Scoring Methodology — Beltway Briefs

Overview

Opportunity Scout scores each ingested solicitation and pre-solicitation on a 0–93% Win Likelihood scale. This is not a statistical win probability — it's a relative priority rank that surfaces the best-fit, lowest-friction work for a mid-market commercial trade contractor (HVAC, electrical, plumbing, painting, roofing, facility maintenance) in Maryland, Virginia, and the D.C. metro.

Scoring runs entirely server-side in the Cloudflare Worker. All scores are computed fresh on each scan and are refined by Claude's read of the actual solicitation documents.

Sources & taxonomy

Every opportunity carries a source tag: state-md (Maryland eMMA), state-va (Virginia eVA), or state-dc (D.C.). When a NIGP or UNSPSC commodity code is present it is auto-detected and stands in for a profile-code match at Step 3. Virginia opportunities are pulled automatically each day from the eVA portal; Maryland and D.C. records are loaded from commodity-code alert emails or a fixed-column upload.

Win Likelihood = Base(25) + Eligibility + Value + Set-aside + Bond + Keyword − No-match penalty + Classification credit − Complexity penalty − Red-flag penalties → clamped 8%–93% Sort rank = (Win Likelihood² ÷ Complexity²) × Trade-tier weight

Step 1 — Exclusion Filter

Before scoring, any opportunity whose title matches an exclusion pattern is flagged "off-profile" and scored 5% (hidden from the ranked view by default). The list is deliberately narrow so building trade work always passes — it screens out only structurally out-of-scope categories.

Category	Example keywords
Heavy civil	road / highway / bridge / dam / dredging / earthwork / site grading / paving
Aviation / aerospace	aircraft, aviation, airframe, rotor, helicopter
Weapons / ordnance	weapon, ordnance, ammunition, missile, artillery
Nuclear / CBRN	nuclear reactor, radiological, biological agent, chemical agent
Non-construction services	catering / dining / meal preparation / agriculture / livestock
Electronics manufacturing	circuit board, PCB, semiconductor, cable assembly
Marine vessels	vessel, ship, submarine, destroyer
Materials-supply packages	materialman, supply only, furnish and deliver only, equipment procurement package, "no on-site installation" — buyer wants equipment made and delivered, not a contractor to install it

Note: "Roof replacement", "HVAC upgrade", "electrical service", "building renovation" etc. are not excluded — commercial building trades are the target.

Step 2 — Base Score & Adjustments

Base

Every non-excluded opportunity starts at 25%.

Eligibility (set-aside / preference)

Condition	Adjustment
Eligible now — SB / unrestricted / SBR (MD) / SWaM (VA) / SBE / CBE (DC) / MBE / DBE	+16%
Future eligible — WOSB / SDVOSB / 8(a) / HUBZone (cert gap)	+3%
SB set-aside or unrestricted channel (overlaps the row above — kept small)	+6%
Bonding requirement (bid / performance / payment bond) — screens out undercapitalized bidders	+4%

Contract value

Calibrated for mid-market trade contract sizes. A first pass — per-client bonding-capacity limits refine this once client profiles are added.

Contract value	Adjustment	Rationale
Unknown / $0	+4%	Unpriced
≤ $50K	+8%	Small job — thin estimator ROI for a mid-market firm
$50K – $250K	+16%	Mid-market sweet spot
$250K – $1M	+14%	Solid project size
$1M – $3M	+4%	More bonding, more competition
$3M – $6M	−8%	Large
> $6M	−18%	Favors regional / national primes

Complexity

Complexity (Low −0% / Medium −5% / High −14%) is decoupled from raw dollar value.

Category	Complexity rule
Janitorial, Grounds/Lawn, Painting — labor-scalable, low code/permit friction	Low up to $2M; Medium above — never High from size alone
HVAC, Electrical, Plumbing, Roofing, Facility Maintenance, unclassified	Value tiers: Low <$500K, Medium $500K–$2M, High >$2M
Any category with a Design-Build, Multi-Site, or Abatement red flag	+1 complexity level (capped at High)

Complexity also drives sort order — the ranked list is ordered by (WinLikelihood² ÷ Complexity²) × TradeTierWeight, so a High-complexity job ranks well below a Low-complexity one at the same win likelihood, and (see Step 3) a premium-trade job outranks a commodity-trade job that is otherwise equal. The tier weight affects sort order only — never the displayed win likelihood.

Step 3 — Trade Tier, Keyword Boost & No-Match Penalty

The opportunity title is scanned for trade keywords. Every matching trade is recorded (a job can be "Electrical and Plumbing Upgrades"); the highest-tier match applies its boost, labels the opportunity, and sets its tier — which also weights the sort order (Step 2). Two matched trades add +4%, three or more +6% (broader scope). A client counts as "in trade" if any matched trade is in their profile.

