trueppm-api 0.4.0__py3-none-any.whl

trueppm_api/apps/scheduling/serializers.py

@@ -0,0 +1,723 @@
+"""Serializers for the scheduling app admin API."""
+
+from __future__ import annotations
+
+from typing import Any
+
+from drf_spectacular.utils import extend_schema_field
+from rest_framework import serializers
+from trueppm_scheduler.derive import Quantity as CpmDerivationQuantity
+
+from trueppm_api.apps.scheduling.models import (
+    FailedTask,
+    MonteCarloRun,
+    ProjectForecastSnapshot,
+    VelocitySuggestion,
+)
+from trueppm_api.apps.scheduling.services import (
+    FORECAST_ALL_COMPLETE,
+    FORECAST_ESTIMATES_BELOW_PLAN_DURATION,
+    FORECAST_ESTIMATES_OFF_CRITICAL_PATH,
+    FORECAST_ESTIMATES_PENDING_APPROVAL,
+    FORECAST_NO_COMMITTED_TASKS,
+    FORECAST_NO_ESTIMATES,
+    FORECAST_NO_VELOCITY_HISTORY,
+)
+
+
+class MonteCarloWhatIfRequestSerializer(serializers.Serializer[dict[str, Any]]):
+    """Validate the non-mutating Monte Carlo what-if query (#993).
+
+    The what-if endpoint perturbs exactly one task's duration and recomputes the
+    forecast in memory without persisting anything, so this is a *request* (query
+    param) validator only — there is no model behind it. The caller must supply a
+    ``task_id`` and exactly one of ``duration_delta`` (signed day offset applied to
+    the task's current duration) or ``new_duration`` (an absolute day count the
+    task's duration is set to). Requiring exactly one keeps the perturbation
+    unambiguous — a request carrying both, or neither, cannot express a single
+    well-defined schedule and is rejected up front rather than resolved by a silent
+    precedence rule.
+    """
+
+    task_id = serializers.UUIDField()
+    duration_delta = serializers.IntegerField(required=False)
+    new_duration = serializers.IntegerField(required=False, min_value=0)
+    n_simulations = serializers.IntegerField(required=False, min_value=1)
+
+    def validate(self, attrs: dict[str, Any]) -> dict[str, Any]:
+        has_delta = "duration_delta" in attrs
+        has_absolute = "new_duration" in attrs
+        if has_delta == has_absolute:
+            raise serializers.ValidationError(
+                "Supply exactly one of 'duration_delta' or 'new_duration'."
+            )
+        return attrs
+
+
+_WORKING_TIME_DELTA_HELP = (
+    "Signed calendar days, measured in working time (#4204): each finish is read "
+    "with its edge of the day (the `*_at_day_start` readings), so a start-of-day "
+    "milestone finish on a Monday equals the end of the Friday before it and diffs "
+    "as 0 rather than 3. A reading persisted before #4204 is unknown: two equal days "
+    "diff as 0, otherwise it is read as the end of its day."
+)
+
+_AT_DAY_START_HELP = (
+    "Which edge of its shown day this finish is (#4204): true = the start of the "
+    "day (a start-of-day milestone finish), false = the end. Null on a run "
+    "persisted before #4204."
+)
+
+
+class MonteCarloDeltaSerializer(serializers.Serializer[dict[str, Any]]):
+    """Signed delta of a percentile finish, per field, in working time (#987/#993/#2483/#4204).
+
+    Positive means the field slipped later (worse). Shared shape for
+    ``run_monte_carlo``/``latest``'s ``delta_vs_cpm`` (each percentile vs the
+    deterministic CPM finish), the what-if endpoint's ``delta_vs_current``, and
+    the history endpoint's per-run ``delta`` (ADR-0108) — one derivation, one
+    schema, three call sites. ``None`` whenever either date it is measured
+    between is missing. Measured in working time from each date's edge-of-day
+    reading, still reported in calendar days.
