FR Catalogue — Full Detail

Click-through target for the FR IDs in the Product Requirements & Roadmap milestone tables. 84 requirements. The single source of truth is docs/portfolio/_data/requirements.yaml; this page only re-arranges it for reading. Edit the _data, then regenerate: python docs/portfolio/_build/generate_fr_catalogue.py.

M0 — Platform Foundation · Jul 2025

FR-APP-07

Web-based 3D inspection viewer (Cesium geospatial scene) ✦

Web-based 3D geospatial scene rendering inspection imagery and point clouds in spatial context — delivered at M0.

  • Capability: Application Surface
  • Theme: 3D viewer and overlays
  • Area: App
  • Quarter: Q3 2025
  • Type: Engineering
  • Priority: Critical
  • Status: Delivered
  • Milestone: M0

FR-APP-08

Operator dashboard — organization / campaign / anomaly overview

Operator overview of organizations, campaigns and anomalies — delivered at M0.

  • Capability: Application Surface
  • Theme: Dashboard and defect gallery
  • Area: App
  • Quarter: Q3 2025
  • Type: Engineering
  • Priority: High
  • Status: Delivered
  • Milestone: M0

FR-APP-09

Visual defect gallery — geo-tagged imagery, annotations & anomaly bounding boxes

Gallery of geo-tagged inspection imagery with annotations and anomaly bounding boxes — delivered at M0.

  • Capability: Application Surface
  • Theme: Dashboard and defect gallery
  • Area: App
  • Quarter: Q3 2025
  • Type: Engineering
  • Priority: High
  • Status: Delivered
  • Milestone: M0

FR-EVI-01

RGB inspection imagery ingestion — EXIF / GPS provenance & dataset scoping

RGB inspection imagery ingestion with EXIF/GPS provenance and per-dataset scoping — delivered at M0.

  • Capability: Evidence Intake
  • Theme: Sensor ingestion
  • Area: Platform
  • Quarter: Q3 2025
  • Type: Engineering
  • Priority: High
  • Status: Delivered
  • Milestone: M0

FR-SEC-04

Multi-tenant authentication & workspace access control ✦

Multi-tenant authentication with workspace/organization membership and role-based access control — delivered at M0.

  • Capability: Security & Compliance
  • Theme: Access control and tenant governance
  • Area: Platform
  • Quarter: Q3 2025
  • Type: Engineering
  • Priority: Critical
  • Status: Delivered
  • Milestone: M0

M1 — AI Foundation · Dec 2025

FR-AI-05

First-generation AI chat assistant — natural-language query over inspection data (early, not-yet-reliable prototype)

First natural-language chat assistant over inspection data, shipped as an early, not-yet-reliable prototype — delivered at M1.

  • Capability: Application Surface
  • Theme: Contextual data chat
  • Area: AI Assistant
  • Quarter: Q4 2025
  • Type: Engineering
  • Priority: High
  • Status: Delivered
  • Milestone: M1

FR-APP-10

Automated agentic task coordination — multi-agent orchestration (planner / executor, tool routing)

Automated multi-agent orchestration (planner/executor, tool routing) coordinating the AI systems behind chat and analysis — delivered at M1.

  • Capability: Application Surface
  • Theme: Agentic task coordination
  • Area: AI Assistant
  • Quarter: Q4 2025
  • Type: Engineering
  • Priority: High
  • Status: Delivered
  • Milestone: M1

FR-APP-12

Gas measurement visualization — sensor-reading heatmap overlay on the 3D scene

Sensor-reading gas heatmap overlay rendered on the 3D scene (colour-mapped concentration / temperature) — delivered at M1.

  • Capability: Application Surface
  • Theme: 3D viewer and overlays
  • Area: App
  • Quarter: Q4 2025
  • Type: Engineering
  • Priority: Medium
  • Status: Delivered
  • Milestone: M1

FR-EVI-02

Multimodal evidence pipeline foundation — thermal / OGI / gas ingest (prototype)

Prototype multimodal pipeline ingesting thermal, OGI and gas evidence, the foundation for the calibrated Q3 connectors — delivered at M1.

  • Capability: Evidence Intake
  • Theme: Sensor ingestion
  • Area: AI Assistant
  • Quarter: Q4 2025
  • Type: Engineering
  • Priority: High
  • Status: Delivered
  • Milestone: M1

M2 — App MVP · Mar 2026

FR-AI-06

Structured dataset-list responses — assistant answers with dataset collections render as DATASET_LIST cards on every certified engine

Assistant answers containing dataset collections are delivered as structured DATASET_LIST data and rendered as dataset cards, uniformly across certified AI engines — delivered at M2.

