Forecasts

Next-48h hourly demand forecast. National by default; pass ?state=<key> for a per-state forecast. Returns the demand_mw point (= p50) with the p10/p50/p90 uncertainty band and a per-IST-day peak head. Honest provenance: source is 'model' only when the trained 48h model is promoted to serving, and served is strictly source == 'model' (otherwise a seasonal-naive / climatology baseline is served with served=false).

{
  "data": {
    "scope": "national",
    "state": "All India",
    "issued_at": "2026-07-04T06:00:00+00:00",
    "as_of": "2026-07-04T06:00:00+00:00",
    "horizon_hours": 48,
    "source": "model",
    "served": true,
    "model_version": "national-48h-v3",
    "coverage": { "band": "p10-p90", "notes": "national-48h-v3; anchor=recent-median" },
    "peaks": [
      { "date_ist": "2026-07-04", "peak_mw": 238412.0, "peak_hour_ist": 15 }
    ],
    "hours": [
      { "target_hour_ist": "2026-07-04T12:00:00", "hour_ist": 12,
        "demand_mw": 236010.0, "p10_mw": 219100.0, "p50_mw": 236010.0, "p90_mw": 252920.0 }
    ]
  },
  "meta": { "scope": "national", "source": "model", "served": true }
}

demand_mw equals p50_mw exactly — the same number every Atlas surface (the forecast page, /states, and the daily email) reads for the same issue. The old /forecast/national and /forecast/state/{state_key} paths are aliases of this endpoint (identical response, retained for back-compat), and /forecast/ledger — the predicted-vs-actual accuracy history — sits behind the same gate.

Hour-by-hour carbon intensity forecast. Each hour is the trailing hour-of-day average intensity for the scope, anchored to the latest observed value with a 6h fade so the series joins the actuals without a step. Any API key may call the national scope.

{
  "data": {
    "items": [
      { "timestamp": "2026-07-04T15:00:00+05:30", "state": "All India", "state_slug": "all-india",
        "carbon_intensity_gco2_kwh": 688.4, "intensity_class": "high",
        "total_generation_mw": 198420.5, "is_forecast": true }
    ],
    "count": 24,
    "forecast_hours": 24,
    "state_slug": null,
    "unit": "gCO2/kWh"
  },
  "meta": { "state": null, "requested_hours": 24, "returned_hours": 24, "order": "desc" }
}

Timestamps are IST (+05:30). The plan window bites here rather than the entitlement: Sandbox, Grid AI, Pro, and Starter clamp at 7 days (168 h), Student and Institutional at 14 days (336 h), and Growth and Enterprise are uncapped.

Implied DAM price curve in monthly buckets, computed from settled IEX day-ahead history — not a forward-looking model output. Each month returns the average, median, P25/P75, min, max and standard deviation of market clearing price, plus an off-peak / solar / evening / night time-of-day profile.

{
  "data": {
    "monthly_series": [
      { "month": "2026-06", "block_count": 2880,
        "avg_mcp": 4218.36, "median_mcp": 3990.0,
        "p25_mcp": 3180.5, "p75_mcp": 5010.25,
        "min_mcp": 1500.0, "max_mcp": 10000.0, "stddev_mcp": 1487.92 }
    ],
    "tod_profiles": {
      "2026-06": {
        "off_peak": { "block_count": 720, "avg_mcp": 3410.18, "median_mcp": 3290.0 },
        "solar":    { "block_count": 1440, "avg_mcp": 3688.42, "median_mcp": 3520.0 },
        "evening":  { "block_count": 480, "avg_mcp": 6120.77, "median_mcp": 5980.0 },
        "night":    { "block_count": 240, "avg_mcp": 4402.11, "median_mcp": 4310.0 }
      }
    },
    "overall_avg_mcp": 4218.36,
    "months_available": 1,
    "lookback_months": 1
  },
  "meta": { "requested_months": 12, "returned_months": 1 }
}

Prices are INR/MWh. Months are IST calendar months (YYYY-MM) returned in ascending order — the axis is semantic, so this route is exempt from the order parameter. Time-of-day buckets are IST: off-peak 00:00–06:00, solar 06:00–18:00, evening 18:00–22:00, night 22:00–24:00.