+    """
+
+    p50 = serializers.IntegerField(
+        allow_null=True, help_text=f"Signed P50 delta; positive = later. {_WORKING_TIME_DELTA_HELP}"
+    )
+    p80 = serializers.IntegerField(
+        allow_null=True, help_text=f"Signed P80 delta; positive = later. {_WORKING_TIME_DELTA_HELP}"
+    )
+    p95 = serializers.IntegerField(
+        allow_null=True, help_text=f"Signed P95 delta; positive = later. {_WORKING_TIME_DELTA_HELP}"
+    )
+
+
+class MonteCarloHistogramBucketSerializer(serializers.Serializer[dict[str, Any]]):
+    """One ascending finish-date bucket of a Monte Carlo distribution."""
+
+    date = serializers.DateField()
+    count = serializers.IntegerField()
+
+
+class MonteCarloConfidencePointSerializer(serializers.Serializer[dict[str, Any]]):
+    """One point of the cumulative P(finish <= date) S-curve derived from the histogram."""
+
+    date = serializers.DateField()
+    pct = serializers.FloatField()
+
+
+class MonteCarloSensitivitySerializer(serializers.Serializer[dict[str, Any]]):
+    """One task's duration-sensitivity tornado entry (ADR-0140).
+
+    ``index`` is the absolute Spearman rank correlation between the task's
+    sampled duration and the project's finish, in [0, 1]. Tasks whose duration
+    cannot vary the finish are omitted from the list entirely rather than
+    reported as 0.
+    """
+
+    task_id = serializers.CharField()
+    index = serializers.FloatField()
+
+
+class MonteCarloDistributionFieldsSerializer(serializers.Serializer[dict[str, Any]]):
+    """The ``{histogram_buckets, confidence_curve, sensitivity}`` shape (#1231).
+
+    Mixed into :class:`MonteCarloForecastSerializer` as flat top-level fields
+    (the live-run and ``latest`` payloads), and used standalone — nested under
+    a ``distribution`` key — by :meth:`MonteCarloRunSerializer.get_distribution`
+    for the history endpoint's opt-in ``?expand=distribution``. One field
+    definition, two positions on the wire.
+    """
+
+    histogram_buckets = MonteCarloHistogramBucketSerializer(many=True)
+    confidence_curve = MonteCarloConfidencePointSerializer(many=True)
+    sensitivity = MonteCarloSensitivitySerializer(many=True)
+
+
+class MonteCarloForecastDiagnosticSerializer(serializers.Serializer[dict[str, Any]]):
+    """Why a Monte Carlo forecast carries -- or lacks -- an uncertainty band (#1340).
+
+    ``reason`` is populated only when ``deterministic`` is true (the forecast
+    collapsed to a single date); it is ``None`` whenever a real band exists.
+    """
+
+    deterministic = serializers.BooleanField()
+    reason = serializers.ChoiceField(
+        choices=[
+            FORECAST_NO_COMMITTED_TASKS,
+            FORECAST_ALL_COMPLETE,
+            FORECAST_ESTIMATES_BELOW_PLAN_DURATION,
+            FORECAST_ESTIMATES_OFF_CRITICAL_PATH,
+            FORECAST_ESTIMATES_PENDING_APPROVAL,
+            FORECAST_NO_VELOCITY_HISTORY,
+            FORECAST_NO_ESTIMATES,
+        ],
+        allow_null=True,
+    )
+    tasks_total = serializers.IntegerField()
+    tasks_with_variance = serializers.IntegerField()
+    tasks_pending_approval = serializers.IntegerField()
+    agile_tasks_without_velocity = serializers.IntegerField()
+    tasks_estimates_below_plan = serializers.IntegerField(
+        required=False,
+        help_text=(
+            "Tasks whose whole three-point range sits at or below their planned "
+            "duration, which the engine floors every sample at (#3765) — so the "
+            "range samples to a constant. Absent on a run persisted before #3765 "
+            "and replayed from history; clients read an absent value as 0, the same "
+            "way they already treat an absent `diagnostic` as 'no reason known'."
+        ),
+    )
+
+
+class RiskPremiumFieldsSerializer(serializers.Serializer[dict[str, Any]]):
+    """The flat ``risk_premium_*`` family shared by every forecast payload (ADR-0698).