  • Capability: Application Surface
  • Theme: Structured assistant responses
  • Area: AI Assistant
  • Quarter: Q1 2026
  • Type: Engineering
  • Priority: High
  • Status: Delivered
  • Milestone: M2

FR-AI-07

Structured image-gallery responses — image-browsing answers carry image identities (id, dataset slug, filename, type) as IMAGE_GALLERY with click-through on every certified engine

Image-browsing answers are delivered as structured IMAGE_GALLERY data carrying image identities (id, dataset slug, filename, type) with click-through to the image viewer; presentation (thumbnails, signed URLs) is resolved by the frontend, uniformly across certified AI engines — delivered at M2.

  • Capability: Application Surface
  • Theme: Structured assistant responses
  • Area: AI Assistant
  • Quarter: Q1 2026
  • Type: Engineering
  • Priority: High
  • Status: Delivered
  • Milestone: M2

FR-APP-11

Provenance-cited chat answers — citations & tool-execution timeline

Chat answers carry citations and a tool-execution timeline tying responses back to source datasets — delivered at M2.

  • Capability: Application Surface
  • Theme: Contextual data chat
  • Area: App
  • Quarter: Q2 2026
  • Type: Engineering
  • Priority: Medium
  • Status: In Progress (Q2)
  • Milestone: M2

FR-OPS-01

Work-order / recommendation export — markdown + CSV

Export of work-order and recommendation packets as markdown and CSV — delivered at M2.

  • Capability: Operator Handoff
  • Theme: Reports and recommendation packets
  • Area: App
  • Quarter: Q2 2026
  • Type: Engineering
  • Priority: Medium
  • Status: Alpha (prototype)
  • Milestone: M2

M3 — AI Q2 Delivery · Jun 2026

FR-APP-02

Contextual data chat Ph.0 — single-turn NL query (classify → SQL execute → report) over workspace data ✦

First phase of conversational querying over inspection data — single-turn only; multi-turn is FR-APP-03.

  • Capability: Application Surface
  • Theme: Contextual data chat
  • Area: AI Assistant
  • Quarter: Q2 2026
  • Type: Engineering
  • Priority: Critical
  • Status: Delivered
  • Milestone: M3

FR-APP-13

Agent failure recovery — Supabase RLS/timeout, Gemini rate-limit/timeout, JWT expiry, and blob-storage retry handled without pipeline crash ✦

Supabase/Gemini/JWT/storage fault handling with graceful degradation, no pipeline crash.

  • Capability: Application Surface
  • Theme: Assistant reliability and safety
  • Area: AI Assistant
  • Quarter: Q2 2026
  • Type: Engineering
  • Priority: Critical
  • Status: Delivered
  • Milestone: M3

FR-APP-14

Contextual chat agent evaluation gate — classifier/executor/reporter accuracy benchmarks required before ship

Classifier/executor/reporter accuracy benchmarks required before ship.

  • Capability: Application Surface
  • Theme: Assistant reliability and safety
  • Area: AI Assistant
  • Quarter: Q2 2026
  • Type: Engineering
  • Priority: High
  • Status: Delivered
  • Milestone: M3

FR-APP-15

Persona-tailored workspace chat scoping — Data Explorer vs Integrity Engineer chat views

Data Explorer vs Integrity Engineer chat views, distinct from FR-SEC-04’s access control.

  • Capability: Application Surface
  • Theme: Contextual data chat
  • Area: App
  • Quarter: Q2 2026
  • Type: Engineering
  • Priority: High
  • Status: Alpha (prototype)
  • Milestone: M3

FR-APP-16

Contextual chat SQL-injection & malicious-input defense — DML/DDL blocking, injection-pattern rejection, workspace-scoped query firewall ✦

DML/DDL blocking, injection-pattern rejection, workspace-scoped query firewall.

  • Capability: Application Surface
  • Theme: Assistant reliability and safety
  • Area: AI Assistant
  • Quarter: Q2 2026
  • Type: Engineering
  • Priority: Critical
  • Status: Delivered
  • Milestone: M3

M4 — Persistent Sensing · Q3 2026

FR-AI-08

Server-side dataset-scope enforcement — scoped assistant queries return only in-scope entities, uniformly across engines

Assistant queries carrying a workspace dataset scope return only in-scope entity identifiers in structured answers (dataset rows, image identities), enforced platform-side at the gateway so the guarantee holds uniformly across all certified AI engines regardless of engine behavior.

