Skip to content

spec(igla): own_model cell 23 verified -- 17 of 17 ML specs compile (Closes #10163) #10164

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
gHashTag merged 2 commits into from
Oct 18, 2026
Merged

spec(igla): own_model cell 23 verified -- 17 of 17 ML specs compile (Closes #10163) #10164

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
82ea8dc
spec(igla): own_model cell 23 verified -- 17 of 17 ML specs compile (…
gHashTag Oct 18, 2026
d1e247f
spec(igla): own_model -- re-take the DELIVER cells' evidence for revi…
gHashTag Oct 18, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
46 changes: 29 additions & 17 deletions specs/igla/own_model.t27
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -61,13 +61,23 @@
; forwardTrain (bind, rotate, the PHI_INV gate, +0.1 * project(context) when it fires, the
; output into the next block); the stream never reads a float parameter, so the gradient
; checks hold with it on. A gate mutant (> for >=) fails 2 tests and a gain-0 mutant fails 1.
; - re-test on 2026-10-18, t27c built from SOURCE_COMMIT, lab checkout at master ba78d6d3c, after
; the PR that made nn/attention compile (#10094): nn/attention 35/35 pass, 9 invariants proved,
; 0 vacuous. The other seven re-run on the same checkout still pass (cross_entropy_loss 5/5, sgd
; 4/4, adam 4/4, adamw 12/12, lamb 4/4, nn/hslm 61/61, hslm/forward_pass 14/14), so 17 of the 17
; ML specs compile and pass, and cell 23 has its passing test. The DELIVER cells' evidence was
; re-taken on the same checkout for the review of that PR: nn/ternary_ste 26/26,
; nn/hslm_backward 15/15, nn/hslm_data 39/39, ml/arena_heap 10/10, ml/file_io_contract 12/12,
; 0 vacuous (cells 16-18); playra/tern-tc-9m and tern-tc-9m-t27 200, tern-tc-100m 401 (cells 7
; and 8); the board /state 200 (cell 21). A review of that PR is what the engine asks of cells
; 7, 16, 17, 18 and 27; REVIEWED turns true only in a later PR that names the approving review.
; - live: GET huggingface.co/api/models/playra/tern-tc-9m and .../tern-tc-9m-t27 returned 200;
; .../tern-tc-100m returned 401, the same as a made-up name, so no 100M is public.
; GET board-production-2dec.up.railway.app/state returned 200 and a made-up path 404; its
; snapshot is dated 2026-10-05, so the board is live and 12 days stale.
; - review: no PR behind any cell has a review, so no DELIVER cell is VERIFIED.
; The engine asks an engineering LEARN step for a source link, a PRACTICE step for a passing test
; and a DELIVER step for a review. So: 1 published, 15 verified, 6 drafts, 5 gaps.
; and a DELIVER step for a review. So: 1 published, 16 verified, 5 drafts, 5 gaps.

; PERCENT is a second, separate number per cell: an estimate of how much of that part exists, with
; its basis in SOURCES. It never feeds a status. The lens and overall figures are plain means, every
Expand Down Expand Up @@ -184,7 +194,7 @@ pub const SOURCES : [27]str = [
"trinity-fpga b2bd5a30, 7682304/7682304 receipts",
"board-production-2dec.up.railway.app/state",
"t27 specs/port/trinity/fpga/openxc7-synth/ternary_matvec.t27",
"t27c test-report on 17 ML specs: 16 compile",
"t27c test-report on 17 ML specs: 17 compile",
"",
"",
"",
Expand All @@ -208,20 +218,20 @@ pub const MISSING : [27]str = [

/// The measurements per cell (see the header for how each was taken).
pub const GROUNDED : [27]bool = [true, true, true, true, true, true, true, false, false, true, true, true, true, true, true, true, true, true, true, true, true, true, true, false, false, false, true];
pub const TEST_PASS : [27]bool = [false, false, false, true, true, true, false, false, false, false, false, false, true, true, true, false, false, false, false, false, false, true, false, false, false, false, false];
pub const TEST_PASS : [27]bool = [false, false, false, true, true, true, false, false, false, false, false, false, true, true, true, false, false, false, false, false, false, true, true, false, false, false, false];
pub const REVIEWED : [27]bool = [false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false, false];
pub const LIVE : [27]bool = [false, false, false, false, false, false, true, false, false, false, false, false, false, false, false, false, false, false, false, false, true, false, false, false, false, false, false];