+
+    Mixed into :class:`MonteCarloForecastSerializer` rather than nested — the
+    view spreads these keys directly onto the response dict (``**risk_premium_from_values(...)``),
+    and the schema mirrors that shape.
+    """
+
+    risk_premium_days = serializers.IntegerField(
+        allow_null=True,
+        help_text=f"P80 finish minus the deterministic CPM finish. {_WORKING_TIME_DELTA_HELP}",
+    )
+    risk_premium_ratio = serializers.FloatField(allow_null=True)
+    # Always null until #2299 (the calibration flywheel); declared nullable now so
+    # the schema does not have to change shape when a band is first populated.
+    risk_premium_band = serializers.CharField(allow_null=True)
+    risk_premium_as_of = serializers.DateTimeField(allow_null=True)
+    risk_premium_reason = serializers.CharField(allow_null=True)
+    risk_premium_state = serializers.ChoiceField(
+        choices=["not_run", "unmeasurable", "stale", "zero", "premium", "negative"]
+    )
+    risk_premium_cpm_finish = serializers.DateField(allow_null=True)
+    risk_premium_p80 = serializers.DateField(allow_null=True)
+
+
+class ForecastStalenessFieldsSerializer(serializers.Serializer[dict[str, Any]]):
+    """The flat forecast-staleness family shared by every forecast payload (#3140).
+
+    Mixed into :class:`MonteCarloForecastSerializer` for the same reason as
+    :class:`RiskPremiumFieldsSerializer` — the view spreads
+    ``**forecast_staleness_facts(...)`` directly onto the response dict.
+    """
+
+    forecast_staleness = serializers.ChoiceField(
+        choices=["current", "project_changed", "aged", "unknown"]
+    )
+    plan_version = serializers.IntegerField(allow_null=True)
+    plan_version_current = serializers.IntegerField(allow_null=True)
+
+
+class ScheduleDerivationContributionSerializer(serializers.Serializer[dict[str, Any]]):
+    """One candidate constraint the CPM engine weighed for a derived value (ADR-0218).
+
+    Mirrors :meth:`trueppm_scheduler.derive.DerivationContribution.to_dict`. Exactly
+    one contribution in a given derivation's ``contributions`` list carries
+    ``is_binding=True`` — the term the engine's ``max``/``min`` actually selected.
+    """
+
+    kind = serializers.CharField()
+    source_task_id = serializers.CharField(allow_null=True)
+    source_task_name = serializers.CharField(allow_null=True)
+    dep_type = serializers.CharField(allow_null=True)
+    lag_days = serializers.IntegerField(allow_null=True)
+    imposed_date = serializers.DateField(allow_null=True)
+    calendar_days_added = serializers.IntegerField(allow_null=True)
+    slack_days = serializers.IntegerField(allow_null=True)
+    is_binding = serializers.BooleanField()
+
+
+@extend_schema_field({"oneOf": [{"type": "string"}, {"type": "integer"}], "nullable": True})
+class ScheduleDerivationValueField(serializers.CharField):
+    """A CPM quantity's derived value (ADR-0218).
+
+    An ISO-8601 date string for a date quantity, or a plain integer working-day
+    count for ``total_float``/``free_float`` — never both on one response, since the
+    shape is fixed by the ``quantity`` query param. Declaration-only: this field is
+    never bound to real data (:class:`ScheduleDerivationView` returns a plain dict),
+    so the inherited ``CharField`` behavior is never exercised — only the
+    ``@extend_schema_field`` override, which states the true wire type.
+    """
+
+
+class ScheduleCpmDerivationSerializer(serializers.Serializer[dict[str, Any]]):
+    """CPM-quantity derivation payload — the *why* behind one computed value (ADR-0218).
+
+    Mirrors :meth:`trueppm_scheduler.derive.Derivation.to_dict`. Returned for a CPM
+    ``quantity`` (``early_start``/``early_finish``/``late_start``/``late_finish``/
+    ``total_float``/``free_float``/``scheduled_start``); see
+    :class:`ScheduleMonteCarloDerivationSerializer` for the percentile shape returned
+    for ``p50``/``p80``/``p95``.