Most-recently published day-ahead 96-block price forecast: p10/p50/p90 plus a price-cap-hit spike probability per 15-minute block, with issuing timestamp, model version and staleness in minutes. This is the block-level product the forward curve is often mistaken for.

DAM forecasts publish once daily at 00:30 UTC (06:00 IST), ahead of the exchange's 11:00 IST bid gate. Starter and Growth keys read this single curve and nothing else of the product — the response reports meta.access: "dam_slice" rather than "full". Every other plan receives 403 upgrade_required.

The rolling 14-day day-ahead-market outlook — one row per IST delivery day, D+1 to D+14, issued once a day after the DAM auction clears (about 06:10 IST). Each row carries both day-average bases: the volume-weighted price (VWAP p10/p50/p90 and mean — what the market actually paid per MWh, the headline) and the unweighted 96-block mean (p05..p95), plus the four time-of-day medians, the probability that the 18–22h evening clears at the ₹10,000/MWh ceiling and the expected number of capped blocks. Every number comes from one joint simulation per delivery day, so blocks, time-of-day averages, cap risk and the day average agree with each other.

{
  "data": [
    { "target_date": "2026-09-16", "horizon_days": 1,
      "vwap_q10": 4269.0, "vwap_q50": 5555.0, "vwap_q90": 5853.0, "vwap_mean": 5290.0,
      "level_q05": 5310.0, "level_q10": 5920.0, "level_q25": 6870.0, "level_q50": 7733.0,
      "level_q75": 8410.0, "level_q90": 9010.0, "level_q95": 9340.0, "level_mean": 7690.0,
      "tod_off_peak_q50": 4210.0, "tod_solar_q50": 2980.0, "tod_evening_q50": 10000.0, "tod_night_q50": 6120.0,
      "tod_evening_q90": 10000.0, "tod_evening_pcap": 0.99,
      "n_cap_mean": 61.4, "n_cap_q10": 44.0, "n_cap_q90": 71.0,
      "day_max_q50": 10000.0, "day_max_q90": 10000.0, "cap_rs_mwh": 10000.0,
      "model_version": "mh-v2", "issued_at": "2026-09-15T00:40:00+00:00" }
  ],
  "meta": { "market": "DAM", "product": "outlook_14d", "as_of": "2026-09-15", "delivery_days": 14,
            "issued_at": "2026-09-15T00:40:00+00:00", "model_version": "mh-v2",
            "freshness": { "issued_at": "2026-09-15T00:40:00+00:00", "staleness_minutes": 412 } }
}

meta.as_of is the last cleared delivery day the issue was built from; horizon_days counts from it. Measured out of sample over 379 delivery days (Sep 2025 – Sep 2026): VWAP median error 9.7% at D+1, about 16% at D+7 and D+14, with the 80% bands covering 75–81% of outcomes; block-level skill is largest at one to three days ahead, and the cap probability is a lower bound while the market tightens.

The 96 fifteen-minute blocks of one delivery day inside the latest 14-day outlook: p05..p95, the mean and the probability that the block clears at the ceiling. 404 no_forecast when the requested day is not in the latest issue.

{
  "data": [
    { "target_ts": "2026-09-21T18:30:00+00:00", "block": 0, "horizon_days": 7,
      "q05": 2420.0, "q10": 2860.0, "q25": 3520.0, "q50": 4400.0, "q75": 5060.0, "q90": 5940.0, "q95": 6600.0,
      "mean_rs_mwh": 4532.0, "p_cap": 0.04, "model_version": "mh-v2" }
  ],
  "meta": { "market": "DAM", "product": "outlook_14d", "target_date": "2026-09-22", "horizon_days": 7,
            "issued_at": "2026-09-15T00:40:00+00:00", "model_version": "mh-v2",
            "freshness": { "issued_at": "2026-09-15T00:40:00+00:00", "staleness_minutes": 412 } }
}

Block 0 starts at 00:00 IST; target_ts is the block start in UTC. Blocks of one simulated day are tied together by the rank pattern of a real historical day, so the evening peak caps together or not at all — read p_cap with the day's n_cap_mean from /outlook/latest.