  • Capability: Application Surface
  • Theme: Assistant reliability and safety
  • Area: AI Assistant
  • Quarter: Q3 2026
  • Type: Engineering
  • Priority: High
  • Status: Q3 Target
  • Milestone: M4

FR-APP-03

Contextual data chat Ph.1

Context-aware chips, workspace scoping, and multi-turn session state.

  • Capability: Application Surface
  • Theme: Contextual data chat
  • Area: AI Assistant
  • Quarter: Q3 2026
  • Type: Engineering
  • Priority: High
  • Status: Q3 Target
  • Milestone: M4

FR-APP-05

Interactive overlays

User-driven asset tags, annotations, and defect boxes mapped in the 3D model.

  • Capability: Application Surface
  • Theme: 3D viewer and overlays
  • Area: App
  • Quarter: Q3 2026
  • Type: Engineering
  • Priority: Medium
  • Status: Q3 Target
  • Milestone: M4

FR-APP-06

Evidence-bound reports generated from a worklist selection — the write-up cites the findings and notes it was built from

An engineer selects findings and notes on the campaign worklist and generates a PDF report that cites that evidence rather than restating it; reports are private by default and listed per campaign or asset.

  • Capability: Operator Handoff
  • Theme: Reports and recommendation packets
  • Area: AI Assistant
  • Quarter: Q3 2026
  • Type: Engineering
  • Priority: Medium
  • Status: Alpha (prototype)
  • Milestone: M4

FR-APP-17

Asset-scoped chat focus — set an engineering asset tag (e.g. AB-106) as chat scope; queries ground on the resolved asset record, its findings, and its geo-vicinity

Set an engineering asset tag (e.g. AB-106) as chat scope; queries ground on the resolved asset record, its findings, and its geo-vicinity.

  • Capability: Application Surface
  • Theme: Contextual data chat
  • Area: AI Assistant
  • Quarter: Q3 2026
  • Type: Engineering
  • Priority: Medium
  • Status: Q3 Target
  • Milestone: M4

FR-APP-22

Agent-operable integrity loop — the finding lifecycle (raise, correct, decide, curate) and actions exposed as tools that act as the signed-in user under RLS, with write guards and no decisions taken on the user’s behalf

Every step of the inspection-evidence loop is reachable by an AI agent through the MCP surface as the signed-in user: reads and writes ride RLS, writes require explicit confirmation, new findings default to “possible”, and the agent is instructed never to decide a finding the user has not judged.

  • Capability: Application Surface
  • Theme: Agent-operable surface (MCP)
  • Area: AI Assistant
  • Quarter: Q3 2026
  • Type: Engineering
  • Priority: High
  • Status: Alpha (prototype)
  • Milestone: M4

FR-CAD-01

CAD geometry via open IFC4 (BIM STEP) standard

Open-standard (BIM STEP) extraction of geometry plus engineering metadata into the Kav AI schema — demonstrated at 100% coverage on one example process unit.

  • Capability: World Model
  • Theme: Engineering context (CAD and P&ID)
  • Area: Platform
  • Quarter: Q2 2026
  • Type: Engineering
  • Priority: High
  • Status: In Progress (Q2)
  • Milestone: M4

FR-CAD-06

DEXPI open-standard P&ID ingestion (equipment, nozzles, piping, connectivity)

Ingest logical P&IDs via the open DEXPI (XML) standard — equipment, nozzles, piping, and connectivity — without proprietary lock-in.

  • Capability: World Model
  • Theme: Engineering context (CAD and P&ID)
  • Area: Platform
  • Quarter: Q2 2026
  • Type: Engineering
  • Priority: High
  • Status: In Progress (Q2)
  • Milestone: M4

FR-CAD-07

Deterministic dual-tagging (legacy CAD ↔︎ operator / DEXPI tags) with asset cross-reference

Rule-based, auditable layer reconciling legacy CAD tags with operator / DEXPI tags and resolving both to the same asset record.

  • Capability: World Model
  • Theme: Asset identity and tag reconciliation
  • Area: Platform
  • Quarter: Q2 2026
  • Type: Engineering
  • Priority: High
  • Status: In Progress (Q2)
  • Milestone: M4

FR-CAD-09

Image-to-asset attribution by depicted tag — OCR reads the equipment tag off the imagery, the engineer confirms or rejects, the decision propagates across RGB/thermal pairs; framed (field-of-view) and nearby (distance) images are context, not attribution

An image belongs to an asset when the tag it depicts has been read (OCR) and an engineer has confirmed it; the decision propagates across the RGB/thermal pair. Below attribution sit two context tiers that are never evidence: framed — the asset lies inside the camera’s field of view (Tier 3 geometry, asset-aware-retrieval PR #4, in review) — and nearby — distance only.

