Standards

Utility standards, ingested as code-actionable rules.

A utility's overhead construction standards are a 700-page PDF today. After ingestion, they're a queryable rule set that fires on every pole the chat agent or QA/QC platform analyzes — with source-page citations and engineer-approved provenance for every rule.

Onboarding compressed
Weeks → hours

an in-house standards-codification effort takes months

Source-cited
100%

every approved rule carries doc + page citations

Topical hints needed
0

agent reads + categorizes from titles + content alone

Engineer-of-record sign-off
1

humans approve every rule before it fires

In active pilot. Pipeline validated end-to-end against multiple real-world standards corpora — from doc-numbered IOU compendiums to federal-style bulletins. Request early access to size an onboarding for your utility.

What sets it apart

Built for how standards actually publish, not how vendors wish they did.

Real overhead-construction manuals are doc-numbered compendiums with no page-number table of contents, federal pubs that express the same limit in completely different units, and binders too large to send to any single AI call. The pipeline handles all of these on day one.

Cold-start, no hand-holding

No "look in this document" hints. The agent reads the manual, fits findings to existing rule families, and proposes new ones when nothing fits — proven against utility standards it had never seen before.

Engineer review queue, not auto-fire

AI extracts; humans approve. Every rule lands in a staging review queue with the source PDF, page citations, and confidence flags before it can fire on a real pole. Bulk-approve the easy ones, walk-through the rest.

Built for how standards actually publish

Doc-numbered IOU compendiums with no page-number TOC. Federal bulletins that express limits as transverse force instead of angle. Multi-section binders too large to ingest in a single pass. The pipeline handles all three on day one.

Self-aware about its limits

On a 1,400-cell conductor table the agent will report "extracted a representative sample; full coverage needs the proposed force-based category to be approved first." Honest partial reporting beats silent dropping.

Source-traceable approvals

Every approved rule carries its source document number and page range. Recognition cards in the chat agent and QA/QC platform link back to the exact standards page that triggered the finding.

Rejection feedback loop

When an engineer rejects an extraction, the rejection reason feeds a tuning report that suggests where to tighten the detector. Misclassifications surface what to fix instead of staying hidden.

How it works

An eight-step pass on every utility.

From PDF drop to live validators. The pipeline runs once per utility at onboarding; the rules it produces fire on every pole forever after.

  1. 01

    Drop a PDF

    Drag the utility's overhead construction manual in. Multi-document compendiums, federal pubs, and large IOU sets all welcome.

  2. 02

    Probe + normalize

    Pre-flight pass that flags oversize or image-heavy PDFs for downsampling so per-pass costs stay bounded.

  3. 03

    Slice for context

    PDFs are sliced along section boundaries (table-of-contents-driven where possible) into chunks sized for a single ingest pass.

  4. 04

    Cold-start ingest

    No instructions, no hints. The agent extracts every rule it can recognize, fits each to a registered rule family, and proposes new ones when none fits.

  5. 05

    Stage to review

    Each rule lands as one validated record in a per-utility staging queue with audit-ready metadata: source doc, source pages, ingest run, confidence flags.

  6. 06

    Engineer review

    Bulk-approve, reject with a tuning reason, or open a single-rule detail with the source PDF inline. Edit-before-approve for small corrections.

  7. 07

    Rules go live

    Approved rules apply to the per-utility live config and are immediately consumable by the chat agent and QA/QC analyzers.

  8. 08

    Validators fire

    Every analyzer that runs against a pole now references the utility's own approved standards — not a generic baseline.

Rule families

Eight rule families, with room to grow.

Standards Console ships with four registered rule families — framing, conductor angle limits, pole attachment separation, loading area definitions — plus four additional rule families that get proposed during ingest and brought online after engineer review. Each utility's idiosyncratic schema slots in via per-utility translators that keep the underlying data shape stable.

conductor angle limits

Per-conductor maximum interior-angle limits keyed by conductor size, loading area, and framing configuration. Also accepts force-per-conductor + wind-span expressions.

pole attachment separation

Minimum separation distances between pole attachments, phases, and structural elements from dimensioned construction figures.

loading area definition

Per-utility loading-area buckets with their weather criteria — temperature, ice thickness, wind pressure, span limits — and any geographic triggers.

framing

Per-utility singleton: angle thresholds, slack-span max lengths, and explicitly disallowed constructions (e.g. raptor-zone restrictions).

guy assembly capacity

Per-assembly guy load ratings.

anchor holding power

Per-anchor holding capacity across expanding, screw, plate, service, rock, and swamp anchor types.

pole setting depth

Minimum pole setting depth by length × soil type. Augments ANSI standard depth with the utility's own published table.

soil classification

Multi-class soil classification table with derating guidance for utilities that publish their own soil schedule.

Framing detector + tuning loop

Geometric truth, then per-utility tuning.

A standards rule that says "triangular framing in raptor zones is unauthorized" is only useful if you can detect triangular framing. Standards Console pairs every rule set with a structural-framing detector that recognizes the geometric pattern from real pole geometry — not from string-matching on free-text labels.

When the detector misclassifies, engineer rejections feed a tuning report that suggests which threshold to tighten. Per-utility threshold overrides land in code, not in config, so they go through CI before they affect production findings.

Detector features

  • Geometric framing detector Recognizes vertical, horizontal, and triangular framing patterns from real pole geometry. Per-utility threshold overrides for the regions where defaults misfire.
  • Two-tier framing recognition Inline hint annotations during chat-side analysis surface framing context without auto-suppressing clearance findings. A persistent engineer-review queue audits independently of any single click.
  • Rejection-tuning corpus Engineer rejections summarize into candidate threshold tweaks for the detector. Tuning is data-driven, not guess-driven.
  • Restricted-zone enforcement When a structure pattern is detected on a pole inside a restricted zone (e.g. a raptor concentration area) and the utility's standards forbid it, the system emits a finding the reviewer can act on.

Onboarding model

A one-time, quoted onboarding fee.

Standards ingestion is a one-time per-utility engagement, scoped to the corpus and the engineering oversight a real onboarding requires. Quotes account for the size of the manual, the number of distinct rule families it carries, the engineer-review effort to land each one, and the white-glove tuning needed to get the validators firing cleanly on the customer's own poles.

Once a utility is onboarded, the rule set runs against every pole the chat agent or QA/QC platform analyzes — included in ongoing platform usage, not billed again per analysis.

Why a quoted fee, not a price list

  • Effort scales with the corpus A 200-page IOU framing section and a 1,200-page multi-section federal binder are very different engagements. We size each one.
  • Engineer review is part of the deliverable Every rule is human-approved before it fires. The fee covers the AI extraction and the engineering oversight that signs off on it.
  • A new rule family may need a new validator When a corpus surfaces a rule type the platform doesn't yet model, we wire the validator. That work is in scope.
  • In-house alternatives take months A standards-codification effort built in-house typically takes a small team months of cross-checking, transcription, and code work. The fee reflects the labor it replaces.

Add your utility's standards.

Tell us which manual you're starting from. We'll come back with a scoped onboarding plan and a quote.