@battlegrid/mcp-server 32.0.84 → 32.0.86
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/README.md +121 -0
- package/dist/index.d.ts +1 -1
- package/dist/index.js +1 -1
- package/package.json +1 -1
package/README.md
CHANGED
|
@@ -38,6 +38,127 @@ Seeing package `31.x` alongside handshake `battlegrid@33.x` — the package **be
|
|
|
38
38
|
|
|
39
39
|
**What this changes for you:** nothing about how you call anything. Upgrading the package no longer waits on a server deploy, and a server deploy no longer strands you on a package that names the wrong contract — reconnect and the announcement follows. **Contract breaking-change notes are no longer keyed to package versions**, since a contract move is no longer a release here; the v11-and-earlier notes below are kept as history, and the live vocabulary is always discovery.
|
|
40
40
|
|
|
41
|
+
## Contract history — v156 (a crowd reading no bar can serve is not a condition; trades name their config version)
|
|
42
|
+
|
|
43
|
+
**v156 is MAJOR: writes v155.1 accepted are refused, and one published result narrows.** No input schema moves.
|
|
44
|
+
|
|
45
|
+
### Refused input — the first reason v156 is MAJOR
|
|
46
|
+
|
|
47
|
+
- **A condition on a crowd reading no decided bar can serve is refused**, entry or exit, with `VALIDATION_ERROR`
|
|
48
|
+
and authoring code `CONDITION_OPERAND_UNSERVABLE`. Two readings are refused:
|
|
49
|
+
- a live crowd header: the eight `…Live` crowd codes (`pickLive`, `upBiasLive`, `crowdAccLive`, `captRateLive`,
|
|
50
|
+
`upLeanZLive`, `upExcessZLive`, `captExcessZLive`, `accEdgeZLive`), written before the game's candle closes;
|
|
51
|
+
- any crowd header on a main candle no Market Grid time range settles on.
|
|
52
|
+
- **Where it refuses:** every strategy write (`commit_strategy_draft`, `restore_strategy`, `fork_strategy`,
|
|
53
|
+
`apply_officer_proposal`) and the radar deploy, rebind and preflight checks (`deploy_agent_to_radar`,
|
|
54
|
+
`commit_radar_deployment_draft`, an agent rebind through `commit_agent_draft`, `preview_radar_resolution`,
|
|
55
|
+
`preview_agent_radar_deploy`, `propose_entry_decision`). Reports and prompts still read those columns.
|
|
56
|
+
|
|
57
|
+
### Narrowed output — the other reason
|
|
58
|
+
|
|
59
|
+
- **`preview_strategy_report`'s `authoring.conditionColumns` no longer lists the refused headers**, so the same
|
|
60
|
+
`detail: "detailed"` request returns fewer condition columns.
|
|
61
|
+
|
|
62
|
+
### Widened acceptance
|
|
63
|
+
|
|
64
|
+
- **A settled crowd exit rule is accepted and deployable.** An exit rule on a settled crowd reading (for example
|
|
65
|
+
`upLeanZ`) now carries the exit role and deploys on a radar slot. Each decided bar reads the sessions settled at
|
|
66
|
+
or before its close. A coin with no Market Grid cell in that window, or no settled session, leaves the rule
|
|
67
|
+
unread, so it never fires. Exit rules on the four convergence readings (`FLOW_ALIGN`, `SMART_RETAIL`,
|
|
68
|
+
`CAPTAIN_CONF`, `CONFIDENCE`) are still refused on the exit lane.
|
|
69
|
+
|
|
70
|
+
### Widened output
|
|
71
|
+
|
|
72
|
+
- **Trades trace to a config version.** `configVersion` is added to `get_agent_explorer`'s entries (a number, the
|
|
73
|
+
agent's current config version), and to `get_public_agent_open_positions`, `get_public_agent_realized_trades`,
|
|
74
|
+
`get_trade_chart` and `get_public_agent_trade_chart` (`number | null`, the version the trade's decision names).
|
|
75
|
+
- **A firing condition exit records the crowd it read.** The condition-exit evidence on `get_signal_log` and
|
|
76
|
+
`get_public_agent_signal_log_detail` gains `crowdSettledAt` (`string | null`).
|
|
77
|
+
|
|
78
|
+
### Changed values — the same schema serves a different value
|
|
79
|
+
|
|
80
|
+
- **A settled crowd header's `conditionHold.refusal` reads `HISTORY`** where it read `NOT_REWINDABLE`. A hold over
|
|
81
|
+
it still counts one bar.