Tier	Trade	Trigger keywords (sample)	Boost	Sort weight
Premium higher ticket, license- & bond-gated, fewer qualified bidders	HVAC	hvac, heating, ventilation, air-condition, chiller, boiler, rooftop unit, RTU, AHU, BAS, DDC, mechanical	+22%	1.0
	Electrical	electrical, switchgear, lighting, generator, transformer, power distribution, conduit, wiring	+22%	1.0
	Plumbing	plumbing, piping, backflow, water heater, fixtures, sewer line, domestic water, grease trap	+22%	1.0
	Roofing	roof(ing), TPO, EPDM, built-up, membrane, shingle, roof replace/restore/recover/repair, waterproofing	+22%	1.0
Recurring	Facility Maintenance	facility/building maintenance, preventive maintenance, repair services, multi-trade, O&M / operations and maintenance	+16%	0.9
Commodity labor-driven, low barrier, incumbent-heavy, price-competitive	Painting	painting, repaint, coating, surface prep, epoxy floor, protective coating, abrasive blast	+12%	0.78
	Janitorial	janitorial, custodial, cleaning, housekeeping, sanitation	+12%	0.78
	Grounds/Lawn	lawn, landscape, grounds, mowing, turf, snow removal	+12%	0.78

No-match penalty: −12%
If no trade keyword matches the title, the opportunity takes a −12% penalty and a 0.6 sort weight. Vague titles are likely misclassified or out-of-profile — unknown scope defaults to skepticism, and stays out of the top of the brief.

Why "O&M" is a boost, not a flag: operations-and-maintenance language signals recurring, multi-trade facility work — core revenue for a facility-maintenance contractor. (In a services-firm scoring model it would be a red flag; here it is deliberately flipped.)

Classification credit

Records carry a UNSPSC or NIGP code, not a matchable profile-code table. An opportunity whose title matched a trade keyword gets a flat +6% classification credit plus +15 signal points — standing in for a direct commodity-code hit. Applies once, on a keyword match.

Step 4 — Red-Flag Deductions

Title patterns that signal structural disadvantage for a mid-market trade contractor. Total deduction is capped at −30%. Up to 2 flag labels show in the opportunity row.

Flag	Trigger keywords (sample)	Deduction
Design-Build	design-build	−12%
Abatement	asbestos, lead abatement, lead-based paint, hazmat abatement, ACM, mold remediation	−12%
Incumbent Signal	incumbent, recompete, follow-on, option exercise, continuation of services	−12%
PLA/Union	project labor agreement, PLA, union labor required	−8%
IDIQ/JOC Vehicle	IDIQ, MATOC, JOC, job order contract, indefinite-delivery, BPA	−8%
Multi-Site Scope	district-wide, campus-wide, enterprise-wide, multiple facilities, portfolio of buildings	−8%
Site Utility	water/sewer main, water distribution system, force main, lift/pump station, manhole rehab, CIPP, sanitary sewer main/collection, utility relocation, stormwater BMP — underground/heavy-civil work that reads as plumbing but isn't a building contractor's scope	−16%

Design-Build, Abatement, and Multi-Site also add one complexity level (Step 2).

Score Signal

Each score carries a Signal indicator (High / Med / Low) — how much data supports the score, not whether it's good or bad.

● High signal Strong trade-keyword match plus the classification credit. The score reflects real information about trade alignment.

● Med signal Partial signal — a keyword match is present but weak, or the title is otherwise ambiguous.

● Low signal No trade-keyword match. A conservative structural estimate — treat it as a flag for closer review, not a reliable likelihood.

Claude AI Analysis

On demand (click "Analyze ✦"), Claude reads the solicitation documents and returns the estimator's decision facts:

3-bullet scope of work — systems, equipment, quantities
Pre-bid walk-through — date, time, location, mandatory/optional (with an .ics calendar button)
Bid bond %, performance/payment bond %, minimum license tier
Q&A cutoff, bid due date/time, buyer contact
Contract type, evaluation basis, past-performance thresholds
Incumbent risk, win factors, red flags, overall fit

Analyses are cached 90 days and restored automatically after a re-scan. All analyzed opportunities are stored in the Analysis Archive.

Known Gaps & Limitations

What the score does not capture:

Market competition: no comprehensive public historical-award dataset for state/local trade work, so competition density is not scored. The scorer's competitorCount input is wired in for a future market-data source (e.g. eVA / Maryland spend data).
Bonding capacity vs. project size: the value curve is a global first pass; a per-client bonding ceiling would refine it materially.
Amendment history: heavily amended solicitations may signal incumbent negotiation or scope churn — not yet scored.
Prevailing wage / certified payroll: normal for public work but adds administrative load; captured in AI analysis, not scoring.
Geography: a specific non-local jobsite that requires travel/mobilization is not penalized unless the title flags multi-site scope.
Contract type (lump sum vs. unit price vs. T&M): visible only in the document — captured in AI analysis, not algorithmic scoring.
Virginia opportunities carry no value, code, or set-aside at scan time — the eVA feed is title + agency + dates only. Every VA opp scores Low complexity, so the pre-analysis VA ranking separates on trade tier, red flags, and bonding signal, not size. AI analysis (and operator review) does the sizing before anything reaches the brief — and when analysis extracts a dollar figure, it's written back to the record so the next scan can rank on size.