Repository navigation
spec(t27f): a research reel through the same topic workflow -- GF16 vs bfloat16 NMSE (Closes #10010) - #10012
Merged
Conversation
…s bfloat16 NMSE (Closes #10010) - topic_engine: keyword_hits + route_brief execute KEYWORD_MATCH on a brief's own text; no reel counts hits by hand any more. - topic_reel (new): the one workflow every topic reel calls -- statuses_earned, gaps_are_named, slides_fit, stage_slides_bare, counts. - reel_t27_basics: data plus course-specific checks; its local copies of the counting and slide-fit functions are gone. - reel_gf16_nmse (new): second domain, research. 27 steps from the NMSE protocol spec, doc, schema and PUB-01/PUB-02; 15 published, 1 verified, 6 drafts, 5 named gaps (the sealed run PUB-02 asks for does not exist). t27c and t27b agree on every test of all seven touched or dependent specs, and on 10 planted faults. Refs gHashTag/trinity#1677 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Contributor
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Closes #10010
Refs gHashTag/trinity#1677
What
gHashTag/trinity#1677 asks for at least two very different domains on one widget and one workflow, "without duplicated orchestration code". This PR adds the second domain and moves the workflow into one place.
topic_engine.t27:keyword_hitsandroute_brief. KEYWORD_MATCH was a sentence, and a reel counted its hits by hand. Now the engine counts them on the brief's own text (lowercase, split on non a-z/0-9, exact words, every repeat counts) and hands the counts to the existingroute. Testkeyword_match_is_executedcovers the edge cases: case, hyphens, plurals ("villas", "tests"), the empty brief and ties. The two demo briefs now route through it as well.topic_reel.t27(new): the one workflow.statuses_earned/first_unearned(every drawn status isearned_statusof its measurements),gaps_are_named(a sourceless cell says what is missing),slides_fit/stage_slides_bare(slide text fits every preset as the host breaks it), and counts. It has 5 tests of its own on fixtures.reel_t27_basics.t27is now data plus the checks that concern the course. Its BRIEF is the source's TITLE + DESCRIPTION verbatim, routed withroute_brief. Its localmeasured_status,count_*,row_count,head_charsandrow_charsare gone, andstage_slides_carry_no_card_textbecame a call to the shared check.reel_gf16_nmse.t27(new): the research reel, "GF16 against bfloat16 NMSE":specs/benchmarks/gf16_bfloat16_nmse.t27,docs/GF16_BFLOAT16_NMSE_PROTOCOL.md,schemas/nmse-protocol-v1.jsonand the PUB-01/PUB-02 roadmap.route_briefcounts gf16 twice and benchmark once, so the reel routes to research.The research map
The three lenses are formats (steps 1-9), protocol (10-18) and claims (19-27). Status counts: 15 published, 1 verified, 6 drafts, 5 gaps.
de2cff88.t27.ai/t27/files/<path>returns 200 for the spec and the schema, and 404 for the doc, FORMAT-SPEC-001.json, the 2026-04-05 results, the crosswalk and the paper draft. A made-up path also returns 404.bench/results/nmse-*.json, and the repo holds none. So these are named gaps, each with a MISSING text:tt-trinity-gammanmse_gf16_bf16.golden.jsonatde484c4fpins seeds but saysmeasured_silicon: false.conformance/gf16_bench_results.json(no seed, no seal) are drawn as a draft. The protocol itself says they do not count.Two compilers, t27c and t27b (Railway lab)
test --check, qemu-aarch64)a_gap_is_named_not_filled,a_status_is_earned_not_chosena_status_is_earned_not_chosena_gap_is_named_not_filledthe_domain_is_routed_not_chosenthe_domain_is_routed_not_chosena_status_is_earned_not_chosenGF16_MANT_BITS = 8both_layouts_are_sixteen_bits,gf16_trades_exponent_for_mantissaa_status_is_earned_not_chosenfirst_unearnednever reports a mismatchtopic_reel.t27a_drawn_status_must_be_the_earned_oneWhen the shared
first_unearnedis broken, the research reel itself stays green: a reel trusts the shared checker, andtopic_reel.t27's own test is what catches the break. That is the intended split.Each plant was restored afterwards, and the lab files match the commit byte for byte (sha256 prefixes
8b37f309,2a873424,adc9acbf,74baa4aa).A first draft of the research reel was refused by both compilers in
every_slide_fits_every_preset. The call headline "One run is missing" is 18 characters, which is over the split card's one-line limit, so it is now "The missing run".Stills
t27f_view(t27-mcp07e2c7c) drew 15 of 15 stills (5 scenes x wide/square/tall) and refused none. The overview's cells are, in step order:#5eead4;#2dd4bf(step 20);#fbbf24;#ff7a45.The focus slide is step 11, "Check phi before any NMSE".
Not claimed
No NMSE value, no ratio and no winner: the reel says that run does not exist. No GIF or MP4. CI reds
cli-tri(#9634) andt27b-native-ratchetare inherited from master.🤖 Generated with Claude Code