|
|
82
|
+
|
|
83
|
+
### Re-worded text
|
|
84
|
+
|
|
85
|
+
- The `preview_radar_resolution`, `propose_entry_decision` and `restore_strategy` descriptions state that an exit
|
|
86
|
+
rule deciding nothing else is read by the exit sweep.
|
|
87
|
+
|
|
88
|
+
## Contract history — v155 (every breakdown section serves its highlights)
|
|
89
|
+
|
|
90
|
+
**v155 is MAJOR: every breakdown envelope gains a required field.** Inputs are unchanged.
|
|
91
|
+
|
|
92
|
+
### Reshaped output — the reason v155 is MAJOR
|
|
93
|
+
|
|
94
|
+
- **Every breakdown envelope carries `highlights`.** This covers the agent and breakdown doors of Cost Economics,
|
|
95
|
+
LLM spend, pipeline blocks, review slices, conditions, Entry Discipline, Decision Quality and Conviction
|
|
96
|
+
Calibration.
|
|
97
|
+
- `highlights` is the section's server-computed findings, in policy order, empty when none fires.
|
|
98
|
+
- Each finding is `{ policy, kind, subject, value, sampleSize }`.
|
|
99
|
+
- `kind` is one of `LARGEST_CONTRIBUTOR`, `PULLS_PARENT_NEGATIVE`, `DEVIATION`, `NEGATIVE_EDGE` or `HALTED`.
|
|
100
|
+
- `subject` is an agent, a brain row (agent and model) or a condition row (strategy, revision, condition, and
|
|
101
|
+
whether you may open that strategy).
|
|
102
|
+
- `value` is the figure the policy read, in the policy's unit, and `sampleSize` is its population (null for a halt).
|
|
103
|
+
- A highlight fires only above its section's readiness gate. Entry Discipline serves an empty list.
|
|
104
|
+
- A strict client validating the envelope must admit the field. Nothing else moves.
|
|
105
|
+
|
|
106
|
+
## Contract history — v154 (Decision Quality and Calibration compare agents and models, in R)
|
|
107
|
+
|
|
108
|
+
**v154 is MAJOR: two published outputs are reshaped and lose fields.** Two tools are added.
|
|
109
|
+
|
|
110
|
+
### Reshaped output — the reason v154 is MAJOR
|
|
111
|
+
|
|
112
|
+
- **`get_agent_decision_quality` answers the brain comparison envelope.** It carries `scope`, `window` (the evidence
|
|
113
|
+
window), `anchor` (`DECIDED`), `minSampleSize` and `rows`.
|
|
114
|
+
- There is one row per model the agent's decisions ran on, never pooled. Each row carries `agent`, `model`, and
|
|
115
|
+
`isCurrentModel`.
|
|
116
|
+
- `rank` is 1-based among the ready rows; it is `null` below the gate.
|
|
117
|
+
- `figures` is the old figure set without `windowDays`.
|
|
118
|
+
- The exit block gains `unmeasuredCount`, `meanExitR`, `meanGrossR` (the ranking figure) and `grossOfCosts`.
|
|
119
|
+
- Read the row with `isCurrentModel`.
|
|
120
|
+
- **`get_agent_conviction_calibration` answers the same envelope** with `anchor` `CLOSED`.
|
|
121
|
+
- `figures` loses `windowDays` and gains `brierScore`, the ranking figure, where lower is better.
|
|
122
|
+
- Every band and deployment-rule group loses `averageNetPnlUsd` and gains `averageGrossR`, with `rTradeCount`.
|
|
123
|
+
Compare groups in R, not dollars.
|
|
124
|
+
|
|
125
|
+
### Added tools
|
|
126
|
+
|
|
127
|
+
- **`get_decision_quality_breakdown` and `get_conviction_calibration_breakdown`** serve the same envelopes. They take
|
|
128
|
+
`{ teamId?, strategyId?, fromRevision?, toRevision?, agentId? }`, for the account, one team, or a strategy cohort
|
|
129
|
+
narrowed to a team or an agent. The result has one row per (agent, model) and takes no period and no grouping.
|
|
130
|
+
|
|
131
|
+
## Contract history — v153 (strategy revision cohorts and one evidence window)
|
|
132
|
+
|
|
133
|
+
**v153 is MAJOR: published outputs lose fields, one tool's input is replaced, and a finding input is narrowed.** Three
|
|
134
|
+
tools are added.
