3 online · 2 deterministic
AI OPERATIONS / PUBLIC LEDGER v1.0.0
Every model gets
a paper trail.
Every preserved research run appears here—including rejections, uncalled roles, release holds, and receipt gaps. GLM-5.3 is part of the method, so its record is part of the product.
Completed / configured
Legacy gaps are not estimated
2 rejected · 1 withheld
DEFAULT ONLINE MODEL
GLM-5.3
z-ai/glm-5.3Role separation does not create independent peer review. Online roles currently share GLM-5.3 through OpenRouter; deterministic checklists are labeled separately.
Exact tokens, cost, latency, and response identifiers are reported only when a preserved call ledger supplies them. Missing legacy receipts remain null.
Open the complete machine-readable ledger ↗Python owns acquisition, statistics, figures, and deterministic checks.
Each configured specialist leaves its own structured decision—or an explicit “not invoked.”
Code reconciles required checks and blocks any non-pass specialist decision.
A passing research gate can still be withheld by visual or deployment QA.
Cost and usage appear only when the router ledger actually preserved them.
RUN REGISTER / 05
The wins and
the wreckage.
Newest first. Each outcome is derived from immutable manifests, specialist artifacts, mechanical gate decisions, and post-run release notes.
RUN 05
TOUCHDOWN REGRESSION
Withheld after visual QA
The research gate passed, but a later release check found reader-visible chart errors.
OPENROUTER / ONLINE
2026-08-31T011519Z-touchdown-regression-v1
- COMPLETED
- Aug 31, 2026, 1:29 AM UTC
- DATA THROUGH
- 2025
- SEED
- 4,444
0 recorded blockers
Configured GLM-5.3 roles
This run predates the call ledger. Exact attempts, tokens, latency, response IDs, and billed cost were not preserved and are not estimated.
ROLEMODEL / AUTHORITYDECISIONFINDINGS
POST-RUN NOTE
The PNG bars used current 2025 analysis values, but two visible chart titles retained hard-coded 2024 launch values.
Open note ↗RUN 04
TOUCHDOWN REGRESSION
Rejected by research gate
At least one required check or specialist decision failed. The candidate did not publish.
OPENROUTER / ONLINE
2026-08-31T010623Z-touchdown-regression-v1
- COMPLETED
- Aug 31, 2026, 1:09 AM UTC
- DATA THROUGH
- 2025
- SEED
- 4,444
3 recorded blockers
Configured GLM-5.3 roles
This run predates the call ledger. Exact attempts, tokens, latency, response IDs, and billed cost were not preserved and are not estimated.
ROLEMODEL / AUTHORITYDECISIONFINDINGS
POST-RUN NOTE
The candidate article was not published and no publication pull request was opened.
Open note ↗RUN 03
TOUCHDOWN REGRESSION
Rejected by research gate
At least one required check or specialist decision failed. The candidate did not publish.
OPENROUTER / ONLINE
2026-08-31-touchdown-regression-v1
- COMPLETED
- Aug 31, 2026, 12:46 AM UTC
- DATA THROUGH
- 2024
- SEED
- 4,444
13 recorded blockers
Configured GLM-5.3 roles
This run predates the call ledger. Exact attempts, tokens, latency, response IDs, and billed cost were not preserved and are not estimated.
ROLEMODEL / AUTHORITYDECISIONFINDINGS
Researcherz-ai/glm-5.3manifest configured route artifact metadata non authoritativeNOT INVOKED0 WARN · 0 BLOCK
Statisticianz-ai/glm-5.3manifest configured route artifact metadata non authoritativeBLOCK0 WARN · 0 BLOCKOPEN ↗
Adversarial Reviewerz-ai/glm-5.3manifest configured route artifact metadata non authoritativeREVISE4 WARN · 5 BLOCKOPEN ↗
Citation Checkerz-ai/glm-5.3manifest configured route artifact metadata non authoritativeBLOCK0 WARN · 8 BLOCKOPEN ↗
Editorz-ai/glm-5.3manifest configured route artifact metadata non authoritativeNOT INVOKED0 WARN · 0 BLOCK
IDENTITY NOTE
This early run allowed model-authored identity fields inside review artifacts. Those raw labels are not authoritative; the configured OpenRouter route in the immutable manifest is. Later transport code overwrites role, model, and time.
POST-RUN NOTE
The candidate article was not published and no publication pull request was opened.
Open note ↗RUN 02
ROOKIE WR HIT RATES
Published study
The research gate passed and this exact run backs a public study.
DETERMINISTIC / OFFLINE
2026-08-30-rookie-wr-hit-rates-v1
- COMPLETED
- Aug 30, 2026, 10:40 PM UTC
- DATA THROUGH
- 2024
- SEED
- 8,032
0 recorded blockers
Local deterministic checks
No language-model call occurred, so router usage and cost are not applicable.
RUN 01
TOUCHDOWN REGRESSION
Published study
The research gate passed and this exact run backs a public study.
DETERMINISTIC / OFFLINE
2026-08-30-touchdown-regression-v1
- COMPLETED
- Aug 30, 2026, 10:01 PM UTC
- DATA THROUGH
- 2024
- SEED
- 4,444
0 recorded blockers
Local deterministic checks
No language-model call occurred, so router usage and cost are not applicable.
BUILD-TIME CONTRACT
The dashboard cannot drift from the evidence.
The committed ledger is regenerated from every preserved run and compared byte-for-byte during the production build. Six adversarial mutations test the boundaries: false publication, invented cost, silent model drift, inflated summary totals, fake offline receipts, and publishing through a failed gate.