All insights

GeM & Procurement

Understanding GeM's Sanity Framework Flags

· 10 minute read

GeM watches for patterns that look like tinkering, splitting, collusion or nonsense prices. Treat flags as questions to answer. Do not role-play an unpublished algorithm.

Every marketplace that spends public money eventually grows a second nervous system. On GeM that system is talked about as sanity checks, integrity monitoring, flags on bids and catalogues. Sellers swap screenshots. Consultants sell certainty. Most of the certainty is invented.

This article stays conceptual on purpose. We will not publish a fake points table. We will not guess GeM's unpublished weights. We will describe the behaviours the platform and the GFR already say they dislike, and how an AI vendor or a procurement officer should behave so that a flag is answerable.

If GeM later publishes a detailed methodology, use that document. Until then, honesty is the only stable strategy. This is not legal advice.

What we can honestly say

GeM is a rules engine sitting on top of public procurement law. It already validates registrations, categories, brand mappings, bid packets and some commercial fields. It also lets humans raise incidents. The incident FAQs define bid tinkering, catalogue tinkering, and bypassing of system validation, and they call the last family Grave.

On top of hard validation, platforms of this size run anomaly detection: prices that make no sense against the catalogue history, quantities that look like a split, sellers who appear and disappear around one buyer, specifications that mention a brand after the brand field was supposed to be generic. Call that a sanity framework if you want. It is a set of questions, not a published credit score.

Flags you should expect, conceptually

Patterns that reasonably attract a question
PatternWhy a monitor would careHow to make it answerable
Catalogue price jumps or collapses overnightLooks like bait or bid-specific tailoringKeep a dated price log and a reason
Many small orders of the same article to one buyerLooks like splittingAsk for one bid; keep the refusal mail
Spec that only your SKU can passLooks like brand-lockingRewrite as a testable capability
Two bidders, same registered address or same rate card typoLooks like a related-party ringDo not bid as twins
On-prem claimed, hosted URL deliveredLooks like catalogue falsehoodDeliver what the listing said
Bypassing a greyed-out field with a workaroundThe FAQ already calls this GraveDo not

Why agent catalogues trip monitors

Novel software sits badly in categories built for printers. Sellers stretch a nearby category. They paste a brand into the specification. They list a hosted demo as if it were the air-gapped article. Each stretch is a sanity problem waiting for a human.

The fix is boring. Petition for a better category. Write specifications as tests. Keep the hosted demo and the on-prem SKU on different listings if they are different articles. Do not hide a model API in a footnote.

A flag is a conversation if you kept papers

Buyers create flags too. A specification that names a commercial product, a bid that is retendered with a narrower brand, a string of under-threshold carts, a PAC with no market scan: each of those is an integrity story waiting for a question. If you are the procurement officer, write the generic test first. If you are the DDO, add the carts before you pass the fourth. The monitor is not only a seller problem.

What to do when something is flagged

  1. Read the message. If it is an incident, switch to the incident playbook and the 7-day clock.
  2. If it is a bid query, answer on the bid portal with documents, not adjectives.
  3. Do not try a second workaround on the field that triggered the check.
  4. Tell your board the facts. A quiet flag that becomes a Grave incident is worse than an awkward week.
  5. Fix the catalogue for the next bid even if this one survives.

Objections from people who want a cheat sheet

Just tell us the formula. We do not have it. Neither do the people selling it in Telegram groups. Optimising a rumour is how you bypass validation.

Everyone tailors the catalogue to the bid. Everyone is not a defence in the incident FAQ. Catalogue tinkering is defined there in plain language.

Our consultant said flags are random. Some checks will feel random from the outside. Process still wants a factual reply. Randomness is not permission to ignore the mail.

Integrity monitoring, whatever its unpublished mechanics, will always care about rings. Two seller ids, one address. Two DSCs, one laptop. A bid in which the L2 firm's rate card has the L1 firm's typo. Founders in a hurry create these patterns with sister companies and friendly resellers. They think they are covering more categories. They are writing a question they will not enjoy answering.