/// What a host passes the viewer, per cell: the earned status code and the step number as a label.
pub const STATUS : [27]u8 = [4, 4, 4, 4, 4, 4, 3, 0, 0, 4, 4, 4, 4, 4, 4, 3, 3, 3, 4, 4, 5, 4, 3, 0, 0, 0, 3];
pub const STATUS : [27]u8 = [4, 4, 4, 4, 4, 4, 3, 0, 0, 4, 4, 4, 4, 4, 4, 3, 3, 3, 4, 4, 5, 4, 4, 0, 0, 0, 3];
pub const LABELS : [27]str = ["1", "2", "3", "4", "5", "6", "7", "8", "9", "10", "11", "12", "13", "14", "15", "16", "17", "18", "19", "20", "21", "22", "23", "24", "25", "26", "27"];

/// Percent of each part that exists, estimated from its source (see the header). Never a status.
pub const PCT : [27]u8 = [60, 70, 80, 80, 60, 100, 50, 0, 0, 100, 100, 20, 80, 65, 95, 75, 55, 40, 70, 15, 60, 40, 94, 0, 0, 0, 50];
pub const PCT : [27]u8 = [60, 70, 80, 80, 60, 100, 50, 0, 0, 100, 100, 20, 80, 65, 95, 75, 55, 40, 70, 15, 60, 40, 100, 0, 0, 0, 50];

/// The t27c test-report counts the PRACTICE cells quote.
pub const ML_SPECS_MEASURED : i32 = 17;
pub const ML_SPECS_COMPILING : i32 = 16;
pub const ML_SPECS_COMPILING : i32 = 17;
pub const CODER_SPECS : i32 = 9;
pub const CODER_SPECS_COMPILING : i32 = 9;

Expand All @@ -240,9 +250,9 @@ pub const ROWS : [20]str = [
"64 Zig files; best PPL 2.96 at ctx 27",
"Those checkpoints were wiped",
"In t27: loss, optimizers, HSLM pass",
"16 of 17 ML specs compile on the lab",
"17 of 17 ML specs compile on the lab",
"9 of 9 IGLA CODER specs compile",
"1 published, 15 verified, 6 drafts",
"1 published, 16 verified, 5 drafts",
"5 gaps, about 54 percent overall",
"Batched f32 HSLM trainer",
"Shadow weights for the whole model",
Expand Down Expand Up @@ -313,15 +323,15 @@ test a_gap_is_named_not_filled {

test a_status_is_earned_not_chosen {
assert statuses_earned(STATUS, DOMAIN, GROUNDED, TEST_PASS, REVIEWED, LIVE);
// the evidence slide's "1 published, 15 verified, 6 drafts" and "5 gaps"
// the evidence slide's "1 published, 16 verified, 5 drafts" and "5 gaps"
assert count_status(STATUS, STATUS_PUBLISHED) == 1;
assert count_status(STATUS, STATUS_VERIFIED) == 15;
assert count_status(STATUS, STATUS_DRAFT) == 6;
assert count_status(STATUS, STATUS_VERIFIED) == 16;
assert count_status(STATUS, STATUS_DRAFT) == 5;
assert count_status(STATUS, STATUS_GAP) == 5;
assert count_true(REVIEWED) == 0;
// every passing test is a PRACTICE step: the t27 bench, MultiPL-E and coder specs (cells 4 to
// 6), loss, optimizers, forward pass, matvec
assert count_true(TEST_PASS) == 7;
// 6), loss, optimizers, forward pass, matvec, and the ML specs that compile (cell 23)
assert count_true(TEST_PASS) == 8;
var c : i32 = 0;
while (c < CELL_COUNT) {
const i : usize = @intCast(c);
Expand All @@ -347,11 +357,13 @@ test the_compile_counts_are_the_measured_ones {
assert p5 == share_pct(CODER_SPECS_COMPILING, CODER_SPECS);
// the compile cell's percent is the measured share, truncated
const share : i32 = share_pct(ML_SPECS_COMPILING, ML_SPECS_MEASURED);
assert share == 94;
assert share == 100;
const p22 : i32 = @intCast(PCT[22]);
assert p22 == share;
// the compiling ML specs are the nine coder specs plus the six that compiled before them
assert ML_SPECS_COMPILING == CODER_SPECS_COMPILING + 7;
// the compiling ML specs are the nine coder specs plus the eight others
assert ML_SPECS_COMPILING == CODER_SPECS_COMPILING + 8;
assert ML_SPECS_COMPILING == ML_SPECS_MEASURED;
assert STATUS[22] == STATUS_VERIFIED;
assert CODER_SPECS <= ML_SPECS_MEASURED;
// the two coder cells that carry numbers (bench, MultiPL-E) are verified by their spec's test
assert TEST_PASS[3];
Expand All @@ -363,7 +375,7 @@ test the_compile_counts_are_the_measured_ones {
test percent_per_lens_is_a_plain_mean {
assert mean_pct(0, 9) == 55;
assert mean_pct(9, 9) == 70;
assert mean_pct(18, 9) == 36;
assert mean_pct(18, 9) == 37;
// the evidence slide's "about 54 percent overall"
assert mean_pct(0, CELL_COUNT) == 54;
}
Expand Down
22 changes: 11 additions & 11 deletions specs/igla/own_model_board.t27
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -48,7 +48,7 @@ pub const LENS_CELLS : i32 = 9;

