PanoramicData.IssueAssessment

Score the quality of issues and tickets from any system — Jira, GitHub, ServiceNow or your own — with composable, weighted, explainable assessments.

You give it a ticket as a JsonDocument and a set of assessment categories; it returns a score per assessment, per category and overall, each with a plain-English explanation of why. Seven ready-made categories cover the usual questions (is there a clear title, a real description, acceptance criteria, an owner, a due date, steps to reproduce?), and ten building blocks let you write your own.

Installation

dotnet add package PanoramicData.IssueAssessment

Targets .NET 10. No dependencies beyond the framework.

Quick start

using System.Text.Json;
using PanoramicData.IssueAssessment;
using PanoramicData.IssueAssessment.BaseCategories;
using PanoramicData.IssueAssessment.Configuration;
using PanoramicData.IssueAssessment.Core;

var assessor = new IssueAssessor(new AssessmentOptions
{
	Categories =
	[
		TitleQualityCategory.Create(),
		DescriptionQualityCategory.Create(),
		AssigneeOwnershipCategory.Create(),
		DueDateTimelineCategory.Create(),
	],
	ExecutionStrategy = CategoryExecutionStrategy.Parallel,
});

using var ticket = JsonDocument.Parse("""
	{
		"title": "Export to CSV drops rows with line breaks",
		"description": "When a description contains a line break, the CSV export silently skips that ticket.",
		"assignee": "alex",
		"assigneeState": "Assigned",
		"dueDate": "2026-11-20"
	}
	""");

// cancellationToken: your request's token, or CancellationToken.None.
IAssessmentResult result = await assessor.AssessAsync(ticket, cancellationToken);

Console.WriteLine($"Overall: {result.Aggregations.Weighted.WeightedAverage:P0}");
foreach (var category in result.Categories)
{
	Console.WriteLine($"{category.CategoryName}: {category.Score.NormalizedScore:P0}");
	foreach (var assessment in category.Assessments)
	{
		Console.WriteLine($"  {assessment.AssessmentName}: {assessment.ExplanationText}");
	}
}

Assessing tickets from Jira, GitHub or ServiceNow

The built-in categories read a small normalized shape:

Property	Meaning
title	One-line summary
description	Full description
acceptanceCriteria	What "done" means
reproductionSteps	Steps to reproduce, ideally as a numbered or bulleted list
assignee	Who owns it
assigneeState	Assigned, InProgress or Owned when someone has taken it
dueDate	Any date or date-time string (2026-11-20, 2026-11-20T17:00:00Z)

Map your platform's JSON to that shape before assessing. Since 1.1 the package includes mappers to NormalizedIssue, which also carries issue type, priority, status, project, components, labels, fix versions and an attachment count:

using PanoramicData.IssueAssessment.Mapping;

NormalizedIssue issue = JiraIssueMapper.FromRestJson(jiraIssueJson, new Uri("https://jira.example.com/"));
// or GitHubIssueMapper.FromRestJson(githubIssueJson), ServiceNowRecordMapper.FromTableJson(recordJson)
using JsonDocument ticket = issue.ToJsonDocument();

JiraIssueMapper reads REST API v2 (wiki text) and v3 (Atlassian Document Format, flattened to wiki text). If you already hold typed objects, such as a Jira client's Issue, set the NormalizedIssue properties directly.

The Examples project has complete, runnable mappers for the Jira REST API v2, the GitHub REST API and the ServiceNow Table API, including two details that matter in practice:

• GitHub has no fields for reproduction steps or acceptance criteria, so they are read from the headed sections that issue templates produce.
• Template comments and collapsed <details> blocks are stripped first. Without that, an issue whose template was never filled in scores as a long, detailed description.

Your own assessments can of course read any JSONPath in the raw ticket directly.

dotnet run --project PanoramicData.IssueAssessment.Examples

Rubrics

A rubric is a whole configuration written as JSON: categories, assessments, named conditions, score caps, grade bands and classifications. It suits teams that share one definition of a good ticket between several tools.

