Agent report card
Measurement Summary and Ledger Record
27 agents · passed 26 · failed 0 · this run (does not run itself) 1 · total checks 456 · measurement date 2026-10-24 19:01 (+03:00)
Prepared on: 2026-10-24 19:01 (+03:00) · Measurement date: 2026-10-24 19:01 (+03:00) · Ledger: karne_kutugu.jsonl (this page's preparation record: row 2)
Agent Self-Test and Ledger Table
| Agent | Task | Last selftest | Result | Status | Last run | Ledger rows (cumulative) | P0 (cumulative) |
|---|---|---|---|---|---|---|---|
| Structure Reader | Measures HTML language, title, meta description, canonical, and schema structure. | 2026-10-24 19:01 | 8/8 | passed | 2026-10-24 16:09 | 605 | 29 |
| Link Reader | Measures internal link graph, broken links, orphan and dead-end pages. | 2026-10-24 19:01 | 14/14 | passed | 2026-10-24 17:03 | 434 | 0 |
| Source Reader | Measures source attribution for numeric and regulatory claims in text. | 2026-10-24 19:01 | 20/20 | passed | 2026-10-24 17:03 | 3741 | 0 |
| Language Reader | Measures sentence length, term explanations, and language parity. | 2026-10-24 19:01 | 19/19 | passed | 2026-10-24 17:03 | 2471 | 0 |
| Speed Reader | Measures page byte weight, resource request count, and image formats. | 2026-10-24 19:01 | 38/38 | passed | 2026-10-24 16:09 | 276 | 0 |
| Boundary Reader | Measures health promotion regulatory boundaries and disclosure statements. | 2026-10-24 19:01 | 13/13 | passed | 2026-10-24 17:03 | 1268 | 146 |
| Bot Reader | Measures robots rules, sitemap declarations, and crawler accessibility. | 2026-10-24 19:01 | 20/20 | passed | 2026-10-24 16:09 | 81 | 1 |
| Seven Readers Link Engine | Combines seven reader findings with cross-rules and measures correlations. | 2026-10-24 19:01 | 18/18 | passed | 2026-10-24 17:03 | 133 | 5 |
| Reader Update Trigger | Detects pages with changed fingerprints and runs relevant readers. | 2026-10-24 19:01 | 24/24 | passed | — | — | — |
| Maintenance Freshness Sweeper | Measures sitemap, schema dates, and content freshness consistency. | 2026-10-24 19:01 | 8/8 | passed | 2026-10-24 17:30 | 511 | 0 |
| Maintenance Link Sweeper | Verifies broken internal and external links and queues actions. | 2026-10-24 19:01 | 12/12 | passed | 2026-10-24 05:33 | 54 | 0 |
| Maintenance Language Pair Sweeper | Measures missing language counterparts and hreflang links across multilingual pages. | 2026-10-24 19:01 | 10/10 | passed | 2026-10-22 12:27 | 51 | 0 |
| Maintenance Report Card Publisher | Collects dated agent self-test results and ledger counts. | 2026-10-24 19:01 | — | this run (does not run itself) | — | — | — |
| Shell API Self-Test | Tests contract compliance of shell API endpoints. | 2026-10-24 19:01 | — | passed | — | — | — |
| Analysis Chain Self-Test | Tests integrity of end-to-end analysis chain steps. | 2026-10-24 19:01 | 8/8 | passed | — | — | — |
| Analysis Core Engine | Tests quote and contact triggered analysis core. | 2026-10-24 19:01 | 19/19 | passed | — | — | — |
| Reaper Drill Self-Test | Tests pre-publication verification and security gates. | 2026-10-24 19:01 | 6/6 | passed | — | — | — |
| Maintenance Orchestra | Tests sequential execution and lock discipline of maintenance members. | 2026-10-24 19:01 | 16/16 | passed | — | — | — |
| Maintenance Executor | Tests safe application of approved actions in the maintenance queue. | 2026-10-24 19:01 | 13/13 | passed | — | — | — |
