How the DiffiWeb Swarm Works
30 specialist agents deliberate in parallel across 5 expert clusters. No LLM calls — pure TypeScript signal-based scoring with confidence-weighted consensus.
Why Two Scores?
Win Probability
On the card
Question answered:“How open is this market?”
High score = fewer competitors can realistically enter → your statistical odds improve.
Swarm Consensus
War Room
Question answered:“Should WE bid on this right now?”
High score = multiple specialist perspectives agree you should pursue this bid.
Consensus Formula
// For each cluster:
cluster_score = Σ(agent.score × agent.confidence) / Σ(agent.confidence)
// Final consensus (5 cluster scores):
consensus = procurement×0.30 + competition×0.25 + technical×0.20 + financial×0.15 + intelligence×0.10
// Agreement level (confidence-weighted std dev):
stdDev < 8 → High · stdDev < 16 → Moderate · else → Low
Verdict Scale
STRONG BID
72–92 %
Multiple clusters aligned. Pursue aggressively.
BID
52–71 %
Positive consensus. Prepare a competitive bid.
CONSIDER
34–51 %
Mixed signals. Assess your specific capacity.
AVOID
8–33 %
Strong negative signals. High risk exposure.
Domain Abstention
Certain agents are domain specialists — they only produce meaningful scores for tenders in their area. When a tender is outside their domain, they abstain by returning score=50, confidence=40.
Because the cluster average is confidence-weighted, an abstaining agent (conf=40) contributes roughly half the vote weight of a domain expert (conf=88–92). This prevents e.g. a Defence Tech specialist from tanking a Construction tender's Technical cluster.
Agents that abstain: Defence Procurement, Medical/Pharma, Energy/Power, Defence Tech, IT/Software
Trigger:category does not contain the agent's domain keywords
Missing Data Behaviour
competition_density = 0
Treated as data-not-scraped (not "zero competitors"). All density-based bonuses are guarded behind density > 0. No false near-monopoly signal.
Affected: MarketCompetition · MarketSaturation · GEMSpecialist · GlobalIntel · MarginCalc
emd = null, emd_ratio = null
Treated as data unavailable (not confirmed absent). Small uncertainty penalty (−4 to −6) replaces the full no-EMD penalty (−14/−15).
Affected: CPO · RegulatoryCompliance · CreditRisk · FinancialRisk
authority stats empty {}
POST handler now loads authorityStats + stateActivity from the same KV cache used by GET. Agents receive accurate data, not empty objects.
Affected: RepeatWinner · PaymentTerms · HistoricalIntel · AuthorityProfiler · MarketCompetition · StateMarketIntel
amount present, no authority benchmark
FinancialRisk uses amount tier as primary signal (₹20Cr+ → +10, ₹5–20Cr → +6, ₹1–5Cr → +3) instead of scoring at neutral base.
Affected: FinancialRisk
All 30 Agents — Scoring Reference
Every signal, every delta. Base score is the starting value before any signals are applied.
Procurement Cluster
Weight 30 % in final consensus
Chief Procurement Officer
EMD compliance, deadline pressure, procedural eligibility
Base
48
EMD ratio ≥ 3 %
Filters 85%+ casual bidders
EMD ratio 2–3 %
Strong financial barrier
EMD ratio 1–2 %
Moderate barrier
EMD ratio 0.4–1 %
Low barrier
EMD ratio < 0.4 %
Open field
No EMD data
Barrier unknown
Days left ≤ 5
Last-minute entries impossible
Days left 6–14
Tight window
Days left > 60
Prolonged window invites competition
Corrigendum
Fewer firms responded initially
Doc fee ≥ ₹5k
Eliminates casual bidders
Conservative base
Standard CPO adjustment (always)
conf=88 when both days_left + emd_ratio present; else 65
State Procurement Advisor
State market conditions, regional vendor density
Base
50
State active tenders ≤ 3
Minimal local competition
State active tenders 4–9
Manageable field
State active tenders ≥ 10
Crowded regional market
Corrigendum
Extension shows complexity
EMD present
Financial eligibility screen
Global tender
State advantage neutralised
Conservative base
Always applied
conf=80 when state count > 0; else 55
Defence Procurement Specialist
TOT/IDDM requirements, DRDO approvals, DPSU eligibility
Base
44
Defence/Military category
DPSU/OFB approval required
Aerospace category
CEMILAC/DRDO mandatory
Security category
Licensed vendors only
Amount ≥ ₹50 Cr
Only established primes qualify
Non-defence category
Returns neutral 50, conf 40
Conservative base
Always applied
conf=92 in domain; conf=40 when abstaining
GEM Portal Specialist
Online procurement dynamics, empanelment barriers, portal saturation
Base
52
competition_level = low