  • Capability: World Model
  • Theme: Asset identity and tag reconciliation
  • Area: AI Assistant
  • Quarter: Q3 2026
  • Type: Engineering
  • Priority: High
  • Status: Alpha (prototype)
  • Milestone: M4

FR-DMR-01

Evidence-grounded finding — a claim that an asset exhibits a condition, grounded in exactly one of a condition class or an API 571 damage mechanism, citing the images that show it

General Industry wording: Evidence-grounded finding — a claim that an asset exhibits a condition, grounded in exactly one of a condition class or a catalogued failure mechanism, citing the images that show it

A finding is a claim about an asset, not a tagged image: the pair (asset, condition) grounded in one of a condition class or an API 571 mechanism (CHECK edm_one_grounding), with the photographs that show it cited as evidence. New findings start as “possible”; only an engineer moves them.

  • Capability: Integrity Analytical Chain
  • Theme: Damage mechanism review on inspection evidence
  • Area: Platform
  • Quarter: Q3 2026
  • Type: Engineering
  • Priority: High
  • Status: Alpha (prototype)
  • Milestone: M4

FR-DMR-02

Screening damage-mechanism review from the asset profile — credible API 571 mechanisms proposed as Tier-2 “possible” findings for an engineer to confirm or dismiss

General Industry wording: Screening failure-mechanism review from the asset profile — credible catalogued mechanisms proposed as Tier-2 “possible” findings for an engineer to confirm or dismiss

Given an asset’s material, service and process unit, the packaged API 571 catalogue proposes the credible mechanisms as Tier-2 possible findings with the screening rationale. A reasoning aid, not a finding source: the 2026-08-13 review deleted type-keyed screening output in favour of evidence-first findings.

  • Capability: Integrity Analytical Chain
  • Theme: Damage mechanism review on inspection evidence
  • Area: AI Assistant
  • Quarter: Q3 2026
  • Type: Engineering
  • Priority: Medium
  • Status: Alpha (prototype)
  • Milestone: M4

FR-DMR-03

Finding correction and evidence curation — re-ground a finding, replace its notes, add or remove cited images, with every edit logged

A finding can be re-grounded on another condition class or mechanism, its notes replaced, and its cited images added or removed; each edit is appended to a sibling log so the current row and its history never disagree.

  • Capability: Integrity Analytical Chain
  • Theme: Damage mechanism review on inspection evidence
  • Area: App
  • Quarter: Q3 2026
  • Type: Engineering
  • Priority: Medium
  • Status: Alpha (prototype)
  • Milestone: M4

FR-INT-03

IDMS bidirectional integration specification ✦

General Industry wording: IDMS / EAM bidirectional integration specification

Bidirectional integration specification with IDMS programs.

  • Capability: Operator Handoff
  • Theme: IDMS and work-management integration
  • Area: Platform
  • Quarter: Q3 2026
  • Type: Product
  • Priority: Critical
  • Status: Q3 Target
  • Milestone: M4

FR-OPS-02

Engineering decision on a finding — confirm / dismiss / reinstate with verbatim reasoning, attributed by name, append-only and superseding; dismissals reversible

The point at which a proposed finding becomes an engineering judgement: confirm, dismiss or reinstate, with the reasoning recorded verbatim against the decider, appended (never overwritten) with a supersedes link. Reinstate returns a finding to “possible”, not to confirmed.

  • Capability: Operator Handoff
  • Theme: Verification queue and campaign hand-off
  • Area: App
  • Quarter: Q3 2026
  • Type: Engineering
  • Priority: High
  • Status: Alpha (prototype)
  • Milestone: M4

FR-OPS-03

Campaign worklist — the findings and notes of one campaign as a single feed ordered by last change, with note kinds from a controlled vocabulary and note → finding promotion

One queue per campaign over findings and notes, ordered by what changed most recently, so an engineer can answer “what have I not looked at” without leaving it. Note kinds come from an organisation vocabulary; a note can be promoted to a finding, never the reverse.

  • Capability: Operator Handoff
  • Theme: Verification queue and campaign hand-off
  • Area: App
  • Quarter: Q3 2026
  • Type: Engineering
  • Priority: Medium
  • Status: Alpha (prototype)
  • Milestone: M4

FR-OPS-04

Actions that name their finding — work orders and observations recorded with a finding reference, status open → scheduled → complete, amendable

A work order or observation is recorded against the campaign with the finding it follows from, so the reason travels with the action; status moves open → scheduled → complete and the note can be amended or re-typed.