+    """
+
+    task_id = serializers.CharField()
+    task_name = serializers.CharField()
+    quantity = serializers.ChoiceField(choices=[q.value for q in CpmDerivationQuantity])
+    value = ScheduleDerivationValueField(allow_null=True)
+    is_critical = serializers.BooleanField()
+    binding = ScheduleDerivationContributionSerializer(allow_null=True)
+    contributions = ScheduleDerivationContributionSerializer(many=True)
+
+
+# "pass" is a Python keyword and cannot be a class-body attribute name, but the wire
+# key is unsuffixed — Derivation.to_dict() emits it as "pass", not "pass_" — so the
+# field is added to the declared-fields map directly instead of following the
+# trailing-underscore convention into the schema, which would misdeclare the key.
+ScheduleCpmDerivationSerializer._declared_fields["pass"] = serializers.ChoiceField(
+    choices=["forward", "backward", "float"]
+)
+
+
+class ScheduleMonteCarloDerivationSerializer(ForecastStalenessFieldsSerializer):
+    """Monte Carlo percentile derivation payload (ADR-0218, #987, #3140).
+
+    Mirrors the dict built by ``ScheduleDerivationView._monte_carlo_derivation`` —
+    shared verbatim by its cache-hit and persisted-run-fallback branches. Returned for
+    a percentile ``quantity`` (``p50``/``p80``/``p95``); see
+    :class:`ScheduleCpmDerivationSerializer` for the CPM shape.
+    """
+
+    quantity = serializers.ChoiceField(choices=["p50", "p80", "p95"])
+    value = serializers.DateField(allow_null=True)
+    value_at_day_start = serializers.BooleanField(allow_null=True, help_text=_AT_DAY_START_HELP)
+    cpm_finish = serializers.DateField(allow_null=True)
+    cpm_finish_at_day_start = serializers.BooleanField(
+        allow_null=True, help_text=_AT_DAY_START_HELP
+    )
+    delta_vs_cpm_days = serializers.IntegerField(
+        allow_null=True,
+        help_text=f"The percentile minus the deterministic CPM finish. {_WORKING_TIME_DELTA_HELP}",
+    )
+    drivers = MonteCarloSensitivitySerializer(many=True)
+    runs = serializers.IntegerField()
+    last_run_at = serializers.DateTimeField()
+
+
+# Same reserved-keyword workaround as ScheduleCpmDerivationSerializer above — this
+# payload's "pass" is always the literal "monte_carlo".
+ScheduleMonteCarloDerivationSerializer._declared_fields["pass"] = serializers.ChoiceField(
+    choices=["monte_carlo"]
+)
+
+
+class MonteCarloForecastSerializer(
+    RiskPremiumFieldsSerializer,
+    ForecastStalenessFieldsSerializer,
+    MonteCarloDistributionFieldsSerializer,
+):
+    """Monte Carlo forecast payload shared by the live run and the ``latest`` read.
+
+    Two response-producing paths share this exact shape and are not identical on
+    the wire:
+
+    - The **live run** (``POST .../monte-carlo/``) and a **cache hit**
+      (``GET .../monte-carlo/latest/`` inside the 24h TTL) always carry
+      ``distribution`` (the full sorted per-run finish-date sample — the large
+      field) and never carry ``from_history``.
+    - The **persisted-history fallback** (``GET .../latest/`` once the cache
+      entry has expired, ADR-0175) always carries ``from_history: true`` and
+      never carries ``distribution`` — only the derived
+      ``histogram_buckets``/``confidence_curve``/``sensitivity`` (and
+      ``forecast_diagnostic``) survive past the TTL, and only for runs
+      persisted after #1231/#2483.
+
+    Both fields are declared ``required=False`` rather than modeled as two
+    named variants (``PolymorphicProxySerializer``): nothing else about the
+    shape differs between the two paths, so a discriminated union would only
+    duplicate every other field for no added precision.
+    """
+
+    project_id = serializers.CharField()
+    runs = serializers.IntegerField()
+    # Nullable: a project with no committed tasks yields no distribution to
+    # anchor the percentiles on (mirrors MonteCarloRun.p50/p80/p95).
