Skip to content
AGENTIC Data Mining

DIBIQ · DLA DIBBS bid intelligence

Know what the last one actually cost.

A working portal for the open DLA DIBBS board. Search today's solicitations, open a stock number, and read the government's own published prices for it — per unit, with the basis on every row.

Where we hold no price for your item, the page says so in those words. Never a zero, never a blank, never an average standing in for a number we do not have.

Open DIBIQ See what it will not do ↓

Four prices, one stock number

Award line

Per unit

Firsthand. Read off the signed award document.

Procurement history

Per unit

Secondhand. Republished by the government inside a later solicitation.

Catalog reference

Per unit

A listed reference price. Not a transaction anyone made.

Contract total

Not per unit

The whole contract's total, repeated unchanged on every line under it.

A blended “market rate”

Refused

Different quantities, different units of issue, and a currency DLA does not print. No mean, median, trend or spread is computed over these rows at any count.

The difference that decides a bid

An order total divided by nothing is not a unit price

DLA publishes an award's total far more often than it publishes a per-unit figure, and it repeats that same total unchanged on every line underneath the contract. A tool that lists those totals is not telling you what the item costs — it is telling you what somebody's whole order came to, with the quantity left out.

Averaging figures of that shape and calling the result a unit price is not a rounding error — it is a different quantity wearing the same label. Measured against our own corpus on 30 August 2026: DLA published a whole-contract total on 58,706 of the 62,422 award records we hold, and a per-unit figure on none of them. Every genuine per-unit price on this platform comes from somewhere else.

So DIBIQ publishes a genuine per-unit price where the government published one, labels every row with which of the four bases it came from, and keeps the four apart. A total is shown as a total. It is never divided, never blended, never quietly promoted into a unit price.

Inside the portal

One stock number, six things worth knowing about it

Every panel reads the same corpus, and every panel distinguishes “we looked and found nothing” from “this was never loaded” from “this is withheld”. Three different answers, never one shrug.

The conduct envelope

Two capabilities are built, and deliberately switched off

Both exist in the query layer. Both are refused server-side, with a stated reason, rather than quietly returning nothing — because a policy refusal and an empty result are different answers and a bidder is entitled to know which one they got.

  • Withheld by owner decision

    Search every item's free text for a characteristic.

    characteristics · find_items

    The free text on this table names nuclear-hardness-critical features and specific weapons systems, on supply classes that are not export-restricted. Looking one item up requires already holding its stock number. An aggregate keyword search over the same text would turn it into an index of which parts are nuclear-survivability-critical — a materially different exposure, and not one we are willing to publish.

  • Held pending validation

    Comparables sharing your exact part number.

    comparable_prices · tier 1

    It would be the strongest tier, so a wrong row displaces a right one instead of sitting beside it. Run against the real cross-reference it matched specifications as though they were part numbers — one coating specification carried by 625 different stock numbers returned six unrelated items spanning $83.70 to $297.84, every one of them labelled “the same physical part”. The fixes are tuned on the single sample that exposed the fault, so the tier stays off until they have been tested against a bigger one.

  • Structural, not a promise

    Fetch it from DIBBS on the customer's behalf.

    no outbound requests

    DIBIQ reads our own ingested corpus and nothing else. It never authenticates against, posts to, or automates any DLA host, and it never acts on a DIBBS session belonging to you. That is enforced structurally: no file in the portal's source may name a DIBBS or DLA host — or any external host at all — and a scan over every one of them fails the build if one appears, so the constraint cannot be softened quietly.

Quoting and submitting stay yours. DIBIQ prepares the evidence; it does not touch the bid board on your behalf.

Depth

Held, not sampled

These are row counts and date ranges from the live corpus, measured on 30 August 2026. They describe the whole holding, not any single item: the earliest price we hold for your stock number is whatever the government published for it, and the portal shows that date on the item itself.

Award history
2014 → today
62,422 DIBBS award records, earliest 11 Aug 2014.
Prior-price history
1980 → today
Earliest award date on file across the corpus: 24 May 1980.
Item catalog
3,885,898
Releasable NIINs from FLIS / PUB LOG.
Item characteristics
27,937,917
Published technical characteristic rows, keyed per item.
Cross-references
10,281,724
Part number to CAGE to stock number entries.
Supplier registrations
853,880
SAM entity registrations, joined to awardee CAGEs.

The item catalog is export-control carve-out filtered before it reaches us, so a stock number missing from it is withheld from public release rather than nonexistent — and the portal says which of those two it means.

Open DIBIQ The underlying datasets