Few empanelled vendors
competition_level = medium
Manageable field
competition_level = high
Large empanelled pool
Density 1–3 (> 0)
Low portal saturation
Density 4–10 (> 0)
Moderate saturation
Density > 20 (> 0)
Crowded portal category
density = 0 or missing
No bonus — data unavailable
No doc fee
Zero barrier, casual submissions
conf=85 when competition_level present; else 65
Public Works Advisor
Civil contractor grades, PWD norms, infrastructure qualification
Base
48
Construction category
Class-A contractor required
Infrastructure category
Approved vendor list
Water/Sanitation
Specialised civil scope
Electrical category
Licensed contractor required
Outside PWD domain
Limited specialist insight
Amount ₹5–200 Cr
Typical PWD scale
Days left ≤ 14
Teams mobilised, late entries minimal
Conservative base
Always applied
conf=86 for Construction/Infra; else 55
Regulatory Compliance Officer
Procedural completeness, EMD compliance, documentation standards
Base
46
EMD + doc fee both present
Full compliance package
EMD only
Financial screening in place
Doc fee only
Mild procedural filter
Neither (data unavailable)
Compliance status uncertain
Corrigendum issued
Regulatory review conducted
Tender expired
Closed, cannot proceed
conf=82 always
Competition Cluster
Weight 25 % in final consensus
Market Competition Analyst
Competitor density, state saturation, market positioning
Base
52
Density = 1 (scraped)
Near-monopoly opportunity
Density 2–3
Very low saturation
Density 4–7
Manageable field
Density 8–15
Moderate saturation
Density 16–25
Crowded market
Density > 25
Highly saturated
Density = 0 (missing)
Guard: no false monopoly bonus
competition_level = low
Authority confirmed restricted pool
competition_level = medium
Moderate classification
competition_level = high
Authority confirmed high
State activity ≥ 20
Top procurement state
State activity 10–19
Active but manageable
State activity ≤ 3
Fewer local bidders
Global tender (not low)
International pool larger
conf=90 when density > 0 or competition_level present; else 58
State Market Intelligence
Regional vendor density, state procurement maturity
Base
52
State count = 0
Not in top markets, minimal vendor base
State count 1–3
Very low activity
State count 4–9
Emerging market
State count 10–19
Significant competition
State count ≥ 20
Top procurement state
Global tender
State intel less relevant
Corrigendum + low state
Even fewer prepared local bidders
conf=82 when state count > 0; else 55
Global Competitive Intelligence
International competition landscape, global bid pool dynamics
Base
50
Global + level=low
Restricted/specialised scope
Global + level=medium
Selective international pool
Global + level=high
International bidder flood
Global + level unknown
Assume large pool
Domestic density 1–3
Minimal concern
Domestic density 4–10
Manageable
Domestic density > 20
Heavily contested
Domestic density = 0
Guard: data unavailable
Conservative base
Always applied
conf=85 for global; 72 for domestic
SME Entry Analyst
Small-firm accessibility, financial barrier assessment, entry volume
Base
52
EMD ratio ≥ 3 %
Most SMEs cannot furnish — pool collapses
EMD ratio 1.5–3 %
Mid-size SMEs filtered
EMD ratio < 1.5 % + no emd
SME flood risk
Doc fee ≥ ₹5k
Excludes micro vendors
Amount < ₹50L
SME territory, high entry
Amount ₹50L–₹5 Cr
Attracts SMEs
Amount > ₹20 Cr
Excludes micro firms
conf=80 always
Repeat Winner Tracker
Historical award patterns, authority repeat procurement cycles
Base
48
Authority awards ≥ 10 + cat ≥ 3
Strong repeat pattern
Authority awards ≥ 5 + cat ≥ 2
Established buyer
Authority awards ≥ 3
Some pattern intelligence
Corrigendum + cat ≥ 2
Incumbent dropped, fresh chance
Authority not in DB
No pattern data
No authority data
Tracking unavailable
conf=84 when authority+stats available; else 42
Market Saturation Analyst
Category-density mapping, segment saturation scoring
Base
52
Density ≤ 2 (> 0)
Near-empty segment
Density 3–5
Low saturation
Density 6–12
Moderate
Density 13–20
Crowded
Density > 20
Heavily saturated
Density = 0 (missing)
Guard: no false bonus
competition_level = low
Confirms saturation analysis
competition_level = high
Corroborates concern
conf=88 when density > 0; else 52
Technical Cluster
Weight 20 % in final consensus
Technical Evaluation Specialist
Category niche, scope complexity, qualification barriers
Base
50
Oil/Gas, Defence, Medical, Aerospace
Only accredited firms qualify
IT, Infrastructure, Railways, Power
Technical certification required