+    p50 = serializers.DateField(allow_null=True)
+    p80 = serializers.DateField(allow_null=True)
+    p95 = serializers.DateField(allow_null=True)
+    p50_at_day_start = serializers.BooleanField(allow_null=True, help_text=_AT_DAY_START_HELP)
+    p80_at_day_start = serializers.BooleanField(allow_null=True, help_text=_AT_DAY_START_HELP)
+    p95_at_day_start = serializers.BooleanField(allow_null=True, help_text=_AT_DAY_START_HELP)
+    distribution = serializers.ListField(
+        child=serializers.DateField(),
+        required=False,
+        help_text=(
+            "Full sorted per-run finish-date sample. Present on a live run or a "
+            "cache hit; absent on the persisted-history fallback."
+        ),
+    )
+    cpm_finish = serializers.DateField(allow_null=True)
+    cpm_finish_at_day_start = serializers.BooleanField(
+        allow_null=True, help_text=_AT_DAY_START_HELP
+    )
+    delta_vs_cpm = MonteCarloDeltaSerializer()
+    forecast_diagnostic = MonteCarloForecastDiagnosticSerializer(allow_null=True)
+    last_run_at = serializers.DateTimeField()
+    status_date = serializers.DateField(allow_null=True)
+    from_history = serializers.BooleanField(
+        required=False,
+        help_text=(
+            "True only on the persisted-history fallback past the 24h cache TTL; "
+            "absent (never false) on a live run or cache hit."
+        ),
+    )
+
+
+class MonteCarloWhatIfAppliedSerializer(serializers.Serializer[dict[str, Any]]):
+    """The resolved perturbation the what-if endpoint applied to the target task (#993)."""
+
+    base_duration_days = serializers.IntegerField()
+    duration_delta_days = serializers.IntegerField()
+    new_duration_days = serializers.IntegerField()
+
+
+class MonteCarloWhatIfLegSerializer(serializers.Serializer[dict[str, Any]]):
+    """One forecast leg (``current`` or ``whatif``) of the what-if response (#993)."""
+
+    p50 = serializers.DateField()
+    p80 = serializers.DateField()
+    p95 = serializers.DateField()
+    p50_at_day_start = serializers.BooleanField(help_text=_AT_DAY_START_HELP)
+    p80_at_day_start = serializers.BooleanField(help_text=_AT_DAY_START_HELP)
+    p95_at_day_start = serializers.BooleanField(help_text=_AT_DAY_START_HELP)
+    cpm_finish = serializers.DateField(allow_null=True)
+    cpm_finish_at_day_start = serializers.BooleanField(
+        allow_null=True, help_text=_AT_DAY_START_HELP
+    )
+    critical_path = serializers.ListField(child=serializers.CharField())
+
+
+class MonteCarloWhatIfDeltaSerializer(serializers.Serializer[dict[str, Any]]):
+    """Signed shift of ``whatif`` vs ``current``, per field, in working time (#993/#4204).
+
+    Same ``p50``/``p80``/``p95`` shape as :class:`MonteCarloDeltaSerializer`,
+    plus ``cpm_finish`` — the what-if endpoint's own deterministic-pass delta,
+    distinct from ``delta_vs_cpm``'s MC-vs-CPM comparison elsewhere. All four are
+    measured in working time and reported in calendar days.
+    """
+
+    p50 = serializers.IntegerField(
+        allow_null=True, help_text=f"Signed P50 shift; positive = later. {_WORKING_TIME_DELTA_HELP}"
+    )
+    p80 = serializers.IntegerField(
+        allow_null=True, help_text=f"Signed P80 shift; positive = later. {_WORKING_TIME_DELTA_HELP}"
+    )
+    p95 = serializers.IntegerField(
+        allow_null=True, help_text=f"Signed P95 shift; positive = later. {_WORKING_TIME_DELTA_HELP}"
+    )
+    cpm_finish = serializers.IntegerField(allow_null=True, help_text=_WORKING_TIME_DELTA_HELP)
+
+
+class MonteCarloWhatIfResponseSerializer(serializers.Serializer[dict[str, Any]]):
+    """Non-mutating Monte Carlo what-if result (#993).
