Files

T

Brummel 0900f3f413 iter embedding-abi-m5.3 (DONE 3/3): time-shard boundary-invisibility proof + friction harvest

Final functional M5 iteration (spec ae905de, plan 67027ab; on
green m5.1 204c171 + m5.2 b724cd1 whose leak-proof resolved via
dbd76e5). Proves the chunk/window-boundary-invisibility claim the
M4 retirement rests on:

- adapter::fold_window — purely additive windowed sibling of
  fold_symbol (same MidPriceStream+Kernel path; stream_tick_windowed).
  fold_symbol/MidPriceStream/Kernel byte-untouched.
- chrono as a DEV-dep only — the time-shard test derives
  (y,m)->Unix-ms bounds (data-server's own precedent) and passes
  raw ms to the bin; the bin is date-math-free; Invariant 1
  unaffected.
- new single-responsibility [[bin]] timeshard_runner — one thread
  per disjoint window, each owning its Kernel (Ctx:!Send =>
  compile-time per-thread-ctx).
- tests/timeshard.rs — spec §3 strength-ordered: (a) per-shard
  (acc,n) BIT-EXACT vs single-thread host fold of that exact window
  in stream order (THE boundary-invisibility proof); (b) Σ partials
  bit-exact by construction; (c) whole-window within REL_TOL=1e-6
  only (cross-shard f64 reassociation is host arithmetic, never a
  kernel property — bit-exact there would be the bug; tolerance
  derived from the recursive-summation error bound). Plus the
  now-deterministic leak-Σ + a journal-only friction timing capture
  (no bench gate, no timing assertion).

Concrete pin: EURUSD 2017-03/04/05 (recon-verified present,
contiguous, no gaps; data-server has no month-listing API). RED was
a genuine deterministic compile failure (declared-but-absent bin
source; Cargo missing-source error vs the plan's predicted env!
error — same cause, no false-green; benign plan-text imprecision in
Concerns).

Boss-verified independently: timeshard determinism 5/5 (no jitter);
full ail-embed suite 0 failed AND 0 ignored across all binaries
(m5.2 symbol_fan_swarm_leak_free stays un-ignored & green — the
saga's resolution intact); isolated embed_rc_global_stats_race
still green; AILang workspace data-server count 0; compiler-surface
diff empty (zero diff to crates/ailang-*, crates/ail/, runtime/,
examples/*.ail, root Cargo.toml, DESIGN.md).

Friction-harvest deliverable (P2 flat-array-decision input):
host-per-tick-FFI ~= 658 ms / 3,192,562 ticks ~= ~206 ns/tick at
real EURUSD volume.

M5 functional work complete; milestone closes next via the
mandatory audit (which also handles the pre-existing
DESIGN.md:2358-2360 "additive M4 concern" drift, out of m5.3 scope).
Includes the per-iter journal, stats, and the INDEX.md line.

2026-05-19 02:40:16 +02:00

5.0 KiB

Raw Blame History

iter embedding-abi-m5.3 — time-shard boundary-invisibility proof + friction harvest

Date: 2026-05-19 Started from: 67027ab0f9 Status: DONE Tasks completed: 3 of 3

Summary

Final functional M5 iteration. Proves the chunk/window-boundary-invisibility claim the M4 retirement rests on: one symbol (EURUSD), three disjoint contiguous single-month windows (2017-03 / 2017-04 / 2017-05), each folded on its own thread owning its own Kernel (Ctx: !Send makes one-ctx-per-thread a compile-time guarantee), each shard's (acc,n) asserted bit-exact vs a single-thread host fold of that exact window in data-server stream order. Adds the windowed adapter sibling fold_window (purely additive after fold_symbol), a new single-responsibility [[bin]] timeshard_runner, and tests/timeshard.rs carrying the spec §3 strength-ordered (a)/(b)/(c) assertions plus the deterministic leak-Σ and a journal-only friction-timing capture (no bench gate, no timing assertion). The shipped m5.2 symbol-fan path (swarm_runner.rs, swarm.rs) and m5.1 core stay byte-untouched and green. Invariant 1 holds (the AILang workspace graph has zero data-server; chrono enters only as an ail-embed dev-dep). Determinism re-verified 6× (5× standalone + 1× in full suite), all identical. Friction-harvest observed: TIMING total_ticks=3192562 elapsed_ms=658 → derived per-tick-FFI cost ≈ 658 ms / 3,192,562 ≈ ~206 ns/tick (the Boss records this as the P2 flat-array-decision input in the close-out journal).