If you need a reseller, document the relationship the way a CPSE vendor list would. Different control, different books, different bid decision. If you cannot say those sentences, bid once. A single honest id is easier to defend than a family of ids that move in formation.

Price signalling is the cousin of the ring. A WhatsApp group where three 'competitors' discuss floors is not market research. It is how an RA becomes a case. Keep your floor in an internal note. Do not send it to a peer 'for sanity'. GeM's sanity is not your group chat.

When a flag or query names a pattern you did not intend — a sudden catalogue drop, a repeated consignee, a specification that sounds like your brochure — answer the pattern, not your feelings. Yes, we dropped the price because inference cost fell, here is the dated cost note. No, we did not write the indent. Here is the mail where we asked the buyer to genericise the tests. Facts first. The monitor, human or machine, can only score what you give it.

A standing playbook so flags stay boring

  • Week 1 of every quarter: review listings against what you actually deliver.
  • Every bid: save the cost model that produced the price.
  • Every specification you help a buyer write: strip brand names unless PAC is real.
  • Every time a field is blocked: stop. Blocked means do not.
  • Every incident-shaped mail: portal first, narrative second.

File note after a query

GeM raised a query / flag on [bid or catalogue] on [date]. We have not inferred an unpublished score. We have answered the question asked, with [documents]. Related listings were reviewed on [date] for consistency with delivered architecture. No workaround to system validation was attempted. Owner: [name].

Prcept AI lists what it ships: on-prem or air-gapped agents, no training on customer data. If a category cannot say that, we ask for a better category rather than bending a printer field. That is the only sanity strategy that still works next year.

How a buyer or seller should act on this

Treat “Understanding GeM's Sanity Framework Flags” as an operating problem, not a thought piece. A P5 Founder who searches “GeM sanity framework” is usually one bid, one CRAC, or one rejection away from a cash event. The file that wins is the one with dates, document names and a named officer — not a paragraph that restates GeM’s homepage.

GeM watches for patterns that look like tinkering, splitting, collusion or nonsense prices. Treat flags as questions to answer. Do not role-play an unpublished algorithm. That is why this explainer ends in artefacts: screenshots of the live portal term, the clause you invoked, and the date you last checked it. GeM, GFR notes and state portals move. A citation without a date is folklore.

Confirm the live GeM FAQ, the current revenue policy and the bid text before you copy any number from a blog — including this one. If the portal and this article disagree, the portal wins. Put the printout in the file.

  • Write the purpose of the buy in one sentence a DDO will sign.
  • Name the route: catalogue, custom bid, bid, RA, CPPP, or state portal.
  • Attach the exemption or preference documents you will actually upload (Udyam, DPIIT, MII, OEM).
  • Record who can accept the consignee receipt and who raises the bill.
  • Do not invent a category, a PAC, or a price-reasonableness story after L1 is public.

Close this loop before the next CAB

Put “Understanding GeM's Sanity Framework Flags” on the next change-advisory or bid-opening agenda as a single line item with an owner. If it cannot earn a line item, it will not earn a control. The owner should be a P5 Founder, not “the vendor.”

Revisit the item when the model, the GeM term, the region, or the SI changes. “GeM sanity framework” is not a one-time workshop. It is a watch item. Date the last check. Unsigned watch items are souvenirs.

Questions this usually raises

What is GeM's sanity framework, exactly?
Public materials and buyer-seller chatter use sanity and integrity language for automated and human checks on bids, catalogues and transaction patterns. GeM has not published a full scoring formula you should treat as official. Describe the idea. Do not invent weights.
If my bid is flagged, am I banned?
A flag is a question. It may slow a bid, trigger a query, or become an incident. Ban, suspension and debarment follow the Incident Management policy and buyer or DoE orders, not a rumour about a score.
Is matching another seller's price a sanity problem?
Identical prices across supposedly independent bidders can look like collusion. Independent cost models that happen to land nearby are normal in a thin market. Keep workings. Do not exchange prices with a competitor.
Can I hire someone to reverse the algorithm?
No useful, honest service does that. Gaming validation is itself a Grave reason in the incident FAQs. Build a clean catalogue and a clean bid instead.

Sources