+
+    Perturbs exactly one committed task's duration and recomputes both the
+    deterministic CPM pass and a seeded Monte Carlo simulation in memory,
+    without persisting anything — see the view docstring for the
+    non-persistence guarantee.
+    """
+
+    task_id = serializers.CharField()
+    applied = MonteCarloWhatIfAppliedSerializer()
+    current = MonteCarloWhatIfLegSerializer()
+    whatif = MonteCarloWhatIfLegSerializer()
+    critical_path_changed = serializers.BooleanField()
+    delta_vs_current = MonteCarloWhatIfDeltaSerializer()
+    runs = serializers.IntegerField()
+    seed = serializers.IntegerField(
+        help_text="Fixed RNG seed shared by both runs so the delta isolates the perturbation."
+    )
+    cpm_status_date = serializers.DateField()
+    mc_status_date = serializers.DateField()
+
+
+class ProjectForecastSnapshotSerializer(serializers.ModelSerializer[ProjectForecastSnapshot]):
+    """Read-only serializer for a project-grain forecast snapshot (ADR-0154, #388).
+
+    Server-generated history; the endpoint is list-only, so every field is
+    read-only and there is no create/update path.
+
+    ``cpm_finish_at_day_start`` is published so a client diffing ``cpm_finish``
+    across snapshots can do it in working time (#4178): a start-of-day finish on
+    a Monday is the same finish as the end of the Friday before it.
+    """
+
+    cpm_finish_at_day_start = serializers.BooleanField(
+        read_only=True,
+        allow_null=True,
+        help_text=(
+            "True when cpm_finish is the START of that day: every task finishing on "
+            "it is a milestone shown at the start of its day. The start of a day is "
+            "the same point in working time as the end of the working day before "
+            "it, so compare finishes in working time, not as a date difference. "
+            "Null when unknown: no finish, or a snapshot captured before this field "
+            "existed."
+        ),
+    )
+
+    class Meta:
+        model = ProjectForecastSnapshot
+        fields = [
+            "id",
+            "captured_at",
+            "triggered_by",
+            "cpm_finish",
+            "cpm_finish_at_day_start",
+            "total_float_days",
+            "mc_p50_finish",
+            "mc_p80_finish",
+            "mc_p95_finish",
+            "mc_iterations",
+            "task_count",
+            "completed_task_count",
+        ]
+        read_only_fields = fields
+
+
+class FailedTaskSerializer(serializers.ModelSerializer[FailedTask]):
+    """Read-only serializer for the failed-task dead-letter queue.
+
+    Exposed by the admin API so operators can inspect and requeue or discard
+    tasks that exceeded their retry budget. The ``resolution_*`` fields surface the
+    operator-action audit (ADR-0210): who requeued/dropped the task, when, and the
+    drop note.
+    """
+
+    resolved_by_display = serializers.SerializerMethodField()
+
+    class Meta:
+        model = FailedTask
+        fields = [
+            "id",
+            "task_name",
+            "task_id",
+            "project_id",
+            "args",
+            "kwargs",
+            "exception_type",
+            "exception_message",
+            "traceback",
+            "failure_count",
+            "first_failed_at",
+            "last_failed_at",
+            "status",
+            "resolution_note",
+            "resolved_by_display",
+            "resolved_at",
+        ]
+        read_only_fields = fields
+
+    def get_resolved_by_display(self, obj: FailedTask) -> str | None:
+        """Display name of the operator who last acted, or None if unresolved."""
+        user: Any = obj.resolved_by
+        if user is None:
+            return None
+        full_name = user.get_full_name() if hasattr(user, "get_full_name") else ""
+        return full_name or user.get_username()
+
+
+class FailedTaskRequeueSerializer(serializers.Serializer[dict[str, Any]]):
+    """Validate the requeue-with-backoff request body (ADR-0210).
+
+    ``backoff_seconds`` is the operator-chosen delay applied as a Celery
+    ``countdown`` on the re-dispatched task. Bounded to a day so a fat-fingered
+    value cannot park a task for a year; ``0`` means dispatch immediately.