The built-in ticket-quality rubric

BuiltInRubrics.PanoramicDataTicketQuality is the rubric Panoramic Data uses for Jira and GitHub tickets. It scores components, description, testability, steps to reproduce, current versus expected behaviour, environment, attachments, documentation, certifications, UI consistency, business case and rollback plan for ordinary tickets, and a separate set of criteria for epics. It grades the result EXCELLENT, GOOD, ADEQUATE, NEEDS IMPROVEMENT or POOR, and caps a ticket below EXCELLENT when a mandatory element (steps to reproduce on a bug, a rollback plan on a change request) is missing.

using PanoramicData.IssueAssessment.Rubrics;

var assessor = new IssueAssessor(BuiltInRubrics.PanoramicDataTicketQuality);
AssessmentReport report = AssessmentReport.From(await assessor.AssessAsync(ticket, cancellationToken));

Console.WriteLine($"{report.Score}/100 {report.Grade}");
report.Findings.ForEach(Console.WriteLine);         // "✓ Steps to Reproduce: 3 numbered steps", ...
report.Recommendations.ForEach(Console.WriteLine);  // what to add to raise the score
Console.WriteLine(report.ToJson());

For Magic Suite (MS) tickets the rubric also scores release notes (ReleaseNotes, and EpicReleaseNotes on epics) and, for tickets that are not epics, the MUST/SHOULD/MAY classification of acceptance criteria and the Verification Evidence section (TicketContract). Both have weight 6 and a mandatory cap of 0.79, and neither applies to a ticket whose resolution is anything other than Fixed or Done. Documentation and Certifications accept a justified No documentation change: <reason> or No certification question: <reason> line in place of the item.

BuiltInRubrics.PanoramicDataTicketQualityJson returns its JSON, a good starting point for your own.

Writing a rubric

{
	"executionStrategy": "series",
	"conditions": {
		"isBug": { "type": "fieldEquals", "path": "$.issueType", "values": [ "Bug", "Defect" ] }
	},
	"categories": [ {
		"name": "Ticket Quality", "description": "Is the ticket ready to work on?", "assessmentExecutionStrategy": "series",
		"assessments": [ {
			"name": "Steps to reproduce", "description": "Bugs say how to reproduce them",
			"appliesWhen": { "type": "ref", "name": "isBug" },
			"recommendation": "Add numbered steps to reproduce.",
			"definition": {
				"type": "tiered",
				"measures": {
					"steps": { "type": "occurrenceCount", "source": { "paths": [ "$.description" ], "section": "reproduc" }, "pattern": "(?m)^\\s*\\d+\\." }
				},
				"tiers": [
					{ "when": { "type": "measure", "measure": "steps", "atLeast": 3 }, "score": 100, "finding": "✓ {steps} steps to reproduce" },
					{ "when": { "type": "sectionPresent", "source": { "paths": [ "$.description" ], "section": "reproduc" } }, "score": 50, "finding": "⚠ Steps are not a numbered list" },
					{ "score": 0, "finding": "✗ No steps to reproduce" }
				]
			}
		} ]
	} ],
	"scoreCaps": [ { "name": "Steps to reproduce", "cap": 0.79, "group": "mandatory",
		"when": { "type": "assessmentScore", "assessment": "Steps to reproduce", "lessThan": 100 } } ],
	"gradeBands": [ { "name": "GOOD", "minimumScore": 0.6 }, { "name": "POOR", "minimumScore": 0 } ]
}

AssessmentOptions options = AssessmentRubric.Load(File.ReadAllText("rubric.json"));

The building blocks:

• Tiered assessments ("type": "tiered"): tiers are tried in order and the first whose when holds gives the score (0–100 by default), the finding and an optional recommendation. The last tier has no condition.
• Measures count things: patternMatchCount, occurrenceCount, wordCount, alphanumericCount, length, itemCount, number, criteriaHeadingCount, sectionPanelCount (how many h2. sections of Jira wiki text are panelable, panelled, unpanelled or carry a panel title that repeats the heading) and criteriaClassification (how many criteria in a section there are, how many are not classified MUST, SHOULD or MAY, and how many are MUST).
• Conditions: all, any, not, ref (a named condition), measure, textMatches, fieldMatches, fieldEquals, fieldPresent, fieldsDiffer (two fields differ, ignoring case, surrounding space and an optional prefix), sectionPresent, placeholder (a section that only says "TBD") and, for caps and classifications, assessmentScore.
• Text sources join one or more fields and can narrow to a section by heading, in Jira wiki or Markdown.
• Placeholders in findings and recommendations: {measure}, {measure:text}, {$.path} and {$.path~pattern} (the first value matching the pattern).
• Applicability: appliesWhen on an assessment or a category. An assessment that does not apply is reported as NotApplicable (with its notApplicableExplanation) and is left out of the score entirely.
• Category weights, score caps (the lowest applicable cap wins), grade bands and classifications (first matching rule gives the value). Named conditions are reported back as flags.