  • Capability: Operator Handoff
  • Theme: Reports and recommendation packets
  • Area: App
  • Quarter: Q3 2026
  • Type: Engineering
  • Priority: Medium
  • Status: Alpha (prototype)
  • Milestone: M4

FR-OPS-05

Campaign hand-off lifecycle — indexing → ready for review → in review → closed, with send-for-review, return-with-reason and a history timeline in both workspaces

The relay between the two role workspaces is a real state: Data Explorer sends a campaign for review, the Integrity Engineer can return it for more evidence with a required reason, and the campaign history is a timeline of who moved it and why. in_review and closed are reserved states.

  • Capability: Operator Handoff
  • Theme: Verification queue and campaign hand-off
  • Area: App
  • Quarter: Q3 2026
  • Type: Engineering
  • Priority: Medium
  • Status: Alpha (prototype)
  • Milestone: M4

FR-PRT-01

Partner-integrated delivery model — single procurement vehicle, partner-provided HITL seat

Single procurement vehicle with a partner-provided HITL seat.

  • Capability: Operator Handoff
  • Theme: Partner-integrated delivery
  • Area: Platform / Commercial
  • Quarter: Q3 2026
  • Type: Product
  • Priority: High
  • Status: Q3 Target
  • Milestone: M4

FR-PRT-02

Reference partnership — OI.Expert × Kav AI integrated proposal template

General Industry wording: Reference partnership — integrated proposal template (industry-appropriate engineering partner)

Reference integrated-partnership proposal template.

  • Capability: Commercial / GTM
  • Theme: Partner program
  • Area: Commercial
  • Quarter: Q3 2026
  • Type: Product
  • Priority: High
  • Status: Q3 Target
  • Milestone: M4

FR-SCN-02

Calibrated thermal ingestion

Ingest and process calibrated thermal imagery.

  • Capability: Evidence Intake
  • Theme: Sensor ingestion
  • Area: AI Assistant
  • Quarter: Q3 2026
  • Type: Engineering
  • Priority: High
  • Status: Q3 Target
  • Milestone: M4

FR-SCN-03

Field survey readings ingestion — gas concentrations with ambient temperature, humidity and pressure, linked to assets as nearby context by distance and survey pass, screened against limits

Structured ingestion of georeferenced survey readings (NO₂, HCl, Cl₂, H₂ with ambient temperature, humidity and pressure). Readings attach to an asset as nearby context by distance and survey pass — never as attribution — and become evidence only when an engineer cites them; a limit table classifies a reading. Ambient temperature is the thermal ΔT reference at Stage 4, humidity a mechanism-plausibility factor at Stage 3.

  • Capability: Evidence Intake
  • Theme: Sensor ingestion
  • Area: AI Assistant
  • Quarter: Q3 2026
  • Type: Engineering
  • Priority: High
  • Status: Q3 Target
  • Milestone: M4

FR-VIS-02

Geo-tagged assets & images in 3D

Anchor inspection images and assets to geo-tagged positions in the 3D model.

  • Capability: World Model
  • Theme: Spatial registration and the 3D scene
  • Area: App
  • Quarter: Q3 2026
  • Type: Engineering
  • Priority: High
  • Status: Q3 Target
  • Milestone: M4

M5 — Engineering Context & Enterprise · Q4 2026

FR-AI-01

Filter Skill calibration & FNR measurement ✦

Calibrate the Filter Skill and measure false-negative rate.

  • Capability: Evidence Confidence
  • Theme: Calibration and confidence gates
  • Area: AI Assistant
  • Quarter: Q4 2026
  • Type: Engineering
  • Priority: Critical
  • Status: Q4 Target
  • Milestone: M5

FR-AI-02

Confidence score calibration protocol

Protocol for calibrating model confidence scores.

  • Capability: Evidence Confidence
  • Theme: Calibration and confidence gates
  • Area: AI Assistant
  • Quarter: Q4 2026
  • Type: Engineering
  • Priority: High
  • Status: Q4 Target
  • Milestone: M5

FR-AI-03

Chain-level consistency gate (Stage 3.5)

Consistency gate across the integrity analytical chain.

  • Capability: Evidence Confidence
  • Theme: Calibration and confidence gates
  • Area: AI Assistant
  • Quarter: Q4 2026
  • Type: Engineering
  • Priority: High
  • Status: Q4 Target
  • Milestone: M5

FR-AI-04

Out-of-distribution (OOD) detector update cadence

Cadence for updating the out-of-distribution detector.

