Building talomnia.com with Talomnia — how this site was built by the system it describes

The case in two minutes #

What the client asked

Build and launch talomnia.com — the public product site in two languages, showing not only the offer but its own execution trace: steps, roles, time, cost, artifacts, refusals and rework. The specification everything started from is a separate Workflow 0; the market study is a separate research Workflow. This one is the build only. The work is IN PROGRESS: the Evidence Gate is not passed and several checks are open.

Who executed

560 execution entries · 20 roles · 120 agent sessions

Capabilities

created: 1027 · reused: 206 · modified: 488

Checks

validated: 548 · rejected: 2 · reworks: 25

Result

The result is the site you are reading: talomnia.com in production, in two languages, carrying the public trace of its own construction. It was built by the operating model it describes, and it is the only case Talomnia has today — there has been no commercial order. The work is not finished: the Evidence Gate is not passed and several checks are open, and that is visible on the status page rather than hidden.

Acceptance status

in progress

Coverage and currency

560 execution entries · snapshot 85828ab43c78

The entry count and digest state which set the totals were computed from. The digest is taken over EVERY figure in this block rather than over selected columns, so it changes when membership changes or when any figure beside it changes. An attestation narrower than its apparent scope is worse than none, because it buys trust it has not earned.

Every figure above is derived from the execution ledger and the artifact matrix below. The conclusions text is written by hand and is pinned to the entry count: if the ledger moves past it, the projection refuses to load rather than publish a stale claim.

When the system stopped itself #

Below is a real lapse caught by the control mechanisms — not a staged demonstration. The ledger entry keeps its failed status: work executed around the contract is not admissible regardless of how sound its artifacts are — the value of this block is the honest refusal.

Failures and rework #

Derived directly from this case's execution ledger: no failed or reworked unit is hidden. Zero is a statement too.

Failed: 2 · Reworked: 25 · 560 of accounted units

Case economics #

Where the money and the time went #

Between one grand total and a ledger of hundreds of rows sits the breakdown: by role, by stage and — the one that matters — how much bought KNOWLEDGE versus how much was EXECUTION on knowledge that already existed. Every bar carries its own confidence marker: measured and reconstructed figures are never added into one number.

Confidence of these sums

Dashed bars are fully reconstructed after the fact; solid bars contain a measured part. The marker at the end of each bar names its measured share.

By role

By stage

A stage’s money comes from the ledger entries of the tasks that stage declares it covers, not from the caption under the timeline step. Stages sharing the same task set are merged into one bar: the ledger does not resolve finer than the task there, and splitting it further would be an invented allocation.

Producing knowledge versus executing on it

Classification rule: an execution unit counts as "producing knowledge" when at least one artifact it created or modified resolves to a Capability Atlas entity — by the same match the artifact matrix below uses. A unit that both produced knowledge and delivered work counts wholly as knowledge production. That is deliberately conservative: the rule can only inflate the knowledge side, so the published knowledge cost is an upper bound and the reuse claim is never flattered by it.

Created and reused

Knowledge elements engaged

474

Uses of them by execution units

4122

Average units per element

8.70×

Elements used by more than one unit

219

Knowledge-artifact touches (created / modified / reused)

82 / 40 / 18

Delivery-artifact touches (created / modified / reused)

947 / 460 / 200

Two different quantities, and not interchangeable: "touches" count what a unit did to an artifact, "uses" count how many units executed on that knowledge element. The reuse claim rests on the second.

Timeline caption reconciled against the ledger

Divergence between a step caption and the ledger (shown, not smoothed over):

• development: in the caption $133.40 / 8.5h, in the ledger $137.24 / 11.0h
• qa-and-customer-review: in the caption $70.85 / 7.7h, in the ledger $250.49 / 28.0h
• table-geometry-and-explanatory-affordance: in the caption $51.74 / 1.9h, in the ledger $62.69 / 2.7h
• the-approval-queue-becomes-one-reviewable-list: in the caption $0.00 / 10m, in the ledger $0.00 / 35m

Stage timeline #

1. 2026-08-19

   capability-discovery

   researcher — capability map — required roles, competencies, skills and blueprints inventoried against what already existed (99 rows, 95 reuse / 4 create) · public repository · 16m / $4.75

2. 2026-08-19

   knowledge-foundation-and-ledger

   developer — private knowledge repository, the Workflow Ledger itself (schema, CLI, mutation proofs) and the public sanitized trace repository with its CI gate · public repository · 60m / $15.43

3. 2026-08-19

   design-research

   design-lead — design research sweeps (developer-infrastructure SaaS, AI products, honesty-and-evidence surfaces) turned into design-system artifacts in the Capability Atlas · public repository · 35m / $9.75

4. 2026-08-19

   architecture-and-executable-knowledge

   developer — stack/database/analytics ADRs, then the architecture as executable artifacts — ontology schemas, the Professional Knowledge Graph, the five-stage resolver, per-element cost attribution · public repository · 3.8h / $55.00

5. 2026-08-19

   database-and-projection

   developer — schema and migrations plus the one-way Git-to-database projection with reproducibility and mutation proofs · public repository · 29m / $7.65

