Skip to content
chartgen

0007 — Complete series serialization: fills, hbars, divlines

Status: Accepted Date: 2026-04-19

Context and Problem Statement

Section titled “Context and Problem Statement”

src/mcp.rs::panel_result_to_series_json — the serializer behind generate_chart under format=series — covered only 4 of the 7 channels that a PanelResult can carry. lines, bars (as histogram), dots (as signals), and hlines were emitted; fills, hbars, and divlines were silently dropped on the wire.

Those three channels are not decorative detail. They are the only way to render a whole class of chartgen’s 33 indicators:

  • fills — shaded regions between two y-series: Bollinger Bands, Keltner Channels, Ichimoku cloud, VWAP bands.
  • hbars — horizontal bars at price levels: VPVR (Volume Profile Visible Range), Session Volume Profile, HVN/LVN nodes, Naked POC, TPO.
  • divlines — two-point diagonal segments: WaveTrend divergences, RSI divergences emitted by Cipher B.

Without them, a JSON-consuming client has to either fall back to the PNG output or skip those indicators entirely. That blocks TradingView-parity rendering on the wire, which is the whole point of format=series.

Decision Drivers

Section titled “Decision Drivers”
  • format=series is documented as the “full per-bar time-series” shape for chartgen. Dropping channels contradicts that contract.
  • TV-parity rendering for BBands-style bands, VPVR-style overlays, and divergence markers requires these channels; they cannot be recomputed client-side without duplicating indicator logic.
  • The change must be additive — existing consumers reading lines, signals, and so on must not see their payload shape shift.

Considered Options

Section titled “Considered Options”

Keep dropping fills, hbars, divlines. Zero work, but locks the frontend out of VPVR, band-fill regions, and divergence lines for every indicator that emits them. Rejected — this is precisely what the PR is meant to fix.

Add a separate format=full that includes these channels, leave series alone. Splits the API surface for no real benefit. series is already described as the full per-bar view; a second “really full” format is a worse contract, not a better one. Clients end up having to know the difference, and documentation has to explain why there are two. Rejected.

Ship the raw bars[] and have the client recompute the missing channels from price data. Wastes work on both sides: the server already has an authoritative indicator engine, and duplicating that engine in the client is expensive and drifts the moment an indicator changes. Rejected.

Emit the three channels in the existing series payload, additively. Matches the documented “full per-bar time-series” contract, keeps the API surface flat, and preserves backward compatibility for current consumers because the new keys are only present when the indicator emits them. Chosen — see below.

Decision Outcome

Section titled “Decision Outcome”

Chosen option: emit fills, hbars, and divlines alongside the existing channels in panel_result_to_series_json. The shape is additive and mirrors the existing conventions:

  • fills: array of { y1: [...], y2: [...] }. Each array is 1:1 with bars[]; pre-warmup values are null via the existing finite_or_null helper. Colors chosen server-side are dropped — the client themes per indicator on its side.
  • hbars: array of { y, height, width, offset, left }. Field names preserve the exact semantics of chartgen::indicator::HBar — y and height in price space, width and offset as 0.0–1.0 fractions of chart width, and the boolean left that picks the anchor edge. Color dropped.
  • divlines: array of { x1, t1, y1, x2, t2, y2, dashed }. Each endpoint carries both x (bar index) and t (unix seconds, or null for synthetic sample data), mirroring the signal pattern. Color dropped.

All three channels are omitted from the per-indicator object when the indicator emits none, matching the convention already used for lines, histogram, signals, and hlines. The summary serializer (panel_result_to_json) is unchanged; summary keeps its last-value semantics.

Consequences

Section titled “Consequences”

Positive. Volume-profile indicators (VPVR, Session VP, HVN/LVN, Naked POC, TPO) are now fully renderable from the JSON payload. Band fills for BBands/Keltner/Ichimoku/VWAP land on the wire. Divergence segments from WaveTrend and Cipher B come through as data rather than being flattened to signals only. API consumers reach parity with the image output for every indicator chartgen ships.

Negative. Wire size grows for indicators that emit many hbars — VPVR with 32 bins adds roughly 1 KB of JSON per indicator. Three new Zod schemas (SeriesFillSchema, SeriesHBarSchema, SeriesDivLineSchema) have to be kept in sync with the Rust serializer, same as the existing per-response schemas (see ADR-0004).

Neutral. Non-breaking for current consumers because the new keys are conditional on non-empty channels. Frontend rendering for the three new channels is out of scope for this change and will land in a follow-up PR.

More Information

Section titled “More Information”