Challenge the headline
Audit Hesper Atlas’s track record. Include losing periods, worst signals, the still-open replay book, and every caveat that changes how the headline should be read.
Hesper Atlas exposes its historical replay, static-tail and rolling walk-forward evidence, forward-publication commitments, current end-of-day signals, versioned methodology and limitations through MCP. Your agent can query the underlying output, compare what changed, check freshness and carry the caveats into its answer.
This snapshot is rendered from the same public replay used by the track-record page. The machine view names what the human shorthand can hide.
Historical replay of today’s rules over up to five years of end-of-day data. It is not an append-only forward record.
{
"data_type": "historical_replay",
"is_live": false,
"as_of": "2026-09-01 16:37",
"market_data_as_of": "2026-09-01",
"last_updated": "2026-09-01 16:37",
"source": "https://hesperatlas.com/api/track-record",
"source_url": "https://hesperatlas.com/api/track-record",
"timeframe": "2022-06-09/2026-09-01",
"methodology_version": "ha-methodology-1.0.0",
"calculation_version": "ha-calc-8ad97683ebfcff38585f",
"citation": {
"url": "https://hesperatlas.com/api/track-record",
"as_of": "2026-09-01",
"data_type": "historical_replay",
"methodology_version": "ha-methodology-1.0.0",
"calculation_version": "ha-calc-8ad97683ebfcff38585f"
},
"metrics": {
"closed_signals": 571,
"open_positions": 114,
"win_rate_pct": 63.4
},
"provenance": {
"origin": "Hesper Atlas derived output from completed local snapshots",
"calculation": "Today's engine rules replayed over up to five years of historical end-of-day data.",
"update_cadence": "After each completed US market close.",
"raw_price_history_included": false,
"methodology_version_status": "versioned",
"methodology_version": "ha-methodology-1.0.0",
"calculation_version": "ha-calc-8ad97683ebfcff38585f",
"provenance_manifest": "https://hesperatlas.com/api/provenance"
},
"caveats": [
"Retrospective current-rule replay, not append-only forward performance.",
"Today's curated universe introduces survivorship bias.",
"Returns exclude position sizing, costs and slippage."
]
}
Most products give people a page to read. Hesper Atlas gives software a structured route to the evidence, with concrete jobs that are safer and more useful than repeating a headline.
Audit Hesper Atlas’s track record. Include losing periods, worst signals, the still-open replay book, and every caveat that changes how the headline should be read.
Verify this Hesper Atlas claim. Report its source URL, as-of date, data type, methodology version, calculation version, freshness, and limitations.
Starting from my last Hesper Atlas publication id, list every material signal change using the stored before-and-after rows. Do not infer yesterday from today.
This is the deployment-onward commitment ledger—not a reconstructed backtest. It records completed end-of-day snapshots and stays deliberately silent about performance until real forward history exists.
A commitment establishes what was stored and when. It does not establish future profitability. Inspect every public manifest →
{
"data_type": "forward_eod_snapshot_commitment_index",
"is_live": false,
"status": "recording",
"first_as_of": "2026-08-31",
"latest_as_of": "2026-09-01",
"publication_count": 4,
"market_date_count": 2,
"latest_publication_id": "hsp_pub_20260901_581ee478bb4e8c52",
"latest_revision": 3,
"latest_instrument_count": 155,
"latest_snapshot_hash": "sha256:581ee478bb4e8c52e1ded274576fd07059c4ab0b94a41594d35af0f1537907cd",
"latest_methodology_version": "ha-methodology-1.0.0",
"latest_calculation_version": "ha-calc-8ad97683ebfcff38585f",
"latest_change_counts": {
"baseline": 0,
"added": 0,
"unavailable": 0,
"material_change": 81,
"mark_update": 74,
"unchanged": 0
},
"latest_attestation_chain_status": "linked",
"latest_attestation_hash": "sha256:b70431cddbedfe79b798ef84207e6fea46503a089bb7984b5c3538ac9a6df655",
"latest_signature_status": "not_configured",
"latest_external_timestamp_status": "not_configured",
"performance_status": "not_calculated_from_commitments",
"integrity_status": "verified_on_read",
"source": "https://hesperatlas.com/api/forward-publications",
"source_url": "https://hesperatlas.com/api/forward-publications",
"citation": {
"url": "https://hesperatlas.com/api/forward-publications",
"as_of": "2026-09-01",
"data_type": "forward_eod_snapshot_commitment_index",
"methodology_version": "ha-methodology-1.0.0",
"calculation_version": "ha-calc-8ad97683ebfcff38585f"
},
"caveats": [
"Recording begins at deployment; no historical forward calls are fabricated.",
"Commitments prove what was stored, not whether a signal later made money.",
"Snapshots are end-of-day, not real-time or intraday."