Every rule is validated when IssueAssessor is constructed, as below. Definitions written in code can sit in a rubric as { "type": "named", "name": "..." } and be supplied to AssessmentRubric.Load(json, namedDefinitions).

The text tools the rubric uses are public too: IssueText.SectionBody, CriteriaHeadings, IsPlaceholder, WordCount and AlphanumericCount.

Concepts

Assessments

An assessment is one check with a score. Ten types are built in:

Type	Scores
PresenceAssessment	Whether a value exists and is not null or blank
RegexAssessment	Whether a string matches a pattern
StringLengthAssessment	Whether a string's length is in an inclusive range
WordCountAssessment	Whether a string's word count is in an inclusive range
NumericRangeAssessment	Whether a number (or numeric string) is in range
DateRangeAssessment	Whether a date is in range; exact format or flexible parsing
EnumAssessment	Whether a string is one of an allowed set
PatternCountAssessment	Whether a pattern occurs between a minimum and maximum number of times
JsonPathSelectorAssessment	Your async predicate over the selected value
CustomCallbackAssessment	Your async callback over the whole ticket

Values are selected with a JSONPath subset: $, $.a.b and $.a[0].b. Each assessment has a Weight (-100 to 100), an optional timeout, and an explanation, which you can replace with ExplanationTemplate.

Categories

A category groups assessments and runs them in series or in parallel. Categories themselves also run in series or in parallel (AssessmentOptions.ExecutionStrategy). In series, a category can carry an ISeriesGateCondition that decides, from its result, whether the remaining categories run at all — for example, skip expensive checks when the title is missing.

Built-in categories

Category	Checks
TitleQualityCategory	Title present, 5–150 characters, 3–30 words
DescriptionQualityCategory	Description present, 20–10,000 characters, 8–1,500 words
AcceptanceCriteriaCategory	Acceptance criteria present and at least 50 characters
AssigneeOwnershipCategory	Assignee present and in an ownership state
DueDateTimelineCategory	Due date present and between yesterday and two years from now
ReproducibilityCategory	Reproduction steps present and written as a list
AiSuitabilityCategory	Title present and at least 15 words of description

The thresholds were checked against 75 recent public issues from dotnet/runtime, microsoft/vscode and PowerShell/PowerShell: detailed bug reports score 0.9–1.0, and placeholder tickets ("Problems", "Doesn't work") score below 0.4. Each Create() returns an ordinary immutable record, so you can adjust any of it with with.

Scores

Every score carries the raw value, its possible range, and a normalized value in [0, 1]:

AssessmentScore score = assessment.Score; // RawScore, NormalizedScore, MinPossibleRaw, MaxPossibleRaw

A category's raw score is the weighted sum of its assessments, normalized against the weighted range; negative weights invert an assessment's contribution. result.Aggregations adds:

• AllAssessments and ByCategory[name]: count, minimum, maximum, average and standard deviation of normalized assessment scores;
• Percentiles: P50, P90, P95 and P99;
• Weighted.WeightedAverage: the average of the category scores.

Failures

result.Failures lists what could not be assessed normally: a value not found (JsonPathNotFound), a timeout, or an exception thrown by your code. A missing value still scores (it gets the failure score), so a ticket without a description simply scores low; filter Failures by Reason if you only want execution errors.

FailureHandling is CollectAll (the default: record and continue) or FailFast (stop the series at the first failure). A category can override it with CategoryFailureHandling.