| Maintenance Decision Engine | Tests prioritization and deduplication rules for maintenance findings. | 2026-10-24 19:01 | 12/12 | passed | — | — | — |
| Reaper-Shield Deception and Security Guardian | Tests trap endpoints, fake-bot filtering, and CGNAT TTL blocking across three sites. | 2026-10-24 19:01 | 8/8 | passed | — | — | — |
| Live Site Health Guardian | Tests live HTTP, security headers, sitemap, and content integrity across three sites. | 2026-10-24 19:01 | 81/81 | passed | — | — | — |
| Technical SEO and GEO Gate Guardian | Tests ten-gate technical SEO, structured data, AI bot access, and author seal. | 2026-10-24 19:01 | 9/9 | passed | — | — | — |
| Atlas Mechanical Context Guardian | Tests the four-kilobyte mechanical context packet and handover schema gate. | 2026-10-24 19:01 | 7/7 | passed | — | — | — |
| Death Layer-1 Leak Gate | Tests pre-publication secret key, identity, and personal data leak rule gate. | 2026-10-24 19:01 | 37/37 | passed | — | — | — |
| Death Name and Filename Leak Dictionary | Tests the dictionary module catching personal name leaks in filenames and text. | 2026-10-24 19:01 | 16/16 | passed | — | — | — |
| Autonomy Seven-Day Scorecard Meter | Tests run and verification counts of agents over the last seven calendar days. | 2026-10-24 19:01 | 20/20 | passed | — | — | — |
Site Breakdown (Cumulative Ledger)
| Site | Ledger rows (cumulative) | P0 (cumulative) |
|---|---|---|
| atlasmaster.app | 2934 | 78 |
| sister site 1 | 3088 | 173 |
| sister site 2 | 1132 | 6 |
| system (site independent) | 0 | 0 |
| other site | 3467 | 87 |
| other site | 2 | 0 |
| All sites + system total | 10623 | 344 |
Reaper evolution scorecard (K6)
| Stage | Date | Trap catch | False alarm | Gold score |
|---|---|---|---|---|
| ADIM0_TABAN | 2026-10-22 | %100 | %0 | %83.8 |
| SMOKE | 2026-10-22 | %90.3 | %0 | %87.5 |
| TARAMA (K01) | 2026-10-23 | %98.3 | %0 | %94.7 |
| TARAMA (K02) | 2026-10-23 | %99.1 | %0 | %93.4 |
| TARAMA (K03) | 2026-10-23 | %98.3 | %0 | %96 |
| TARAMA (K04) | 2026-10-23 | %100 | %0 | %94.4 |
| TARAMA (K05) | 2026-10-23 | %99.1 | %0 | %94.7 |
| TARAMA (K06) | 2026-10-23 | %97.8 | %0 | %96 |
| TARAMA (K07) | 2026-10-23 | %100 | %0 | %96 |
| TARAMA (K08) | 2026-10-23 | %97.8 | %0 | %95.3 |
| TARAMA (K09) | 2026-10-23 | %98.7 | %0 | %94 |
| TARAMA (K10) | 2026-10-23 | %96.9 | %0 | %94.4 |
| TARAMA (K11) | 2026-10-23 | %99.1 | %0 | %93 |
| TARAMA (K12) | 2026-10-23 | %99.1 | %0 | %95.7 |
| COLAB_PII (P1) | 2026-10-23 | %98.7 | %0 | %95 |
| COLAB_PII (P2) | 2026-10-23 | %97.4 | %0 | %93.7 |
| COLAB_PII (P3) | 2026-10-23 | %97.8 | %0 | %91 |
| DEATH_ALAN_VAL (ALAN1) | 2026-10-23 | — | — | — |
| DEATH_ALAN_MUHUR (VETO1) | 2026-10-23 | %96.7 | %2.5 | %98.6 |
| DEATH_ALAN_AD_AYAR_DONDU (AD1) | 2026-10-23 | %100 | %1.4 | %100 |
| DEATH_ALAN_AD_MUHUR_SET_URETILDI (AD1) | 2026-10-23 | — | — | — |
| DEATH_ALAN_MUHUR (AD1) | 2026-10-23 | %85 | %51 | %60.7 |
ADIM0_TABAN is the baseline of today's auditor: a sealed 20-question trap set, two runs; the answer key was not shown to the model. TARAMA (K01–K12), COLAB_PII (P1–P3), and DEATH_ALAN (VETO1, AD1) rows record Death and Reaper leak, trap, and agreement evolution measurements. An evolution verdict needs above-threshold results in two consecutive stages; one row is not evolution.
