submit_review

Science

The structured outcome of a review: a verdict, a QuakeML event type, a confidence, a solution-quality label, evidence that points at this session's ledger, and optional falsification checks and configurations tried. The server validates the record mechanically. It does not judge whether the seismology is right.

Evidence items must cite a tool_call_id from this session that succeeded. When a JSON Pointer path is given it must resolve in that call's stored payload. The value field is free text a reader compares; v1 does not check it against the pointed-at number. An accepted review with balanced evidence and verdict uncertain is a correct outcome, not a failure.

Profiles that require a review to close the session do so only through this tool. A rejected attempt counts against max_attempts. Structural rules (escalation text, proposed solution, rejection cause, rerun menu) are listed in the generated contract; they are checkability rules, not scientific ones.

Event types follow the QuakeML vocabulary (https://quake.ethz.ch/quakeml/): earthquake, quarry blast, not existing, and the other listed types.

Contract

Generated from the live registry (ToolSpec). Field names, types and defaults come from the input and payload models; they are not hand-typed.

Group	review
Class	—
Tool version	1.0.0
Highest declared tier	F
Timeout	5 s
Budget key	none
readOnlyHint	false
idempotentHint	false

Input

Field	Type	Required	Notes
verdict	string ∈ {confirm, reject, uncertain, reclassify, needs_more_data}	yes
event_type	string ∈ {earthquake, quarry blast, explosion, mining explosion, induced or triggered event, not existing, other event, not reported}	yes
rejection_cause	string ∈ {teleseism_leakage, noise_artefact, mis_association, duplicate, out_of_network, other} | null	no	default null
confidence	number	yes	≥ 0; ≤ 1
solution_quality	string ∈ {good, fair, poor, not_assessed}	yes
untrustworthy_params	array[string ∈ {origin_time, latitude, longitude, depth, magnitude}]	yes
proposed_solution	object (kind catalog | solution) | null	no	default null
recommended_action	string ∈ {publish, queue_for_human, discard, rerun}	yes
rerun_config	object | null	no	default null
rerun_config.picker	string ∈ {phasenet, eqtransformer} | null	no	default null
rerun_config.picker_threshold	number ∈ {0.1, 0.2, 0.3, 0.5} | null	no	default null
rerun_config.filter_band	string ∈ {none, 0.5-5, 1-20, 2-8, 2-15} | null	no	default null
rerun_config.channel	string ∈ {Z, horizontals, all} | null	no	default null
rerun_config.depth_mode	string ∈ {free, fixed_5, fixed_10, fixed_15} | null	no	default null
rerun_config.grid	string ∈ {shallow, deep} | null	no	default null
rerun_config.associator	string ∈ {gamma, pyocto} | null	no	default null
rerun_config.phase_set	string ∈ {all, p_only} | null	no	default null
rerun_config.arrival_model	string ∈ {nll_grid, taup_iasp91} | null	no	default null
rerun_config.time_axis	string ∈ {absolute, reduced_p} | null	no	default null
rerun_config.normalize	string ∈ {per_trace, common} | null	no	default null
evidence	array[object]	yes
evidence[].claim	string	yes
evidence[].tool_call_id	string	yes
evidence[].path	string | null	no	default null
evidence[].value	string	yes
evidence[].supports	string ∈ {confirm, reject, reclassify, uncertain}	yes
evidence[].weight	string ∈ {strong, moderate, weak}	yes
falsification	array[object]	yes
falsification[].check	string	yes
falsification[].tool_call_ids	array[string]	yes
falsification[].result	string	yes
falsification[].passed	boolean	yes
configurations_tried	array[object]	yes
configurations_tried[].tool_call_id	string	yes
configurations_tried[].hypothesis	string	yes
configurations_tried[].config	object	yes
configurations_tried[].result	string	yes
escalation_reason	string | null	no	default null
notes	string	yes

Payload

Field	Type	Required	Notes
review_id	integer	yes
accepted	boolean	yes
session_status	string ∈ {open, submitted, failed_to_submit}	yes
record_digest	string	yes

Example request

{
  "verdict": "reject",
  "event_type": "not existing",
  "rejection_cause": "teleseism_leakage",
  "confidence": 0.86,
  "solution_quality": "poor",
  "untrustworthy_params": [
    "depth",
    "magnitude"
  ],
  "proposed_solution": null,
  "recommended_action": "discard",
  "rerun_config": null,
  "evidence": [
    {
      "claim": "7 of 9 picks fit the us7000abcd P arrival within 15 s",
      "tool_call_id": "tc-0009",
      "path": "/best/matched_fraction",
      "value": "0.78",
      "supports": "reject",
      "weight": "strong"
    }
  ],
  "falsification": [
    {
      "check": "jackknife max shift below 5 km",
      "tool_call_ids": [
        "tc-0012"
      ],
      "result": "max_shift_km 6.2",
      "passed": false
    }
  ],
  "configurations_tried": [
    {
      "tool_call_id": "tc-0011",
      "hypothesis": "grid artefact",
      "config": {
        "grid": "shallow",
        "depth_mode": "free"
      },
      "result": "depth 8 km, rms 0.29"
    }
  ],
  "escalation_reason": null,
  "notes": "opinions belong here and nowhere else"
}

Example payload

{
  "review_id": 77,
  "accepted": true,
  "session_status": "submitted",
  "record_digest": "sha256:3456789abcdef0123456789abcdef0123456789abcdef0123456789abcdef012"
}

What this tool does not tell you

What this tool does not tell you: whether the review is right. It tells you whether it is checkable. An accepted review with balanced evidence and verdict uncertain is a correct outcome, not a failure.

Errors named for this tool

Code	When
E_SCHEMA	R1 — schema, enums, confidence in 0..1, unknown properties
E_EVIDENCE_INVALID	R2 empty evidence, or R3 unknown/failed/foreign tool_call_id or path
E_REVIEW_INVALID	R4–R8 structural rules (rule in detail)
E_PARAM_NOT_IN_MENU	R8 rerun_config value is off-menu
E_SESSION_CLOSED	R9 attempt budget spent, token expired, or the session is already closed