  • Capability: Evidence Confidence
  • Theme: Calibration and confidence gates
  • Area: AI Assistant
  • Quarter: Q4 2026
  • Type: Engineering
  • Priority: Medium
  • Status: Q4 Target
  • Milestone: M5

FR-AI-09

Structured image-analysis responses — surface-analysis answers carry UUID-correlated per-image findings as IMAGE_ANALYSIS_RESULT with viewer annotation overlays, on engines that declare the capability

Image-analysis answers (surface conditions today; thermal, OCR, and quality skills on the same contract later) are delivered as structured IMAGE_ANALYSIS_RESULT data — exactly one entry per requested image UUID with a typed per-image status and normalized-geometry findings, no storage references (CONTRACT-CDC-001 v2.4 §7.8) — and rendered as annotation overlays in the gallery image viewer. Certified by the surface-analysis capability row: skipped for engines that do not declare it, blocking for engines that do. Platform side (contract, Web-owned media resolution, kavai-image-skills runner, certification row, viewer overlays) delivered 2026-08-08; no engine declares the capability yet.

  • Capability: Application Surface
  • Theme: Structured assistant responses
  • Area: AI Assistant
  • Quarter: Q4 2026
  • Type: Engineering
  • Priority: Medium
  • Status: Q4 Target
  • Milestone: M5

FR-AI-10

On-demand image analysis — a viewer control requests analysis of a selected image, authorized per image, without an assistant deciding to call a tool

A permitted user can request analysis of a selected image directly from the viewer, without asking an assistant and without depending on a model choosing to call a tool. The request is authorized per image, executes the canonical image-skill runner, and returns findings correlated to the requested UUIDs.

  • Capability: Application Surface
  • Theme: On-demand and durable image analysis
  • Area: AI Assistant
  • Quarter: Q4 2026
  • Type: Engineering
  • Priority: Medium
  • Status: Q4 Target
  • Milestone: M5

FR-AI-11

Annotation suggestion review — accept, reject, or recategorize model proposals; acceptance is annotation QA, not integrity confirmation

Model-proposed annotations are presented as suggestions distinguishable from stored annotations, and a permitted reviewer can accept, reject, or change the category of each one, or accept an explicit set at once. Accepting an annotation is annotation quality assurance — it does not confirm an integrity finding, authorize field action, or satisfy the operational human-in-the-loop requirement.

  • Capability: Application Surface
  • Theme: On-demand and durable image analysis
  • Area: AI Assistant
  • Quarter: Q4 2026
  • Type: Engineering
  • Priority: Medium
  • Status: Q4 Target
  • Milestone: M5

FR-AI-12

Durable collection analysis — a bounded run that outlives the page, with eligibility preview, resumable progress, cancellation, and per-image failures

A permitted user can analyze a selected set of images as one run that outlives the page. Before it starts the run states how much work is eligible, already done, and to be retried; while it runs progress is readable after a reload; and the user can cancel remaining work. Per-image failures are reported individually rather than collapsed into an overall success.

  • Capability: Application Surface
  • Theme: On-demand and durable image analysis
  • Area: AI Assistant
  • Quarter: Q4 2026
  • Type: Engineering
  • Priority: Medium
  • Status: Q4 Target
  • Milestone: M5

FR-AI-13

Analysis provenance and review audit — every attempt records its producer; every review decision is appended, never overwritten

Every analysis attempt records what produced it — backend, model identity or weights digest, skill and vocabulary version, thresholds, timing, and run identity — sufficiently to reproduce and audit the result. Every review decision is appended, never overwritten, so that a reject later changed to an accept leaves both decisions, their actors, and their times in the record.

  • Capability: Evidence Confidence
  • Theme: Analysis provenance and quality gates
  • Area: AI Assistant
  • Quarter: Q4 2026
  • Type: Engineering
  • Priority: High
  • Status: Q4 Target
  • Milestone: M5

FR-AI-14

Detection quality gate — no backend or skill revision becomes a production default without meeting the agreed protocol on a frozen, independently annotated set

A skill revision or detection backend does not become a production default until it meets the agreed measurement protocol on a frozen, independently annotated evaluation set — stated unit of evaluation, class-match and IoU rules, confidence-threshold selection, per-class and aggregate reporting, and an adoption threshold. Review history from accepted and rejected suggestions is training evidence and is not an evaluation set — it observes only what the current model proposed, so it cannot measure missed defects.