Timeouts and cancellation

Timeouts can be set per assessment, per category and for the whole run (GlobalTimeout). A timeout is recorded as a TimeoutExceeded failure and the result is partial, so a slow custom check cannot block an assessment indefinitely. Cancelling the token you pass to AssessAsync throws OperationCanceledException, as usual.

Regular expressions run against ticket text, which is untrusted input, so every match has a 2-second timeout (RegexCache.MatchTimeout). A pattern that backtracks catastrophically fails that one assessment instead of hanging.

Chaining assessments with virtual extensions

In a series, an assessment can derive a value that later assessments score. Set AllowVirtualExtension and a unique VirtualOutputName; whatever the definition returns as VirtualExtensionOutput is stored under that name, and every later built-in assessment can read it as $.<name> (the ticket's own values always win).

public sealed class FirstSentence : IAssessmentDefinition
{
	public Task<AssessmentExecutionResult> ExecuteAsync(
		JsonDocument input,
		JsonDocument? virtualExtension,
		CancellationToken cancellationToken)
	{
		string description = input.RootElement.GetProperty("description").GetString() ?? "";
		string sentence = description.Split('.')[0];
		return Task.FromResult(new AssessmentExecutionResult(
			RawScore: 1,
			ExplanationText: "Extracted the first sentence.",
			FailureReason: null,
			InnerException: null,
			VirtualExtensionOutput: JsonDocument.Parse(JsonSerializer.Serialize(sentence))));
	}
}

AssessmentCategory summary = new()
{
	Name = "Summary",
	Description = "Is the opening sentence a useful summary?",
	AssessmentExecutionStrategy = AssessmentExecutionStrategy.Series,
	Assessments =
	[
		new Assessment
		{
			Name = "Extract",
			Description = "First sentence of the description",
			Definition = new FirstSentence(),
			AllowVirtualExtension = true,
			VirtualOutputName = "summary",
		},
		new Assessment
		{
			Name = "Summary length",
			Description = "The opening sentence has 5 to 40 words",
			Definition = new WordCountAssessment { JsonPath = "$.summary", MinimumWordCount = 5, MaximumWordCount = 40 },
		},
	],
};

Set CaptureVirtualDocument = true to get the final virtual document back in result.VirtualDocument. Virtual extensions need series execution; configuring one in a parallel category is rejected at construction.

Validation

Configuration is validated when IssueAssessor is constructed — invalid JSONPath or regex, inverted ranges, out-of-range weights, non-positive timeouts, duplicate category or output names — so a bad configuration fails at start-up with a message naming the category and assessment, never halfway through a batch.

Batches and caching

AssessManyAsync assesses an IAsyncEnumerable<JsonDocument> and streams the results.

Pass an IAssessmentCacheProvider to the constructor to reuse results. InMemoryAssessmentCache is thread-safe, bounded (least recently used entries are evicted first) and supports an optional time-to-live. Keys combine a SHA-256 of the ticket with a description of the configuration that does not depend on the current culture; callbacks, evaluators, gates and custom definitions are keyed by instance, so two different callbacks never share an entry.

var assessor = new IssueAssessor(options, new InMemoryAssessmentCache(new InMemoryAssessmentCacheOptions
{
	MaxEntries = 10_000,
	TimeToLive = TimeSpan.FromHours(1),
}));

Extending

• A new kind of check: implement IAssessmentDefinition (as in the virtual extension example), or use CustomCallbackAssessment for a one-off.
• Conditional execution: implement ISeriesGateCondition.
• A shared cache (Redis, a database): implement IAssessmentCacheProvider.

Performance

Measured over 3,000 assessments of 75 real GitHub issues with all seven built-in categories (Release build, one thread): median 0.22 ms, 95th percentile 0.49 ms, 99th percentile 0.95 ms per ticket.

Quality

• 815 unit tests; 100% line coverage of the library.
• The built-in rubric reproduces the scores, grades, findings and recommendations of the PowerShell engine it replaced on all 74 of that engine's test fixtures and on 320 real Jira tickets.
• Builds with zero warnings under TreatWarningsAsErrors and the latest-recommended analyzers.
• CodeQL and Codacy analysis on every push.

Documentation

• Design document: architecture and the reasoning behind it.
• Changelog
• Contributing and security policy

License

MIT. See LICENSE.