. Capacity costs are passed through to customers via rate adjustment clauses 6.

WithheldEffect of 2026/27 prices on 2027 construction starts — no verbatim match found in corpus after two validation passes. insufficient_evidence
4 Provenance chain● Verified
claim
≈40 GW contracted data-center capacity, YE2024
source
Dominion Energy, Inc.
document
Form 10-K, fiscal 2024 · SEC EDGAR
chunk
chunk · page · item
verbatim
not shown — in a delivered answer this rung holds the exact substring of the source document that the claim was matched against

open document → · review aidmatch: exact

Challenge (private review aid)Add to caseExport · DOCX PPTX MD

Sources · 6passages
  1. 1PJM Interconnection2025/2026 Base Residual Auction Report2
  2. 2PJM Interconnection2026/2027 Base Residual Auction Report1
  3. 3FERCOrder accepting settlement, Docket EL25-461
  4. 4Dominion EnergyForm 10-K, fiscal 20243
  5. 5Virginia SCCDominion 2024 Integrated Resource Plan2
  6. 6Virginia SCCRate adjustment clause filings1

Fig. 1 · an answered question with its provenance chain · illustrative

01 · Method

How a claim gets made

Retrieved text is synthesised into an answer, then each sentence is checked against the source chunk for a verbatim match. The check fails closed: a claim that cannot be matched is withheld and reported, never softened.

How a claim gets madeSource, ingestion, chunk and embed, retrieval, synthesis, validator, citation. The validator withholds a brief when a claim cannot be matched verbatim to source text. 1,759 briefs passed and 601 were withheld.Source17,155 tracked9,569 producingIngestionfetch · dedupeparse · normaliseChunk · embed1,693,979 docs→ chunks + vectorsRetrievalvector search · reranktop-k per sectorSynthesisdraft answerclaim spans markedValidatorverbatim matchfail-closed · 2 triesCitation100.0% resolve11,251 issued1,759 · 75%601 · 25%Withheldinsufficient_evidenceclaim → source → doc→ chunk → verbatim
Briefs through the validator: 1,759 passed · 601 withheld · 25% failure rate, published on purposecounted 2026-10-31 02:11 UTC

Every stage, in plain terms →

02 · Corpus

What the corpus is

Two kinds of documents. A curated intelligence corpus of filings, reports, notices and transcripts; and bulk reference corpora that are records, not signals, kept for lookup and cross-checking.

17,155

sources tracked

catalogued across 12 sectors

9,569

producing

≥ 1 document ingested

counted 2026-10-31 01:47 UTC

1,693,979 documents100%
Bulk reference · 43 sources, largest firstrecords, not signals
  • OSV.dev

    182,399 docs

  • MITRE CVE

    110,429 docs

  • Exploit Database

    46,700 docs

  • NIST NVD

    42,230 docs

  • SEC EDGAR (submissions+XBRL+FTS+bulk)

    20,595 docs

  • NIST National Vulnerability Database

    20,440 docs

  • VulDB Recent

    13,750 docs

  • crates.io API

    8,145 docs

  • Open Charge Map

    5,453 docs

  • MITRE CVE Program (cvelistV5)

    5,166 docs

  • Artifact Hub Helm

    4,017 docs

  • Hex.pm Elixir Erlang

    3,457 docs

  • Hugging Face Trending Models

    3,162 docs

  • CISA KEV JSON API

    1,765 docs

  • Allego

    1,453 docs

  • PeeringDB API

    1,335 docs

  • CelesTrak GP API

    1,042 docs

  • Epoch AI Models DB

    1,035 docs

  • EIA Hourly Grid Monitor

    871 docs

  • FIRST EPSS

    116 docs

  • EIA Weekly Natural Gas Storage

    26 docs

  • NHTSA vPIC API

    26 docs

  • EIA fuel prices

    15 docs

  • abuse.ch SSL Blacklist (SSLBL)

    13 docs

  • EIA Weekly Petroleum Status Report

    12 docs

  • abuse.ch Feodo Tracker

    6 docs

  • EIA Drilling Productivity Report

    4 docs

  • EIA Monthly Energy Review

    4 docs

  • blocklist.de/lists/all.txt (MISP OSINT)

    1 docs

  • cybercrime-tracker.net - all (MISP OSINT)

    1 docs

  • diamondfox_panels (MISP OSINT)

    1 docs

  • ELLIO: IP Feed (Community version) (MISP OSINT)

    1 docs

  • firehol_level1 (MISP OSINT)

    1 docs

  • http://cybercrime-tracker.net gatelist (MISP OSINT)

    1 docs

  • http://cybercrime-tracker.net hashlist (MISP OSINT)

    1 docs

  • ip-block-list - snort.org (MISP OSINT)

    1 docs

  • IPsum (aggregation of all feeds) - level 1 - lot of false positives (MISP OSINT)

    1 docs

  • List of malicious domains in Poland (MISP OSINT)

    1 docs

  • malshare.com - current all (MISP OSINT)

    1 docs

  • OpenPhish url list (MISP OSINT)

    1 docs

  • SERPRO: Reputation IP Feed (MISP OSINT)

    1 docs

  • Spamhaus DROP list

    1 docs

  • VXvault - URL List (MISP OSINT)

    1 docs

Bulk records (vulnerability registries, filing indexes, package and model catalogues) are never cited as intelligence on their own; they anchor identifiers and dates in claims drawn from the curated corpus.