Per-task notes

iter embedding-abi-m5.3.1: windowed adapter + chrono dev-dep + bin declaration — fold_window(server, symbol, from_ms, to_ms) appended to ail-embed/src/adapter.rs immediately after fold_symbol, before the #[cfg(test)] mod tests block (windowed sibling, same MidPriceStream+Kernel path, differs only by stream_tick_windowed); chrono = "0.4" added to [dev-dependencies] + [[bin]] timeshard_runner declared in ail-embed/Cargo.toml. --lib regression check 2 passed; 0 failed (exactly the plan's prediction). Spec compliant, quality approved, 0 re-loops.
iter embedding-abi-m5.3.2: the time-shard E2E test (RED) — ail-embed/tests/timeshard.rs created verbatim from the plan (boundary-invisibility (a)/(b)/(c) + leak-Σ + friction capture). RED observed: the declared-but-absent [[bin]] timeshard_runner source makes the test binary fail to build. Spec compliant, quality approved, 0 re-loops. DONE_WITH_CONCERNS — see Concerns (RED-text discrepancy, same cause).
iter embedding-abi-m5.3.3: the timeshard_runner bin (GREEN) — ail-embed/src/bin/timeshard_runner.rs created verbatim (one thread per window, each building its own Kernel; prints RESULT/TIMING). Step 2 GREEN (1 passed; 0 failed, ran 2.71s against real /mnt data — did NOT skip). Step 3 determinism ×5: 5/5 ok. Step 4 full suite: 0 failed AND 0 ignored across all binaries (m5.2 symbol_fan_swarm_* un-ignored & green). Step 5: cargo metadata data-server count = 0; path filter empty. Step 6: embed_rc_global_stats_race 1 passed; 0 failed. Spec compliant, quality approved, 0 re-loops.

Concerns

iter embedding-abi-m5.3.2 (DONE_WITH_CONCERNS): the Task 2 Step 2 RED surfaced as error: couldn't read \src/bin/timeshard_runner.rs`: No such file or directoryrather than the plan's predictederror: environment variable `CARGO_BIN_EXE_timeshard_runner` not defined at compile time. Same root cause (the declared bin timeshard_runnertarget has no source until Task 3), same trigger, same resolution (disappears exactly when Task 3 lands the bin) — Cargo fails at the missing-source check before reaching theenv!` expansion, so the surface error differs but the RED is genuine, deterministic, and not a false-green. Observation only; the plan's intent (a deterministic compile-failure RED isolating Task 2 from Task 3) is fully satisfied.

Known debt

(none introduced by this iter)
Pre-existing docs/DESIGN.md:2358-2360 "additive M4 concern" drift is explicitly out of m5.3 scope (spec §"Out of scope" / Boss decision) and was deliberately not touched — the milestone-close audit handles it, not this iter.

Blocked detail

(not applicable — Status DONE)

Files touched

ail-embed/src/adapter.rs (modified — additive fold_window)
ail-embed/Cargo.toml (modified — chrono dev-dep + timeshard_runner bin)
ail-embed/Cargo.lock (modified — single line: chrono direct-dep edge; automatic, strictly required by the Cargo.toml dep add, already resolved transitively via data-server)
ail-embed/src/bin/timeshard_runner.rs (created)
ail-embed/tests/timeshard.rs (created)

All within the plan's declared file set and the Step 5b allowed path scope (ail-embed/**, docs/, bench/orchestrator-stats/). Zero diff to crates/ailang-*, crates/ail/, runtime/, examples/*.ail, root Cargo.toml, docs/DESIGN.md.

Stats

bench/orchestrator-stats/2026-05-19-iter-embedding-abi-m5.3.json

5.0 KiB Raw Blame History Unescape Escape