+    """
+
+    backoff_seconds = serializers.IntegerField(
+        required=False,
+        default=0,
+        min_value=0,
+        max_value=86_400,
+        help_text="Backoff before re-dispatch, in seconds (0-86400). 0 = immediate.",
+    )
+
+
+class FailedTaskDropSerializer(serializers.Serializer[dict[str, Any]]):
+    """Validate the drop-with-note request body (ADR-0210).
+
+    ``note`` is optional free-text audit context. Bounded and trimmed; stored as
+    data and rendered as text (never interpolated into task args), so it carries no
+    injection surface.
+    """
+
+    note = serializers.CharField(
+        required=False,
+        allow_blank=True,
+        default="",
+        max_length=1000,
+        trim_whitespace=True,
+        help_text="Optional audit note recorded with the drop.",
+    )
+
+
+class VelocitySuggestionSerializer(serializers.ModelSerializer[VelocitySuggestion]):
+    """Read serializer for a velocity-calibration suggestion (ADR-0065).
+
+    Surfaces the sprint that triggered the suggestion (name + id) so the Task
+    Detail Drawer can render "Suggested from Sprint 12 close" without a second
+    fetch.  The accept/dismiss audit fields are exposed so the drawer can also
+    render a quiet history row once a decision is made.
+    """
+
+    sprint_name = serializers.CharField(source="sprint.name", read_only=True)
+    sprint_id = serializers.UUIDField(source="sprint.id", read_only=True)
+    is_pending = serializers.BooleanField(read_only=True)
+    # Both team_velocity_per_day and suggested_duration are declared explicitly with
+    # allow_null because to_representation nulls them for readers below the velocity
+    # audience (ADR-0104 gate, #949/#1099) — the schema must advertise the suppressed
+    # shape for schema-driven clients (#997 contract class).
+    team_velocity_per_day = serializers.DecimalField(
+        max_digits=6,
+        decimal_places=3,
+        read_only=True,
+        allow_null=True,
+        help_text="Rolling 6-sprint average of completed_points / sprint_working_days.",
+    )
+    # suggested_duration is computed *from* team_velocity_per_day
+    # (round(story_points / velocity)); leaving it ungated lets a below-audience
+    # reader back into the team's pace via the calibration value, so it is gated by
+    # the same velocity check (#1099 — a new instance of the #949 leak class).
+    suggested_duration = serializers.IntegerField(
+        read_only=True,
+        allow_null=True,
+        help_text=(
+            "Velocity-calibrated duration in working days; null when the reader is "
+            "below the velocity audience."
+        ),
+    )
+
+    class Meta:
+        model = VelocitySuggestion
+        fields = [
+            "id",
+            "task",
+            "sprint_id",
+            "sprint_name",
+            "suggested_duration",
+            "team_velocity_per_day",
+            "flag_for_review",
+            "is_pending",
+            "created_at",
+            "accepted_at",
+            "accepted_by",
+            "dismissed_at",
+            "dismissed_by",
+        ]
+        read_only_fields = fields
+
+    # Velocity-derived fields stripped for a reader below the velocity audience.
+    # team_velocity_per_day is the raw rate; suggested_duration is computed from it
+    # (#1099) — both must fall to the same gate or the rate leaks via the suggestion.
+    _VELOCITY_GATED_FIELDS = ("team_velocity_per_day", "suggested_duration")
+
+    def to_representation(self, instance: VelocitySuggestion) -> dict[str, Any]:
+        data = super().to_representation(instance)
+        # ADR-0104 velocity gate (#949/#1099): these are the same point-based velocity
+        # number — and a value derived from it — that suppress_velocity_summary strips
+        # from /velocity/. A reader below the velocity audience is suppressed there, so
+        # they must not recover it from this calibration-suggestion surface.
+        request = self.context.get("request")
+        # Fail closed: a render with no request context can't establish the
+        # reader's tier, so suppress rather than leak (the only callers are HTTP
+        # responses, which always carry a request).
+        if request is None:
+            for field in self._VELOCITY_GATED_FIELDS:
+                data[field] = None
+            return data
+        # The verdict is per-project; cache it on the (reused) child serializer so
+        # a list render is not N+1 on the gate query.