Fig. 2 · curated vs bulk reference, proportional · counted 2026-10-31 01:47 UTC

03 · Workflow

Discover, monitor, investigate, decide and defend

Four stages of analyst work, each served by a product object. Opportunity and risk are lenses that run through all four, not separate features.

Discover

Ask a question; get a cited answer.

Monitor

Follow sectors and entities; every brief is dated.

Investigate

Follow an entity across sectors.

Decide & defend

Compose the case; export with provenance.

Product object

Ask

quick ~$0.01 · deliberate ~$0.10

Briefs & watchlist

dated · per sector · validated

Entity dossier & graph

typed edges · cross-sector

Cases, exports & forecasts

DOCX PPTX XLSX MD · manifest

Opportunity lens

Find who is positioned where a market is moving.

Track emerging demand signals by sector.

Map counterparties and dependencies.

Put a probability and a source on the upside.

Risk lens

Surface exposures before they are priced in.

Read the withheld count alongside the passed one.

Trace concentrations and single points of failure.

Every number in the memo carries its citation.

04 · Entities

One entity across sectors

An entity is resolved once and linked across all twelve sectors. A relationship carried by an identifier registry — SEC, GLEIF, Wikidata — is drawn on that provenance. A relationship a model extracted from text is drawn only once it has been marked verified and every source document behind it still resolves. Everything else is stored and not drawn, so the number of relationships shown is smaller than the number held.

ego network · depth 18 edges shown
Ego network of one entity across sectorsOne entity at the centre with 8 typed relationships to counterparties in Tech, Materials, Energy, Defense, Health and Comms. 4 edges are verified and drawn solid; the other 4 are candidates and drawn dashed. Illustrative shape, not measured data.TECHPlatform vendorTECHSector rivalMATEComponent makerENERPower supplierDEFEProcuring agencyHEALClinical software unitTECHDeveloper-tools unitCOMMNetwork carrierEntity
RelationCounterpartySectorState
invested inPlatform vendorTechverified
competitor ofSector rivalTechcandidate
customer ofComponent makerMaterialsverified
customer ofPower supplierEnergycandidate
supplier toProcuring agencyDefenseverified
parent ofClinical software unitHealthverified
parent ofDeveloper-tools unitTechcandidate
supplier toNetwork carrierCommscandidate

illustrative shape · no real entity, no measured value · the live graph draws only edges the corpus supports and marks candidates as dashed

Fig. 3 · ego network with typed edges · illustrative

One real entity, its verified facts and the trust gate →

05 · Forecasts

Forecasts you can score

A forecast is scored only if it was bound, at issue, to a named public series, an explicit threshold and a deadline that can settle it. The bound ones are scored with a Brier score when they resolve, and forecasts on the public prediction-market ledger are also sealed as SHA-256 commitments before settlement. Forecasts issued before the binding rule stay on the record and are published as unscoreable rather than dropped. The track record is short and we say so.

Illustrative only: the shape of a posterior distribution over a binary outcome. This is not a forecast on the record, and it is about no company.

sector: cyber · horizon 365d · point estimate 62% · interval 54–69%

Probability distribution for the sample forecast; point estimate 62 percent.0%20%40%60%80%100%pprobability of acquisition within 12 months

Resolution

A forecast is scored only if it was bound, at issue, to a named public series and an explicit threshold. The resolver records what the series shows and publishes the Brier score, win or lose.

illustrative sample distribution · not a forecast on the record

Public record

190

forecasts on record

18

resolved

18 resolved — too few to read a calibration curve. The scoreboard is public from day one rather than hidden until it looks good.

queried 2026-10-31 02:11 UTC

Forecasts on the public prediction-market ledger are sealed as a salted SHA-256 commitment before the market settles and revealed afterwards, so a changed number would produce a different hash. Forum Forecast outputs in the product are timestamped rows, not hashed commitments.

The public scoreboard →·The commit–reveal ledger →

06 · Coverage

Twelve sectors, one graph

Each sector has its own sources, brief and analysts' vocabulary. Entities and claims are shared, so a question in one sector can draw on documents from another. Before launch the brief engine runs every Monday at 06:00 UTC (every Monday, Wednesday and Friday at 06:00 UTC after launch); a brief whose claims fail validation is withheld rather than published, so each desk shows the date of its own latest validated brief.