|
|
135
|
+
|
|
136
|
+
### Refused input — one reason v153 is MAJOR
|
|
137
|
+
|
|
138
|
+
- **`get_agent_entry_discipline` takes `{ agentId }` alone.** `windowStartAt` and `windowEndAt` are refused. The read
|
|
139
|
+
covers the evidence window, the last 90 days ending at the read.
|
|
140
|
+
- **`submit_officer_review`'s CONDITION finding requires `strategyId` and `strategyRevision`**, copied from the
|
|
141
|
+
`get_agent_conditions` row it cites. A finding naming no row of the agent's condition evidence is refused.
|
|
142
|
+
|
|
143
|
+
### Reshaped output — the other reason v153 is MAJOR
|
|
144
|
+
|
|
145
|
+
- **`get_agent_entry_discipline` answers the scoped-section envelope `{ entryDiscipline }`.** Its `adoption` is one
|
|
146
|
+
row per strategy revision the decisions were read under, and it gains `unstampedDecisions`.
|
|
147
|
+
- **`get_agent_review_analytics` drops `condition`.** Cited conditions are `get_agent_conditions`' rows, keyed by
|
|
148
|
+
strategy revision. Its flat window fields give way to `window`.
|
|
149
|
+
- **Every breakdown envelope's `window` gains `policy`.** It is `PERIOD` or `EVIDENCE`.
|
|
150
|
+
- **Every breakdown envelope's `scope` gains a `STRATEGY` arm.** The arm carries the cohort, every stored revision
|
|
151
|
+
with the axes it changed, and an optional team or agent narrowing.
|
|
152
|
+
- **The pipeline blocks figures' `railRefusals` and `halt` are nullable.** They are null at a strategy scope. The
|
|
153
|
+
figures gain `unstampedBlocks`.
|
|
154
|
+
|
|
155
|
+
### Added tools
|
|
156
|
+
|
|
157
|
+
- **`get_agent_conditions`** and **`get_conditions_breakdown`** serve one row per (strategy, revision, condition key):
|
|
158
|
+
the checklist verdict against what each citing decision became.
|
|
159
|
+
- **`get_entry_discipline_breakdown`** serves Entry Discipline at a strategy cohort.
|
|
160
|
+
- **`get_pipeline_blocks_breakdown`** also accepts a strategy cohort. It then serves only the conditions stage.
|
|
161
|
+
|
|
41
162
|
## Contract history — v152 (LLM spend, pipeline blocks and the review's market slices are scoped sections)
|
|
42
163
|
|
|
43
164
|
**v152 is MAJOR: two published outputs lose fields, and one tool refuses a call it used to answer.** Six tools are
|
package/dist/index.d.ts
CHANGED
|
@@ -47,7 +47,7 @@ import { Client, StreamableHTTPClientTransport } from '@modelcontextprotocol/cli
|
|
|
47
47
|
* being asked. Move it for a change to THIS package — a proxy fix, a dependency bump, a docs
|
|
48
48
|
* correction. Never move it to track the server.
|
|
49
49
|
*/
|
|
50
|
-
export declare const PACKAGE_VERSION = "32.0.
|
|
50
|
+
export declare const PACKAGE_VERSION = "32.0.86";
|
|
51
51
|
export declare const DEFAULT_URL = "https://mcp.battlegrid.trade/mcp";
|
|
52
52
|
export interface EnvConfig {
|
|
53
53
|
apiKeys: string[];
|
package/dist/index.js
CHANGED
|
@@ -49,7 +49,7 @@ import { Client, SdkHttpError, StreamableHTTPClientTransport } from '@modelconte
|
|
|
49
49
|
* being asked. Move it for a change to THIS package — a proxy fix, a dependency bump, a docs
|
|
50
50
|
* correction. Never move it to track the server.
|
|
51
51
|
*/
|
|
52
|
-
export const PACKAGE_VERSION = '32.0.
|
|
52
|
+
export const PACKAGE_VERSION = '32.0.86';
|
|
53
53
|
export const DEFAULT_URL = 'https://mcp.battlegrid.trade/mcp';
|
|
54
54
|
const MAX_RETRIES = 3;
|
|
55
55
|
const RETRY_DELAYS_MS = [2000, 4000, 8000];
|