pub const JOB : str = "own_model";
pub const STAGE : str = "own_model";
pub const MSG : str = "1 published, 15 verified, 6 drafts, 5 gaps; 54%";
pub const MSG : str = "1 published, 16 verified, 5 drafts, 5 gaps; 54%";
/// What MSG says once its numbers are taken out.
pub const MSG_WORDS : str = " published, verified, drafts, gaps; %";
pub const OVERALL_PCT : u8 = 54;
Expand All @@ -59,12 +59,12 @@ pub const KEY_CELLS : str = "cells";
pub const KEY_STATUS : str = "status";
pub const KEY_PCT : str = "pct";

pub const LENS_PCT : [3]u8 = [55, 70, 36];
pub const CELL_STATUS : [27]u8 = [4, 4, 4, 4, 4, 4, 3, 0, 0, 4, 4, 4, 4, 4, 4, 3, 3, 3, 4, 4, 5, 4, 3, 0, 0, 0, 3];
pub const CELL_PCT : [27]u8 = [60, 70, 80, 80, 60, 100, 50, 0, 0, 100, 100, 20, 80, 65, 95, 75, 55, 40, 70, 15, 60, 40, 94, 0, 0, 0, 50];
pub const LENS_PCT : [3]u8 = [55, 70, 37];
pub const CELL_STATUS : [27]u8 = [4, 4, 4, 4, 4, 4, 3, 0, 0, 4, 4, 4, 4, 4, 4, 3, 3, 3, 4, 4, 5, 4, 4, 0, 0, 0, 3];
pub const CELL_PCT : [27]u8 = [60, 70, 80, 80, 60, 100, 50, 0, 0, 100, 100, 20, 80, 65, 95, 75, 55, 40, 70, 15, 60, 40, 100, 0, 0, 0, 50];

/// The size of the compact metrics object, in bytes, computed from the keys and values above.
pub const METRICS_BYTES : i32 = 625;
pub const METRICS_BYTES : i32 = 626;

pub fn is_digit(ch: u8) -> bool {
return ch >= '0' and ch <= '9';
Expand Down Expand Up @@ -261,8 +261,8 @@ test the_tally_sums_to_twenty_seven {
assert count_status(CELL_STATUS, STATUS_OPEN) == 0;
assert published + verified + drafts + gaps == CELL_COUNT;
assert published == 1;
assert verified == 15;
assert drafts == 6;
assert verified == 16;
assert drafts == 5;
assert gaps == 5;
}

Expand All @@ -273,11 +273,11 @@ test the_message_is_the_tally {
const n : i32 = @intCast(MSG.len);
assert n <= MSG_MAX;
// one number wrong, a word changed, or a number added: each refused
assert msg_matches("1 published, 14 verified, 6 drafts, 5 gaps; 54%", STATUS, overall) == false;
assert msg_matches("1 published, 15 verified, 6 drafts, 5 holes; 54%", STATUS, overall) == false;
assert msg_matches("1 published, 15 verified, 6 drafts, 5 gaps; 54% of 100", STATUS, overall) == false;
assert msg_matches("1 published, 15 verified, 5 drafts, 5 gaps; 54%", STATUS, overall) == false;
assert msg_matches("1 published, 16 verified, 5 drafts, 5 holes; 54%", STATUS, overall) == false;
assert msg_matches("1 published, 16 verified, 5 drafts, 5 gaps; 54% of 100", STATUS, overall) == false;
assert msg_matches(MSG, STATUS, 48) == false;
assert nth_number(MSG, 1) == 15;
assert nth_number(MSG, 1) == 16;
assert nth_number(MSG, 4) == 54;
assert nth_number(MSG, 5) == -1;
}
Expand Down
Loading