Supply/Procurement (general)
Very low technical barrier
Other categories
Default
Amount ≥ ₹100 Cr
Large scope
Amount ₹10–100 Cr
Mid-tier
Amount < ₹50L
Micro tender
Authority category share ≥ 40 %
Proven alignment
Authority category share ≥ 20 %
Moderate alignment
conf=82 when category present; else 50
Civil & Infrastructure Expert
Construction grades, PWD norms, infrastructure qualification
Base
50
Construction category
Class-A mandatory
Infrastructure category
Approved panel list
Water/Sanitation
Specialised civil scope
Electrical works
Licensed contractor required
Outside civil domain
Limited insight
Amount ₹10–500 Cr
Typical civil project scale
Authority has civil history
Category alignment confirmed
conf=90 for Construction/Infra; else 52
Medical & Pharma Specialist
DPCO compliance, drug licensing, hospital procurement norms
Base
52
Medical/Pharma category
DPCO/drug licensing mandatory
Health-adjacent category
Regulated procedures
Amount ₹50L–₹50 Cr
Typical pharma range
Amount > ₹50 Cr
DPIIT-approved suppliers only
Non-medical category
Returns neutral 50, conf 40
conf=90 in domain; 40 when abstaining
Energy & Power Expert
OISD/PESO certifications, oil & gas vendor requirements
Base
52
Oil/Gas/Energy category
OISD/PESO mandatory
Electrical/Power category
CEIG approval required
Infrastructure overlap
Partial domain
Amount ≥ ₹20 Cr
Only major certified contractors
Outside energy domain
Returns neutral 50, conf 40
conf=90 for Oil/Energy; 72 for Infrastructure; 40 when abstaining
Defence Technology Evaluator
TOT/IDDM requirements, DRDO/DGQA approvals, defence prime standards
Base
44
Defence/Military category
TOT/IDDM: extremely restricted pool
Security category
Licensed, police-verified only
Amount ≥ ₹50 Cr
Tier-1 prime contractors only
Non-defence category
Returns neutral 50, conf 40
Conservative base
Maximum caution (always)
conf=92 in domain; 40 when abstaining
IT & Software Evaluator
CMMI/ISO certifications, system integration requirements
Base
52
IT/Software/Technology category
CMMI Level 3+ or ISO 27001
Consulting/Services category
Professional certification
Amount ≥ ₹10 Cr
Experienced integrator required
competition_level = low
Proprietary/restricted category
Outside IT domain
Returns neutral 50, conf 40
conf=86 in domain; 40 when abstaining
Financial Cluster
Weight 15 % in final consensus
Financial Risk Officer
Bid economics, amount fit, financial exposure
Base
42
Amount fits authority avg (0.4–2.5×)
Normal financial risk
Amount << authority avg (< 0.15×)
Lower financial risk
Amount >> authority avg (> 5×)
Unusually large
No authority benchmark → Amount ≥ ₹100 Cr
High-value, selective pool
No authority benchmark → Amount ₹20–100 Cr
Significant value
No authority benchmark → Amount ₹5–20 Cr
Mid-tier contract
No authority benchmark → Amount ₹1–5 Cr
Routine value
EMD ≥ ₹1 Cr
Financially stable bidders only
No EMD data
Barrier status unknown
Doc fee ≥ ₹5k
Financial qualifier
No amount
Pricing intelligence unavailable
Conservative base
Always applied
conf=78 when amount present; else 45
Working Capital Analyst
Liquidity requirements, capital lock-up, bid financing burden
Base
50
Amount ≥ ₹100 Cr
Substantial working capital required
Amount ₹10–100 Cr
Manageable for mid-size firms
Amount ₹1–10 Cr
Low capital lock-in
Amount < ₹1 Cr
Minimal capital requirement
No amount disclosed
Unquantifiable
EMD ≥ ₹50L
Ties up capital during bid period
Days left ≤ 14
Short exposure window
Conservative base
Always applied
conf=80 when amount present; else 48
Credit Risk Officer
Counterparty credit screening, payment reliability, financial barriers
Base
48
EMD ≥ ₹1 Cr
Creditworthy counterparties only
EMD present (< ₹1 Cr)
Participation threshold in place
EMD ratio present
Some credit filter active
No EMD data
Credit screening status uncertain
Doc fee ≥ ₹10k
Screens financially weak participants
Authority awards ≥ 10
Reliable payment track
Conservative base
Always applied
conf=76 always
Margin Calculator
Pricing power, margin potential, competition-driven cost pressure
Base
46
Density 1–5 (> 0)
Low competition = pricing power
Density 6–12 (> 0)
Moderate competition, margin possible
Density > 12 (> 0)
Margin compression expected
Density = 0 (missing)
Guard: no false pricing signal
Amount ≥ ₹20 Cr
Absolute margin potential significant
competition_level = low
Premium pricing achievable
competition_level = high
Margin compression likely
Conservative base
Always applied
conf=72 always
FX Risk Analyst