6. 2026-08-19

   development

   knowledge-curator — the site itself — every navigation section in two languages, Support Center integration, the Capability Atlas with search and relations, bilingual content, the plain-language layer, the evidence components · public repository · 8.5h / $133.40

7. 2026-08-19

   devops-and-deployment

   deployment-operator — edge, TLS and web-server configuration, staging and production, backups, deploy drift detection — and the disclosed out-of-contract deploy attempt that the gate caught · public repository · 3.0h / $38.76

8. 2026-08-19

   qa-and-customer-review

   reviewer — three customer acceptance reviews (R1, R2, R3 visual) taken as findings, dispositioned into executable subtasks, plus the orchestration record of the fleet that executed them · public repository · 7.7h / $70.85

9. 2026-08-19

   production-and-evidence-gate

   developer — re-execution of the remaining scope through resolved Knowledge Contracts, Evidence Gate checks, projection refreshes and the Launch Validation Report — NOT finished; this stage is why the workflow status is «in progress» · public repository · 2.9h / $12.99

10. 2026-08-20

    honest-accounting-under-review

    developer — The ledger learns to say "not measured": NULL instead of zero through the schema and the projection, the writer refusing to derive a total over an unmeasured component, and the subagent case that exposed it. Cost covers 1 of 7 entries — in the rest it is not measured, and they are not counted as zeros. · public repository · 1.1h / $1.19

11. 2026-08-20

    design-capability-research-and-binding

    researcher — Audit of the design capability and its contract binding, two research lanes over the sources the customer named, the design wall registered in the ontology, and the contract issued so the capability finally binds. · public repository · 2.1h / $30.25

12. 2026-08-20

    design-instruments-and-budget

    architect — Live confirmation of the reported defects, the measured baseline, the performance budget declared BEFORE any animation existed, and the composited-contrast instrument that measures colour as the page actually paints it. Cost covers 3 of 7 entries — in the rest it is not measured, and they are not counted as zeros. · public repository · 42m / $8.85

13. 2026-08-20

    motion-and-composition

    design-lead — The motion and visual-critique skills with their graph revisions, the ambient motion layer, and the first-screen composition fix — the critique registry learning to report an absence rather than only a defect. · public repository · 1.8h / $95.76

14. 2026-08-20

    interface-defects-and-evolution-notes

    developer — The R8 interface fixes with a check behind each invariant, and artifact evolution notes established as a standing policy rather than a one-off. · public repository · 1.1h / $2.80

15. 2026-08-20

    investor-surface-and-contract-integrity

    resolver-operator — The investor narrative and its honesty gates, the union rebase of the authority journal that chose no side, and the contract re-pinned after the graph snapshot moved. · public repository · 39m / $14.99

16. 2026-08-20

    deployment-and-runner-reliability

    devops — Deploys through the broker with a staging rehearsal, production releases with acceptance measured on the live surface, and the ambiguous runner label that had been quietly sending jobs to the wrong machine. · public repository · 2.0h / $14.62

17. 2026-08-20

    evidence-coverage-and-the-missing-accounting

    evidence-auditor — Publishing the work this page had been omitting, and the gate that stops it happening again: 71 entries and $522.97 of customer-review work brought onto the page, every derived figure recomputed, and a coverage assertion inside the projection transaction that refuses to publish a set which does not account for its own source. Includes this row — the work of publishing the missing accounting is itself billable and appears in the accounting it publishes. · public repository · 29m / $56.09

18. 2026-08-20

    long-form-typesetting-and-the-gates-that-never-read-the-text

    architect — Review R9: the artifact that says how prose becomes HTML, a page type for a document meant to be read, and the finding that two design gates had never looked at the text they were guarding. · public repository · 1.1h / $40.83

19. 2026-08-20

    table-geometry-and-explanatory-affordance

    developer — Review R10: the two table geometries that stop headers breaking mid-word, an affordance that explains what a column means, and the glossary vocabulary behind it — both suites written red on the pre-fix tree before anything was changed. · public repository · 1.9h / $51.74

20. 2026-08-20

    context-durability-of-the-lane-fleet

    architect — Review R11: the orchestrator ran lanes to a full context window. Prior art in the framework was read before authoring — a measurement instrument, the capability artifacts, the graph and resolution, and the fleet-durability work that followed from it. · public repository · 31m / $92.74

21. 2026-08-20

    production-release-and-surface-verification

    deployment-operator — A production release measured before it was touched and verified on the live surface afterwards: the served stylesheet fingerprint, the release symlink SOURCE_COMMIT, and the edge observed rather than assumed. · public repository · 15m / $127.41

22. 2026-08-20

    correction-cycle-and-republish

    evidence-auditor — The customer returned R13 on process, not substance: two right fixes to one CI file had collided, and the accepted work needed rebasing onto them. Resolving it surfaced three more tasks the ledger held and no page published — which is the third time in one session this gate caught the defect it was built for. Cost is not stated: the operator's review time is real and was not metered, so the total is null rather than a number that would count a person's work as zero. · public repository · not measured — this step predates the Workflow Ledger

