Historical Market Context Engine
Pre-computed
historical context
for every
S&P 500 symbol.
Updated daily.
Outcome distributions. Reliability scores. Match quality. LLM-ready commentary. Delivered as structured JSON — for quant traders and AI tool builders who need historical market context without building the infrastructure themselves. The engine runs every close. You access the output via API.
~500
S&P 500
symbols
symbols
8×4
Lookback ×
forward windows
forward windows
4,000+
Analyses updated
daily
daily
10+yr
Historical
match depth
match depth
{ "meta": { "data_through": "2026-06-17", "timeframe": "1d", "universe": "sp500", "lookback_periods": 15 }, "benchmark": { "symbol": "AAPL", "exchange": "NASDAQ", "name": "Apple Inc.", // context spans all 8 lookbacks, not just this file's "context": { "10": { ... }, "15": { ... }, ... } }, // evidence base for this lookback "analysis": { "evidence_count": 4, "unique_symbols": 4, "diversity": { "score": 1.0 }, "match_quality": { "quality_tag": "strong" }, // all 4 forward windows live in one file "forward_outcome": { "5": { "returns": { "positive_return_rate": 0.50, "median_return": 0.0531, "percentiles": { "p05": -0.0102, "p50": 0.0531, "p95": 0.1505 } }, "tags": { "direction": "mixed", "consistency": "low", "reliability": "low", "risk": "elevated", "outlier": "normal" }, // LLM-ready, no preprocessing needed "commentary": { "headline": "Mixed historical outcomes on limited evidence", "summary": "Similar setups appeared 4 times historically. Over the next 5 trading days, outcomes were mixed, with a positive-return rate of 50% and a median move of +5.3%. The sample was limited, so this should be treated as weak historical evidence.", "takeaway": "Historical evidence is limited, so this setup should be treated as exploratory rather than actionable." } }, "10": { ... }, "15": { ... } } }, // every individual historical analog, unordered "matches": [ { "symbol": "ALB", "similarity_score": 0.384, ... }, ... ]}
↓ scroll
What we produce
Raw price data
is everywhere.
Pre-computed historical
market context isn't.
Alpha Vantage, Polygon, yFinance — they all give you OHLC. What none of them give you is the analytical layer on top: which historical conditions are statistically similar to today, what happened next across the full distribution, and how much you can trust the evidence.
Konseki runs that analysis every market close across ~500 S&P 500 symbols, 8 lookback periods, and 4 forward windows. The output is structured JSON, ready for dataframes, backtests, and LLM prompts. No preprocessing. No infrastructure to build.
| Raw data providers | Konseki | ||
|---|---|---|---|
| ✗ | OHLC price history | ✓ | Pre-computed historical context |
| ✗ | Win rate (if any) | ✓ | Full distribution P05 P95 |
| ✗ | No sample transparency | ✓ | Match count + date range visible |
| ✗ | No match quality signal | ✓ | Reliability + match quality scores |
| ✗ | Requires LLM preprocessing | ✓ | LLM-ready commentary field |
| ✗ | Build the analysis yourself | ✓ | 4,000+ analyses updated daily |
How the engine works
01 ──
Daily data ingestion
OHLC data across the full S&P 500 universe. 10+ years of historical matches. Ingested and processed every market close.
02 ──
Cross-market pattern matching
The engine scans all stocks across all history for conditions statistically similar to today — not just within the same ticker. Broader evidence base, stronger signal.
03 ──
Distribution analysis
Forward returns computed across P05 through P95 for each match set. MAE, MFE, reliability score, and match quality — not a headline win rate.
04 ──
Structured JSON output
Self-documenting JSON with a consistent schema across all symbols and lookback combinations. Ready for pandas, your backtesting layer, or direct LLM injection.
What each output contains
forward_outcome.{N}.returns.percentiles { p05 p95 }
Full percentile distribution of forward returns, for every forward window. Understand the shape of what historically happened next — not just the average or a single predicted value.
evidence_count + unique_symbols + diversity
Every claim is backed by a visible sample. See exactly how many historical analogs were found, across how many distinct symbols, and how diverse the sample is in time and symbol — evaluate the evidence yourself.
match_quality { quality_tag + component scores }
Match quality scored 1–5 across six dimensions per match — shape, trend, volatility, range, volume, risk — not a single opaque number. Full transparency into the matching methodology.
max_adverse_excursion + max_favourable_excursion
Path matters as much as final outcome. MAE and MFE show how far price historically moved against and for each setup before resolution.
forward_outcome.{N}.commentary
Natural language summary structured for direct LLM prompt injection — headline, summary, distribution, risk, reliability, match quality, and takeaway. Give your model historical context, not raw numbers to interpret itself.
tags { direction · consistency · reliability · risk · outlier }
Five machine-readable signal tags per forward window, for programmatic triage. Filter and route across 500 symbols without parsing every full distribution.
Who uses Konseki
Quant traders
Your backtests just got a historical context layer.
You already think in outcome distributions and historical evidence. You already validate ideas statistically. Konseki produces the historical context you'd spend days building — 500 symbols, 8 lookback periods, updated daily. Consume the output, not the infrastructure.
- →Load directly into pandas — consistent schema across all symbols
- →Filter by reliability and match quality programmatically
- →Cross-reference with your own strategy signals
- →Full P05–P95 distribution — not a single win rate
- →8 lookback periods: 5 / 10 / 15 / 20 / 25 / 30 / 40 / 50 days
# load historical context into a dataframeimport requests, pandas as pd r = requests.get( "https://api.konseki.io/v1/analysis/AAPL-NASDAQ", params={"lookback":15}, headers={"X-API-Key": "ea_live_..."})fwd = r.json(()["analysis"]["forward_outcome"]["5"])dist = pd.Series(fwd["returns"]["percentiles"])
AI tool builders
Historical context as an ingredient for your agent.
Building a trading assistant, research agent, or analysis workflow? Konseki outputs are structured for direct LLM prompt injection. The forward_outcome.commentary block is written for model consumption. Give your agent historical context, not raw price data.
- →LLM-ready
commentaryblock in every forward window - →Consistent JSON schema — no per-symbol quirks
- →Two clean endpoints, API key auth, clear rate limits
- →Machine-readable tags for agent routing logic
- →MCP server integration planned
# inject historical context into an LLM promptcontext = fwd["commentary"]["summary"] prompt = f"""Historical context for AAPL (5-day forward):{context} Given this context, analyse the currentposition sizing decision for AAPL..."""
Pricing
Free
$0 / month
Full S&P 500 access, no symbol restrictions. The daily request cap works out to roughly 250 symbols at full depth — plenty to evaluate the product properly.
- ✓Full S&P 500 universe — query any symbol
- ✓All 8 lookback periods, all 4 forward windows
- ✓Full JSON output — every field
- ✓Updated daily
- ✓2,000 requests per day~250 symbols at full depth
Pro
$29 / month
The full S&P 500 universe, every lookback, every day — with room to spare for historical date queries. Cancel any time.
- Everything in Free, plus:
- ✓30,000 requests per day15× Free
- ✓Full daily snapshot covered 7× over — plenty left for historical dates
- ✓Priority support
Free and Pro cover the S&P 500 universe at the limits above. Need broader coverage or a higher request ceiling? Get in touch.
Historical market context engine
We compute it.You consume it.
Free tier available. No credit card required to start.