Currency exposure, hedging costs, international payment risk
Base
52
Domestic INR tender
Zero FX exposure
Global + stable currency country
Manageable hedging
Global + volatile currency
FX volatility risk
Global + competition=low
FX premium factored in without squeeze
Global + competition=high
Open competition + FX = margin pressure
conf=84 always
Payment Terms Analyst
Authority payment reliability, state budget cycles, receivables risk
Base
50
Authority awards ≥ 15
Highly reliable payment track
Authority awards 8–14
Adequate payment history
Authority awards 3–7
Limited but positive
Authority not in DB
Payment terms risk elevated
State activity ≥ 15
High-budget state, timely cycles
Conservative base
Always applied
conf=78 when authority+stats available; else 50
Intelligence Cluster
Weight 10 % in final consensus
Historical Intelligence Officer
Authority award history, repeat procurement patterns
Base
50
Authority awards ≥ 16
Highly active procurer
Authority awards 8–15
Sufficient for analysis
Authority awards 3–7
Limited but usable
Authority awards 1–2
Sparse, extrapolated
State share ≥ 50 % of awards
Strong regional alignment
State share 25–50 %
Moderate alignment
Category repeat ≥ 2 (awards ≥ 5)
Repeat buyer in category
Authority not in DB
No pattern intelligence
No authority
Analysis skipped
conf=85 when authority+stats available; else 35
Authority Profiler
Procurement profile depth, buyer intelligence coverage
Base
50
Authority awards ≥ 20
Deep procurement profile
Authority awards 12–19
Strong intelligence
Authority awards 6–11
Moderate depth
Authority awards 2–5
Thin but present
Authority diversified (≥ 3 categories)
Patterns well-established
Authority not in DB
Blind spot
No authority
Profiling not possible
conf=88 when authority+stats available; else 40
Seasonal Timing Analyst
Budget cycles, calendar-based competition patterns
Base
52
Closes in March
Year-end rush, flood of tenders
Closes in February
Pre-year-end push
Closes in January
Post-holiday lull
Closes Oct/Nov
Mid-year cycle, stable
Closes Jun/Jul
Fresh budget cycle
Closes Aug/Sep
Steady mid-year
conf=75 when closing_date present; else 45
Awardee Pattern Tracker
Incumbent vendor cycles, repeat procurement signals
Base
50
Category repeat ≥ 4
Vendor relationships established
Category repeat 2–3
Buyer-vendor relationship forming
Corrigendum + cat ≥ 2
Incumbent may be disqualified
Authority in DB, limited cat data
Partial data
Authority not in DB
No awardee database
conf=80 when authority+stats available; else 42
Regional Budget Analyst
State fiscal health, budget allocation cycles, payment risk
Base
52
State activity ≥ 15
Strong fiscal health signal
State activity 8–14
Adequate budget allocation
State activity 3–7
Moderate activity
State activity ≤ 2
Payment timelines uncertain
Amount > 0 and ≤ ₹50 Cr
Within typical state allocation
Global tender
Regional analysis less applicable
conf=72 when state count > 0; else 48
Repeat Procurement Tracker
Re-tender signals, procurement cycle intelligence
Base
50
Corrigendum + authority awards ≥ 5
Serious procurement, second-chance window
Corrigendum (any authority)
Re-tender = fresh opportunity
Category repeat ≥ 3
Predictable repeat cycle
Corrigendum on un-profiled authority
Opportunistic re-bid
Authority not in DB
Cycle intelligence unavailable
No authority
Tracking not possible
conf=80 when authority+stats available; else 42
Common Questions
Does the Swarm use AI / LLMs?
No. Every agent is a deterministic TypeScript function. Scores are instant, reproducible, and fully explainable — every delta on this page is the exact code.
Why does my 76 % Win Probability tender get 63 % from the Swarm?
They answer different questions. Win Probability measures market barriers (how many competitors can enter). Swarm asks whether you specifically should bid — weighing execution risk, timing, capital, and domain fit. A gap is expected and informative.
What is "confidence" on each agent?
Each agent self-reports domain confidence (0–100). Off-domain agents return conf=40–55 and abstain at score=50. Domain experts return conf=85–92 and drive their cluster average. This prevents e.g. DefenceTech from distorting a Construction tender.
Why are all scores clamped at 8–92 %?
Hard floor of 8 prevents division-by-zero edge cases. Hard ceiling of 92 prevents false certainty — no tender is ever a guaranteed win. The real-world range for informative bids is 34–78 %.
Click any tender card → Swarm button to start the war room