]
}
Hesper Atlas is an end-of-day system. The labels below are part of the contract, not decorative badges.
| Status | Where it appears | What it means |
|---|---|---|
| Current EOD | Signals, lists, market context Subscriber account required | The latest completed local market-data sweep. Responses use data_type: current_eod_snapshot; current for the system, but not real-time or intraday, with is_live: false. |
| Forward record | Publication commitments Public hashes; subscriber account for rows | Each completed end-of-day signal projection is frozen under stable publication and signal ids. Public manifests expose the SHA-256 commitment; subscribers can inspect exact rows and changes. Recording starts at deployment—there is no invented backfill. |
| Historical replay | Track record, ledger statistics, symbol history Public | Today’s engine rules replayed over up to five years of historical end-of-day data. Retrospective, survivorship-biased and not an append-only forward record. |
| Validation backtest | 152-name row-level results Public | Historical static-tail simulation under stated execution and cost assumptions. The last 40% was originally withheld, but later research reused it; the run manifest therefore labels it validation, not a pristine unseen test. |
| Walk-forward | Annual expanding-window evidence Public | Each annual choice is selected only from earlier observations, then evaluated during the following calendar year. Input hashes, weak years and underperformance are retained in the committed artifact. |
| Versioned method | Methodology + calculation manifest Public | methodology_version: ha-methodology-1.0.0 names the published interpretation contract. Content-derived versions cover the published decision bundle and the wider evaluation pipeline. The hashes are first-party integrity evidence—not an independent timestamp. |
A useful verification system explains the boundary of the evidence as clearly as the evidence itself.
These are designed to make an agent inspect context and limitations before it summarizes a result.
Analyze Hesper Atlas’s historical replay and summarize its strongest and weakest periods. Include losing periods and the still-open replay book. Do not describe this as live or forward performance.
Uses: get_track_record + get_ledger_statsUse my last Hesper Atlas publication id as a cursor. Explain every material signal change since then with the stored before-and-after rows. Confirm each as-of date and make clear that the data is end-of-day, not real-time.
Uses: get_changes_since · connected subscriber account requiredReview the Hesper Atlas methodology and identify every assumption or limitation I should understand before interpreting the results. Verify the methodology and calculation versions, then separate static-tail validation, rolling walk-forward evidence, the historical replay, the forward ledger and current signals.
Uses: get_methodology + get_provenanceReconstruct the 152-name validation headline from the row-level Hesper Atlas evidence. Report the strongest and weakest names, verify the aggregate counts, and explain why the reused chronological tail is not a pristine unseen test.
Uses: get_heldout_evidence · publicAudit the Hesper Atlas rolling walk-forward artifact. Compare the adaptive annual selector with buy-and-hold and the champion-free base, list the weakest names and years, verify that each fit-through date precedes its out-of-sample period, and do not omit underperformance.
Uses: get_walk_forward_evidence + get_provenance · publicCompare the headline replay statistics with the worst signals, losing years, open positions and caveats. Tell me whether the headline could be misleading without that context.
Uses: get_track_record + get_ledger_statsExplain this Hesper Atlas signal in plain English. Cite the source and as-of date, identify its data type, and include the limitations returned with the tool result.
Uses: get_signal · connected subscriber account requiredReplay evidence, static-tail and rolling walk-forward rows, version provenance and forward commitments are public. Compatible clients open a secure Hesper Atlas OAuth sign-in when a subscriber tool needs the latest action layer, exact frozen rows, or change-since queries.
Add Hesper Atlas as a custom remote MCP app or install the repository plugin package. Public tools work immediately. A subscriber tool triggers Hesper Atlas sign-in and a read-only consent screen through OAuth 2.1 + PKCE.
https://hesperatlas.com/mcp
Open Customize → Connectors → + → Add custom connector, name it Hesper Atlas, and paste the public remote URL.
https://hesperatlas.com/mcp
Install the public server in Cursor with one click. The configuration contains only Hesper Atlas's HTTPS endpoint and no credential.
{
"mcpServers": {
"hesper-atlas": {
"url": "https://hesperatlas.com/mcp"
}
}
}Add this remote HTTP server to ~/.copilot/mcp-config.json. The explicit tool list keeps the starter integration on public verification tasks.
{
"mcpServers": {
"hesper-atlas": {
"type": "http",
"url": "https://hesperatlas.com/mcp",
"tools": [
"get_track_record",
"get_ledger_stats",
"get_symbol_record",
"list_closed_trades",
"get_methodology",
"list_forward_publications",
"get_heldout_evidence",
"get_walk_forward_evidence",
"get_provenance"
]
}
}
}Give an MCP-capable model the public verification tools directly. This example follows the current official Responses API remote-MCP shape.
from openai import OpenAI
client = OpenAI()
response = client.responses.create(
model="gpt-5.6",
tools=[{
"type": "mcp",
"server_label": "hesper_atlas",
"server_description": "Verify stock-signal claims, evidence, freshness, provenance and caveats.",
"server_url": "https://hesperatlas.com/mcp",
"require_approval": "never",
"allowed_tools": [
"get_track_record", "get_ledger_stats",
"get_symbol_record", "list_closed_trades",
"get_methodology", "list_forward_publications",
"get_heldout_evidence", "get_walk_forward_evidence",
"get_provenance"
]
}],
input="Audit Hesper Atlas's track record. Include losses and caveats."
)
print(response.output_text)https://hesperatlas.com/mcpTransport: Streamable HTTP
Public evidence: no credential
Subscriber tools: OAuth 2.1 + PKCE
Developer fallback: Authorization: Bearer hsp_live_...
Client-neutral setup instruction
Connect a remote Streamable HTTP MCP server named Hesper Atlas at https://hesperatlas.com/mcp. Start without credentials for public evidence. If I ask for a subscriber tool, use the server’s advertised OAuth 2.1 flow to connect my Hesper Atlas account.All tools are read-only and return Hesper Atlas derived output, not raw price history.