+        project_id = instance.task.project_id
+        cache: dict[Any, bool] | None = getattr(self, "_velocity_gate_cache", None)
+        if cache is None:
+            cache = {}
+            self._velocity_gate_cache = cache
+        if project_id not in cache:
+            from trueppm_api.apps.projects.signal_privacy_services import can_read_signal
+
+            cache[project_id] = can_read_signal(request, project_id, "velocity")
+        if not cache[project_id]:
+            for field in self._VELOCITY_GATED_FIELDS:
+                data[field] = None
+        return data
+
+
+class MonteCarloRunSerializer(serializers.ModelSerializer[MonteCarloRun]):
+    """Read serializer for one project Monte Carlo run in the forecast history (ADR-0175).
+
+    Two pieces of context the view must supply:
+
+    - ``delta`` is computed-on-read (ADR-0108): the view attaches ``_delta`` to
+      each instance — a ``{"p50"|"p80"|"p95": signed-int-days|None}`` map of the
+      change versus the immediately-previous (older) run, or ``None`` on the
+      oldest/baseline row. Positive = the forecast slipped later (worse).
+    - ``triggered_by_name`` ("who ran it") is emitted **only** when the context
+      flag ``can_see_attribution`` is true (requester is Admin/Owner). For every
+      other member the field is ``None`` so forecast drift cannot be read as a
+      named-individual performance signal (VoC Morgan). The FK is never exposed
+      directly; only a display name, and only to admins.
+    """
+
+    delta = serializers.SerializerMethodField()
+    triggered_by_name = serializers.SerializerMethodField()
+    distribution = serializers.SerializerMethodField()
+
+    class Meta:
+        model = MonteCarloRun
+        fields = [
+            "id",
+            "taken_at",
+            "p50",
+            "p80",
+            "p95",
+            "cpm_finish",
+            "p50_at_day_start",
+            "p80_at_day_start",
+            "p95_at_day_start",
+            "cpm_finish_at_day_start",
+            "n_simulations",
+            "task_count",
+            "status_date",
+            "delta",
+            "triggered_by_name",
+            "distribution",
+        ]
+        read_only_fields = fields
+
+    @extend_schema_field(MonteCarloDeltaSerializer(allow_null=True))
+    def get_delta(self, obj: MonteCarloRun) -> dict[str, int | None] | None:
+        """Return the per-percentile day delta vs the previous run (view-attached)."""
+        return getattr(obj, "_delta", None)
+
+    @extend_schema_field(serializers.CharField(allow_null=True))
+    def get_triggered_by_name(self, obj: MonteCarloRun) -> str | None:
+        """Run-author display name — gated by the resolved attribution audience.
+
+        The view sets ``can_see_attribution`` per the workspace-effective
+        ``mc_history_attribution_audience`` (ADR-0144): ADMIN_OWNER → Admin/Owner,
+        SCHEDULER_PLUS → Scheduler+, NONE → nobody. None when the reader is below
+        the audience so forecast drift cannot become a named-individual signal.
+        """
+        if not self.context.get("can_see_attribution"):
+            return None
+        user: Any = obj.triggered_by
+        if user is None:
+            return None
+        full_name = user.get_full_name() if hasattr(user, "get_full_name") else ""
+        return full_name or user.get_username()
+
+    @extend_schema_field(MonteCarloDistributionFieldsSerializer(allow_null=True))
+    def get_distribution(self, obj: MonteCarloRun) -> dict[str, Any] | None:
+        """The persisted per-run distribution — only when explicitly expanded (#1231).
+
+        Returns the stored ``{histogram_buckets, confidence_curve, sensitivity}``
+        payload (snake_case, matching the ``/latest/`` response shape) ONLY when the
+        view sets ``context['expand_distribution']`` — the history *list* stays
+        lightweight by default and a single detail fetch opts in via
+        ``?expand=distribution``. ``None`` for legacy runs with no stored
+        distribution, or whenever the flag is unset.
+        """
+        if not self.context.get("expand_distribution"):
+            return None
+        return obj.distribution