  • Capability: Evidence Confidence
  • Theme: Analysis provenance and quality gates
  • Area: AI Assistant
  • Quarter: Q4 2026
  • Type: Product
  • Priority: High
  • Status: Q4 Target
  • Milestone: M5

FR-ANO-01

Cross-modal anomaly detection

Detect anomalies across thermal, OGI, and gas modalities.

  • Capability: Evidence Confidence
  • Theme: Detection models and reasoning
  • Area: AI Assistant
  • Quarter: Q4 2026
  • Type: Research
  • Priority: High
  • Status: Q4 Target (research-gated)
  • Milestone: M5

FR-ANO-02

Physical AI reasoning & remediation

Reason from a finding to API 571 damage mechanism and remediation.

  • Capability: Evidence Confidence
  • Theme: Detection models and reasoning
  • Area: AI Assistant
  • Quarter: Q4 2026*
  • Type: Research
  • Priority: High
  • Status: Q4 Target (research-gated)
  • Milestone: M5

FR-APP-04

Chat with 3D map

Sync the chat interface with the 3D viewport — location fly-to and camera control.

  • Capability: Application Surface
  • Theme: Contextual data chat
  • Area: App
  • Quarter: Q4 2026
  • Type: Engineering
  • Priority: High
  • Status: Q4 Target
  • Milestone: M5

FR-CAD-02

CAD version tracking and diff visualization

Track CAD versions and visualize differences.

  • Capability: World Model
  • Theme: Engineering context (CAD and P&ID)
  • Area: App
  • Quarter: Q4 2026
  • Type: Engineering
  • Priority: High
  • Status: Q4 Target
  • Milestone: M5

FR-CAD-03

As-built vs as-designed comparison

Compare the as-built model against as-designed CAD.

  • Capability: World Model
  • Theme: Engineering context (CAD and P&ID)
  • Area: App
  • Quarter: Q4 2026
  • Type: Engineering
  • Priority: High
  • Status: Q4 Target
  • Milestone: M5

FR-CAD-04

Engineering change notification

Notify on engineering changes to drawings.

  • Capability: World Model
  • Theme: Engineering context (CAD and P&ID)
  • Area: Platform
  • Quarter: Q4 2026
  • Type: Engineering
  • Priority: Medium
  • Status: Q4 Target
  • Milestone: M5

FR-CAD-08

Additional CAD formats (RVT / DGN) beyond IFC4

Support additional CAD import formats (Revit, Bentley DGN) alongside the IFC4 pathway.

  • Capability: World Model
  • Theme: Engineering context (CAD and P&ID)
  • Area: Platform
  • Quarter: Q4 2026
  • Type: Engineering
  • Priority: Medium
  • Status: Q4 Target
  • Milestone: M5

FR-INT-01

OPC UA SCADA connector

General Industry wording: OPC UA control-system / historian connector

Read-only OPC UA connector to SCADA systems and historians.

  • Capability: Evidence Intake
  • Theme: Process data (read-only SCADA)
  • Area: AI Assistant
  • Quarter: Q4 2026
  • Type: Engineering
  • Priority: High
  • Status: Q4 Target
  • Milestone: M5

FR-INT-02

P&ID database (SQL) connector — direct read (distinct from DEXPI ingestion, FR-CAD-06)

Direct read of P&ID tag data via SQL for line and instrument linkage — distinct from DEXPI open-standard ingestion (FR-CAD-06).

  • Capability: World Model
  • Theme: Engineering context (CAD and P&ID)
  • Area: App
  • Quarter: Q4 2026
  • Type: Engineering
  • Priority: Medium
  • Status: Q4 Target
  • Milestone: M5

FR-INT-04

SAP PM certified connector

Certified connector to SAP Plant Maintenance.

  • Capability: Operator Handoff
  • Theme: IDMS and work-management integration
  • Area: Platform
  • Quarter: Q4 2026
  • Type: Engineering
  • Priority: High
  • Status: Q4 Target
  • Milestone: M5

FR-MDA-01

Solomon Associates benchmarking

General Industry wording: Industry RAM benchmarking integration (Solomon for hydrocarbons; equivalents for power, metals, pulp & paper, chemicals)

Benchmark corrosion rates and life estimates against Solomon data.

  • Capability: Integrity Analytical Chain
  • Theme: Benchmarking
  • Area: AI Assistant
  • Quarter: Q4 2026
  • Type: Engineering
  • Priority: High
  • Status: Q4 Target
  • Milestone: M5

FR-MDA-02

Synthetic data generation

Generate synthetic OGI data via physics-based plume simulation.