LoRA fine-tuned agent fleet (36 adapters)
| Task adapter | qwen7b (loss / status) | gemma9b (loss / status) | qwen14b (loss / status) |
|---|---|---|---|
| 1. Tool-Calling Specialist | loss 0.6124 · 154 MB | loss 0.6735 · 206 MB | loss 0.8486 · 263 MB |
| 2. Context and Handover Reader | loss 1.3295 · 154 MB | loss 0.3588 · 206 MB | loss 0.4397 · 263 MB |
| 3. Screen and Session Handover | loss 1.5063 · 154 MB | loss 0.4843 · 206 MB | loss 0.6946 · 263 MB |
| 4. Death Site and Leak Auditor (K2) | initialized (A100 queue) · 154 MB | loss 0.5221 · 206 MB | loss 0.6483 · 263 MB |
| 5. Enki Note and Self-Iteration Voice | initialized (A100 queue) · 154 MB | loss 0.4330 · 206 MB | loss 0.5524 · 263 MB |
| 6. Enlil Schema and Audit Guardian | initialized (A100 queue) · 154 MB | loss 0.4885 · 206 MB | loss 0.5615 · 263 MB |
| 7. Architectural Code Synthesizer | initialized (A100 queue) · 154 MB | loss 0.4057 · 206 MB | loss 0.4846 · 263 MB |
| 8. Visual and Video Production Auditor | initialized (A100 queue) · 154 MB | loss 0.4544 · 206 MB | loss 0.5906 · 263 MB |
| 9. Site SEO and GEO Reader | initialized (A100 queue) · 154 MB | loss 0.4617 · 206 MB | loss 0.5855 · 263 MB |
| 10. QAJAN Hypothesis and North Watchdog | initialized (A100 queue) · 154 MB | loss 0.4785 · 206 MB | loss 0.6076 · 263 MB |
| 11. Clinical Case Brief Builder | initialized (A100 queue) · 154 MB | loss 0.5032 · 206 MB | loss 0.6365 · 263 MB |
| 12. Shared Terminal Classifier | initialized (A100 queue) · 154 MB | loss 0.5452 · 206 MB | loss 0.6641 · 263 MB |
Across three base models (qwen7b, gemma9b, qwen14b), 12 task adapters (36 adapters total) are verified on disk. Training losses are measured for gemma9b (12/12) and qwen14b (12/12); on qwen7b the first 3 adapters are trained and the remaining 9 are queued for the A100 pass.
Method
Each agent's self-test command is executed in a separate subprocess and the final verification token and gate counts are recorded. P0 is the highest priority label used by the readers and maintenance members; the P0 figure here counts ledger rows accumulated over time, including later-resolved findings, and is not a count of open violations. The preparation date is when the page was generated; the moment it went live is recorded separately in the ledger as a 'yayinlandi' (published) row once the live page fingerprint is verified. Because the reader and maintenance ledgers are append-only and idempotent, the ledger row and P0 columns represent cumulative totals rather than point-in-time open counts. Agents that do not write to a ledger or tokens that do not carry gate counts are marked with '—' without fabricating unmeasured numbers.