Reads the bid. Holds the call.
A four-layer single-dispatch read of incoming bid packages against the operator's articulated thresholds. Information not judgment. The tool surfaces; the BD calls. Built by an operator's hand for the operator's desk — and tuned, line by line, against the corpus that the BD will actually read on a Tuesday morning.
Procore flattens the third-state distinction. Document Crunch flattens the third-state distinction. The CLC vertical-IP carries it canonically: applies · does not apply · genre out of scope · corpus silent. The schema is the edge. The Scan is what reads it.
Four positions. One stance. One read.
Bid Scan reads top-down — Position 1 catches the eye, Position 2 tells the story, Position 3 cites the corpus, Position 4 names the gaps. Each position does one job and hands off cleanly.
The Score Card
— Catches the eye before the coffee gets cold.
Composite verdict grid. Seventeen items, one icon each. Ten seconds from open to “yeah, look at item nine.”
The Executive Summary
— Tells the story without telling the BD what to do.
One to three paragraphs. Names the load-bearing items. Flags the anomalies as hypotheses. Aggregates the stacked-burden triggers that actually fired.
load_bearing_items[] · anomalies[] · stacked_burdens_triggered[]. The hand-off the BD reads first.The Per-Item Reports
— Shows the work. Cites the corpus. No black box.
Every extracted value carries a doc_id, section_id, page, and quoted text. The BD doesn’t take Scan’s word for it. The corpus is footnoted.
What To Watch For
— Names what isn’t there. Says what would unblock it.
Substrate-gap-propagation findings. The Cardinal-edge defensible-edge: cite the gap, then name the acquisition that closes it. Surfaces checklist gaps the BD might not have flagged.
Three voices. One schema.
Same data shape; three operator-facing stances. The BDM sit-down picks which one runs. All three score 14/14 against rubric v2 on Wagner.
Strict Extraction Specialist
Corpus-anchored, paranoid about silence. Lowest fabrication risk; engineering-leaning sentences. Best fit for BD culture that wants Scan as a forensic read of the corpus and nothing more.
— Sonnet 4.6 · 14/14 rubric · judge 2/2/2Structural-Inference Reader
Extracts where anchored; infers regime applicability where structure carries the signal. Provenance-tagged inferences. Best fit for BD culture that wants Scan to do the structural deduction the BD usually does in their head.
— Sonnet 4.6 · 14/14 rubric · judge 2/2/2BD-Voice Mirror
Same data, BD-friendly cadence, named operator actions (“pull from Building Connected,” “confirm SDI eligibility”). Pulls the lever to the surface. Best fit for an actionable read at the desk.
— Sonnet 4.6 · 14/14 rubric · judge 2/2/2The four reads on a single tool.
The BD's framing question on any Bid Scan output is one of these four. Bid Scan is built so each one has a clean answer.
A single-dispatch reader of incoming bid packages. Takes (contractor profile, bid corpus), returns a typed JSON conforming to scan-output-v0. Four output layers; one stance.
One model dispatch. Tool-use forced emission against the schema. Pre-registered rubric gates the output. LLM-judge scores narrative quality on three axes. Renders to self-contained HTML for the BD's desk.
For the BD reading the bid Tuesday morning. Calibrated to one reference operator at a time; profile JSON is the single per-client edit surface. Built from inside the work, not above it.
Industry-standard tools flatten the third-state distinction. Substrate gaps go unflagged; regime applicability gets binary. CLC Bid Scan carries the typed third-state canonically and names what acquiring missing substrate would unblock.
“Reads the bid. Holds the call.
Cites the corpus. Names the gap.
The BD's tool, on the BD's desk, before the bid due date.”
— CLC BID SCAN · v0 · pre-BDM-sit-down · 2026-05-06