  • Capability: Evidence Confidence
  • Theme: Detection models and reasoning
  • Area: AI Assistant
  • Quarter: Q4 2026*
  • Type: Research
  • Priority: High
  • Status: Q4 Target (research-gated)
  • Milestone: M5

FR-RBI-01

API 581 inspection interval calculation

General Industry wording: RBI inspection interval calculation (API 581 in hydrocarbons; ISO 31000-aligned methodologies in other sectors)

Calculate API 581 inspection intervals from risk scores.

  • Capability: Integrity Analytical Chain
  • Theme: API 581 risk and inspection intervals
  • Area: AI Assistant
  • Quarter: Q4 2026
  • Type: Engineering
  • Priority: High
  • Status: Q4 Target
  • Milestone: M5

FR-RBI-02

Equipment class boundary of automation

Define which equipment classes are in scope for automated RBI.

  • Capability: Integrity Analytical Chain
  • Theme: API 581 risk and inspection intervals
  • Area: Platform
  • Quarter: Q4 2026
  • Type: Product
  • Priority: High
  • Status: Q4 Target
  • Milestone: M5

FR-SCN-01

OGI sensor ingestion

Ingest and process optical gas imaging (OGI) data from the campaign.

  • Capability: Evidence Intake
  • Theme: Sensor ingestion
  • Area: AI Assistant
  • Quarter: Q4 2026
  • Type: Engineering
  • Priority: High
  • Status: Q4 Target
  • Milestone: M5

FR-SEC-01

SOC 2 Type II certification ✦

Achieve SOC 2 Type II certification.

  • Capability: Security & Compliance
  • Theme: Certification and compliance management
  • Area: Platform
  • Quarter: Q4 2026
  • Type: Engineering
  • Priority: Critical
  • Status: Q4 Target
  • Milestone: M5

FR-SEC-02

Customer cloud tenant deployment

Deploy into a customer-managed cloud tenant.

  • Capability: Deployment Profile
  • Theme: Customer cloud tenant
  • Area: Platform
  • Quarter: Q4 2026
  • Type: Engineering
  • Priority: High
  • Status: Q4 Target
  • Milestone: M5

FR-SEC-03

Compliance management

Compliance management and reporting (e.g. EPA Quad-O).

  • Capability: Security & Compliance
  • Theme: Certification and compliance management
  • Area: App
  • Quarter: Q4 2026
  • Type: Engineering
  • Priority: High
  • Status: Q4 Target
  • Milestone: M5

FR-SEC-05

Tenant governance of analysis data — suggestions, decisions, and derived datasets stay tenant-scoped, deletable, and out of shared training without written authorization

Model suggestions, review decisions, and any dataset, embedding, checkpoint, or evaluation export derived from customer imagery are tenant-scoped by default, covered by the export and deletion workflows, and excluded from cross-tenant or shared training without explicit written authorization. Derived artefacts remain traceable to their source tenant so that a deletion obligation can be assessed rather than assumed.

  • Capability: Security & Compliance
  • Theme: Access control and tenant governance
  • Area: Platform
  • Quarter: Q4 2026
  • Type: Engineering
  • Priority: High
  • Status: Q4 Target
  • Milestone: M5

FR-VIS-01

3D CAD model overlay

Overlay the 3D CAD model onto the facility world model for as-built spatial context.

  • Capability: World Model
  • Theme: Spatial registration and the 3D scene
  • Area: App
  • Quarter: Q4 2026
  • Type: Engineering
  • Priority: Medium
  • Status: Q4 Target
  • Milestone: M5

FR-XSC-01

Cross-source correlation engine — promoted to named primitive (tag / match / score / surface) ✦

Tag, match within 2m, score, and surface multi-source findings.

  • Capability: Evidence Confidence
  • Theme: Cross-source correlation
  • Area: AI Assistant
  • Quarter: Q4 2026
  • Type: Engineering
  • Priority: Critical
  • Status: Q4 Target
  • Milestone: M5

FR-XSC-02

Multi-source confirmed TPR > 98% / FPR < 2% target reporting

Report TPR > 98% / FPR < 2% for multi-source confirmed findings.

  • Capability: Evidence Confidence
  • Theme: Cross-source correlation
  • Area: AI Assistant
  • Quarter: Q4 2026
  • Type: Engineering
  • Priority: High
  • Status: Q4 Target
  • Milestone: M5

Generated by docs/portfolio/_build/generate_fr_catalogue.py from docs/portfolio/_data/requirements.yaml. Regenerate after editing the _data.