State of the Spec: Q4 2026

A building product is chosen twice. The second time is the purchase, the one with a price on it. The first time is the specification, written months earlier by an engineer who picked a real product and built the section around it. Everything after that, the bid, the submittal, the substitution fight, is other manufacturers arguing with a decision that has already been made.

This is the first State of the Spec, a quarterly count of that first decision, taken from the public bid documents SpecIndex reads. It counts positions: the places in a specification where a brand is named, classified the way the document itself classifies them, as the basis of design, as an approved equal, or as a proprietary naming that allows no equal at all. Three terms carry this edition. A spec position is one place in a specification where a brand is named. The basis of design is the product a section is built around. A proprietary position is a naming that allows no equal at all. The population is stated wherever a figure appears, and the method is stated in full in section 7. Nothing in this edition is a sample. Every scope named below was counted completely.

The headline findings are three. First, 14.5 percent of basis of design positions are written proprietary (90 of 622): roughly one design seat in seven is closed on its face, before any equal is considered. Second, there are 6.8 approved equal positions for every basis of design position (4,216 against 622): the equal list is where most named positions live, and it is crowded. Third, the typical basis of design section names 2 manufacturers on its approved list (median, across the same 622 sections): the list beside the design seat is usually short, even though the count of equal positions across the market is large. Those two facts together are the shape of the spec market. Few seats, short lists, and a long tail of brands named somewhere as an equal and almost nowhere as the design.

1. The headline numbers

Proprietary share, 14.5 percent. A basis of design position is classified proprietary when the section's own language names the product and permits no substitution path. The full openness distribution across the 622 basis of design positions in this count: or equal 489 (78.6 percent), proprietary 90 (14.5 percent), performance 28 (4.5 percent), unspecified 15 (2.4 percent). Most public sections cannot legally close, and most do not. The 14.5 percent that do are the positions a competitor cannot win at the submittal stage, because they were never open.

Approved equals per basis of design position, 6.8. Approved equal positions are 87.1 percent of all 4,838 citable positions in the count. One section naming a basis of design product and a list of equals produces one design position and several equal positions, so this ratio is a property of positions, not of projects. On a project basis, the same population holds 1,421 projects.

Approved list size, median 2. The distribution behind that median matters: 41.2 percent of basis of design sections name one manufacturer or fewer on the parsed approved list, 59.8 percent name three or fewer, and the mean is 9.9, pulled up by a small number of very long federal lists (the maximum is 169). The median is the honest central figure, and it is the number a rep should hold in mind. The list you are trying to join usually has two names on it.

A note on verification. These figures were first measured on 13 October 2026 for the SpecIndex home page. This edition re counted the entire population from the raw rows on 14 October, all 150 brand and state scopes, and the recount returned the identical set: 4,838 positions, the same 622 design seats, the same 90 proprietary, the same distribution, the same counts for every brand in every state. A quarterly series that cannot survive a recount is not a series. This one starts by surviving one.

2. The corpus this edition counts

The position statistics above describe a deliberately bounded population: 15 national brands, named below, across the ten states where the index holds the most documents. The wider index they sit inside, measured the same day across all 50 states, is this:

Documents searched is the count of public bid documents the index can search and cite, which is the measure every position in this edition rests on. Projects tracked is the wider register of projects the index follows, most of which have no public document to read. Counts describe the public bid documents SpecIndex has read. Privately bid work produces no public specification and is not represented anywhere in this edition. This frame counts the 50 states; the index also searches documents in jurisdictions beyond the 50 states, and those totals are additional to the frame shown here.

3. Divisions: where the design seat is held, and where it is not

The panel's positions by CSI division, same population as the headline count:

Three readings stand out. Division 23, HVAC, carries the most citable positions in the panel (1,425) and closes a fifth of its design seats (19.6 percent proprietary). Division 09, Finishes, holds the most design seats outright (289 of the 622), because its panel brands, Armstrong and Sherwin-Williams, are named as the design basis more often than they are listed as equals. Division 26, Electrical, is the panel's most list driven division: 93.4 percent of its positions are approved equals, and its brands hold the design seat in only 74 of 1,126 positions. A manufacturer reading its own division in this table is reading the structure of its fight. In Division 26 the fight is for the seat, because almost nobody in the panel holds it. In Division 23 the fight is against closure, because a fifth of the seats cannot be contested at all.

4. States: where the documents are

Documents concentrate the way construction concentrates, unevenly. The ten states holding the most documents in the index on 14 October 2026, with the panel's citable positions in each:

Texas holds 1,431 of the panel's positions in these ten states, nearly three in ten, on 315 projects. New York tracks the most projects of any state in the frame (233,172) and converts a smaller share of them into documents, which is why it ranks sixth by documents while ranking first by projects tracked. New York and Virginia land on the same projects with documents count, 2,171 each, a coincidence in the record rather than an error in the table. Virginia is new to this table: it did not sit in the ten highest document states when the baseline was measured on 13 October, and it enters at fifth. Its book also reads the most closed of any state in the frame. Across the panel's 261 Virginia positions, 69 are basis of design and 17 of those are proprietary, a 24.6 percent proprietary share against the panel's 14.5 percent. Pennsylvania, which Virginia displaces, reads 8,405 documents and holds 247 panel positions, 45 of them basis of design, 8 proprietary. This edition keeps the headline statistics on the October population so the series stays comparable, and reports the current frame here and in section 6, so the change of frame is on the record rather than absorbed silently.