07 · Cases

Work that leaves as a file

A case is the container you build an argument in. Add a brief, an entity, a forecast or a note to it from the brief, dossier and forecast pages, and record why each one is there. When the case is ready it composes into four files, each carrying a provenance manifest so the reader can check any line back to its source.

Marking a case decided freezes its manifest as a dated record of what the decision was made on. Reopening the case does not delete it.

What goes into a case

  • Briefs
  • Entities
  • Forecasts
  • Notes
One casethesis · items · a note against each item · a frozen manifest at decision
  • Memo.docx

    The case's thesis, its claims in order, and the note recorded against each one.

  • Board pack.pptx

    One slide per finding, each carrying the citations behind it.

  • Binder.docx

    The memo plus a verified-fact table for every entity in the case.

  • Comps.xlsx

    The case's entities as a comparison sheet, one row each, values with their source.

Composed from cited claims11,251 citations · 100.0% resolve · counted 2026-10-31 02:03 UTC

every file carries forum.case_manifest.v1briefs also export .md and print-ready .html

Fig. 4 · one case, four files, one base of cited claims · counted 2026-10-31 02:03 UTC

08 · Readers

Who uses it

Boutique consultant

Asks the client's question on Monday, checks each citation, and drops the answer into a Deep Dive with the manifest attached.

Ask · Deep Dive · manifest

Trade journalist

Reads each new sector brief, opens a claim's provenance chain, and quotes the verbatim source rather than the summary.

Brief · provenance · verbatim

Government-affairs staff

Follows dockets and filings in two sectors, receives withheld counts alongside passed ones, and exports a risk memo for the VP.

Monitor · Risk memo

Developer

Calls the API and receives claims with citation objects and a validator result; builds an agent that refuses to ship an uncited number.

API · claims[] · validator{}

09 · Trust

The numbers we publish about ourselves

11,251

citations issued

100.0% resolve to a live document

queried 2026-10-31 02:03 UTC

1,759 / 601

briefs passed / withheld

25% failure rate, published on purpose

queried 2026-10-31 02:11 UTC

190 / 18

forecasts on record / resolved

commit–reveal ledger, public

queried 2026-10-31 02:11 UTC

9,569

sources producing

of 17,155 tracked

queried 2026-10-31 01:47 UTC

POST /v1/query → { sections[], sources[], validator{}, usage{} }API reference →

Check the numbers without us

The figures on this page are served as JSON from the same rollups the page reads, each field carrying the instant it was counted and a degraded list naming any read that did not return. Nothing on this site is a fixture that these endpoints would contradict.

10 · Changes

What changed, and when

Drawn from the ledger the platform is built against, with plain-English entries written over it where a ledger record would not read as anything — releases, data corrections and decisions alike. A correction is published the same way a feature is.

All 54 entries →

  1. 2026-10-20

    Decision

    Lead decision under FD-40: ingestion fetchers run their launch cadence pre-launch (scrape 2x/day, api daily, rotation 4x/day, EDGAR daily)

  2. 2026-10-15

    Release

    SP-27B: /app/entities shows Watched with this week's documents; /app/watchlist is a page (301 to Today removed at merge)

  3. 2026-10-14

    Release

    SP-12C: docs-freshness check judged on first-parent history (REVIEW-SP-12C F1); ci python job at fetch-depth 0

  4. 2026-10-14

    Release

    SP-13C: consent banner, one subprocessor list, DSAR intake handler, workspace deletion routine, export AI marker, Terms v1.1 draft

  5. 2026-10-14

    Release

    SP-17C: true .docx Memo/Binder, behavioural money/export tests, cited-paragraph validator rule, digest + entity-resolution fixes

  6. 2026-10-14

    Release

    SP-21C: product objects code half - Answer/Watch/Case/Export (D1-D4 SQL in hand-off, signed manifests, /app/today, suggest-only evidence)

ledger as of 2026-10-30

11 · Pricing

Self-serve, from free

Every tier includes citations and the provenance chain. API queries are metered at about $0.01 quick and $0.10 deliberate.

Free

$0

Top-3 items per brief, 30-day archive, 50 queries/mo.

Start free

Pro

$39/ month

Full Forum Brief, unlimited speculative queries, full archive, Network basic.

Start free

Team

$199/ month

5 seats, Forum Forecast, full Network, exports.

Start free

Business

$600/ month

15 seats, SSO, uncapped API reads, DPA + 1-year audit log.

Talk to us

Enterprise

from $1,500/ month

25+ seats, dedicated CSM, custom SLAs.

Contact sales

Full pricing and limits →

The free tier includes cited answers, sector briefs and the provenance chain.