5. The panel, brand by brand

Aggregate position counts for the 15 panel brands, summed over the ten headline states. These are counts of named positions in public bid documents, published with the method in section 7. They are not market shares, and a zero in the proprietary column means none of that brand's design seats in this population were classified proprietary, nothing more.

The two finishes brands hold 307 of the panel's 622 design seats between them, which is the Division 09 result from section 3 seen from the other side. The electrical brands show the opposite profile: Eaton, Siemens, Legrand, and Schneider Electric together hold 123 design seats against 1,314 equal positions. Being named is common. Being the design is rare. The distance between those two columns is the distance between a brand a specifier has heard of and a brand a specifier has chosen.

6. What changed since the October baseline

One day separates this edition's count from the baseline it verifies, and the honest answer is: in the position layer, nothing. The recount matched the baseline exactly, scope by scope and brand by brand, which is reported in section 1. The corpus around the positions moved, as it does every day: documents searched rose in Texas (21,651 to 22,234) and Ohio (7,553 to 8,638) and read lower in Pennsylvania, with smaller moves elsewhere. Because those counts moved in both directions, this edition treats them as two reads of a changing index, not as growth or decline, and draws no state level change claim from them. The frame change they produced, Virginia in and Pennsylvania out, is handled openly in sections 4 and 7 rather than by quietly rebasing the headline numbers.

Recomputed on today's ten states instead of the October ten, the headline statistics read: proprietary share 15.3 percent (99 of 646), approved equals per basis of design position 6.5, median approved list 2, over 4,852 citable positions on 1,414 projects. The gap between that and the headline figures is the size of the frame's effect, published so a reader can judge it. The next edition, Q1 2027, is the first that can show a full quarter of movement in the position series. Until then, the series stands at two points, and they coincide.

7. How this was counted

1. Positions were pulled on 14 October 2026 through the SpecIndex index tools, on the same counting basis the brand check uses: server side counts over citable brand mention rows, where a position's total is its basis of design plus approved equal rows, and proprietary means a basis of design row whose classified openness is proprietary.
2. The panel is 15 national brands: Carrier, Trane, Daikin, Johnson Controls, and Honeywell in Division 23; Acuity, Eaton, Legrand, Schneider Electric, and Siemens in Division 26; Kohler and Moen in Division 22; Armstrong and Sherwin-Williams in Division 09; Owens Corning in Division 07.
3. The headline population is those brands across California, Texas, Florida, North Carolina, Pennsylvania, New York, Washington, Illinois, Georgia, and Ohio, the ten states of the 13 October baseline. Virginia was counted as an eleventh state for the state table and the sensitivity figures, because it entered the ten highest document states on 14 October.
4. Every scope was enumerated completely at 200 rows per pull. In all 165 scopes the rows pulled equal the server reported total. No scope reached the tool's row cap.
5. Every headline figure was computed two ways, from the server counts and from the pulled rows, and the two agree in all cases. Where a median or percentile is reported, it is computed from the pulled rows; percentiles use the nearest rank.
6. The corpus frame (projects tracked, documents searched, projects with documents) was measured per state across all 50 states on the same day and summed. Documents searched counts documents the index can search and cite; it is not a count of files in storage. Frame counts are reads of a live index, and they move in both directions between measurement days as documents are added, reprocessed, and withdrawn: Pennsylvania's documents searched read 12,634 on 13 October and 8,405 on 14 October. This edition records that movement here, as a property of indexing a live corpus, and draws no state level change claim from it.
7. The window is positions held in the index on the measurement day. Position rows carry no date the specification was written, and project announcement dates in the population run from 2022 through 2026.

Method note. The count behind every figure in this edition is the 15 brand panel named in point 2, enumerated completely rather than sampled: 150 brand and state scopes, 4,838 citable positions in all. The window is 13 and 14 October 2026, the headline population being the ten states of the 13 October baseline, counted in full on the first day and recounted in full on the second. Each basis of design position carries one of four openness readings, taken from the section's own language: or equal, where the section names the design product and allows named equals in its place; proprietary, where the section names the product and permits no substitution path; performance, where the section is written to what the product must do rather than to a product name; and unspecified, where the language settles none of these. Questions about the method go to Asif Hussain at [email protected].

8. What this edition did not measure

Three things a reader might expect are absent on purpose, because the data does not carry them honestly.

How far ahead of bid the design is set. There is no written date in the position data, and the available proxy, project announcement to bid due date, fails at row level: bid due dates exist for under a fifth of the baseline population's projects, and most of those windows run backwards, because the announcement field records a later portal event rather than the day the spec was set. Any number published here would describe portal lag and call it design timing. It is not published.

A single path measure beyond proprietary. A basis of design section whose parsed list names one manufacturer looks closed, but most such sections carry or equal language that expressly allows a substitute, and some proprietary sections carry long parsed lists. The list count and the language disagree in both directions, so the proprietary classification is the only closure measure this edition stands behind, and it is already section 1's first figure.

Everything outside the panel and the public record. Fifteen large national brands are not a draw of all manufacturers, and regional brands may read differently. Privately bid work produces no public specification and appears nowhere in these counts. A brand absent from a table here is absent from the public documents counted, which is a statement about the documents, not about the market.