23. 2026-08-20

    edge-cache-purge-and-fleet-scope

    devops — Review R12: the deploy never purged the Cloudflare edge, so a corrected origin sat behind a stale edge copy. The gap measured first, the purge step argued rather than assumed, a cache policy written down, and the scope of the fleet it applies to reported. · public repository · 1.7h / $34.95

24. 2026-08-20

    composition-findings-closed-worst-first

    developer — The 43 critique findings measured on production, reproduced exactly before any change and closed worst first, including the WCAG 2.2 AA target-size work. Cost is not stated: no entry in this stage carries a measured total. · public repository · not measured — this step predates the Workflow Ledger

25. 2026-08-20

    acceptance-review-and-mutation-arms-rebuilt

    customer-acceptance — An acceptance review that refused the merge on a red CI and separated the two reds rather than averaging them, then rebuilt the mutation arms and added positive controls so a proof could not pass by tripping the wrong check. Cost covers 4 of 5 entries — in the rest it is not measured, and they are not counted as zeros. · public repository · 12m / $17.54

26. 2026-08-20

    the-webfont-that-never-loaded

    developer — The 2px RU overflow that had been blocking main was not a flake in the usual sense: every @font-face rule declared font-family: var(--font-display), which is invalid in a descriptor, so all six faces were dropped and text width had been a property of the reader own machine since TALO-0079. · public repository · 1.0h / $24.50

27. 2026-08-20

    unversioned-stylesheet-and-the-asset-version-gate

    sre — The root language chooser wrote its own document shell instead of going through the shared page template, which is how it kept linking an unversioned stylesheet; the asset-version gate that now refuses that, and the residual declared rather than left implicit. · public repository · 38m / $73.78

28. 2026-08-20

    production-deploy-and-font-acceptance

    deployment-operator — The pre-deploy surface captured before the switch because it could not be recreated afterwards — document.fonts.size was 0 on both language trees — then the deploy refused once, landed, and was accepted by measurement on the live surface. · public repository · 16m / $76.71

29. 2026-08-20

    rebase-onto-the-font-fix-and-re-measure

    evidence-auditor — Rebasing the edge-purge work onto the tree carrying the font fix and re-proving the harness there rather than trusting a verdict measured on the broken stylesheet, plus a re-measurement of the stale edge copy. · public repository · 9m / $9.74

30. 2026-08-20

    rebase-remeasure-and-republish

    evidence-auditor — Re-proving the coverage figure against the ledger as it stands rather than as it stood: the published red-before number had itself gone stale, understating the gap threefold. Also the cycle in which my own landing script appended 14 duplicate rows — caught by the ledger validator and by the projection loader independently, before anything reached a page. Cost is not stated: the operator's review time is real and was not metered. · public repository · not measured — this step predates the Workflow Ledger

31. 2026-08-20

    edge-purge-rebased-onto-the-font-fix

    deployment-operator — The edge-purge work rebased onto the tree that carries the font fix, because its merge base predated it and a verdict measured on the broken stylesheet proves nothing; then deployed, with sanitization wired onto every PR and an append-only correction to the ledger. · public repository · 16m / $65.60

32. 2026-08-20

    landing-page-proves-the-deploy

    evidence-auditor — The briefs premise did not survive contact — production was already at the expected commit, established on the authority that reads the prod host rather than on the brief. The landing page was then measured directly, an acceptance instrument built, and the residuals exposure window turned into a number instead of a worry. · public repository · 18m / $10.98

33. 2026-08-20

    discovery-surface-and-the-authoring-hole

    architect — Review R14: re-measured the reviews three unbindable instances instead of accepting them, and two of the three were already reachable. The discovery surface was then authored and implemented, and the hole that let an artifact be written without becoming bindable was closed behind it. · public repository · 45m / $32.33

34. 2026-08-20

    instruments-declared-and-a-real-blind-spot

    reviewer — Every instrument made to declare how it can be blinded — and the declaration was not paperwork: forcing it exposed a gate that, pointed at a page which does not exist, exited zero and printed its full success message. The blind spot was behind the missing declaration, which is the argument for the rule. · public repository · 17m / $51.00

35. 2026-08-20

    final-rebase-and-merge-readiness

    evidence-auditor — The last re-measurement before merge, computed against the ledger as it stood at merge time rather than as it stood when the work began — four more tasks had landed and appeared on no page. Also the cycle in which the portable finding got its evidence: the same coverage failure has now been observed four times in unrelated domains, the sharpest being a syntactic position rather than a list of files. Cost is not stated: the operator's review time is real and was not metered. · public repository · not measured — this step predates the Workflow Ledger

36. 2026-08-20

    discovery-surface-live-and-two-gates-that-refused-first

    deployment-operator — The discovery surface reached production, and the release was refused twice on the way by gates from its own lane before it was allowed through — a deploy that has never been refused has not been tested. · public repository · 16m / $115.76

37. 2026-08-20

    figures-derived-rather-than-restated

    evidence-auditor — The freshness gate bound one field while every other published figure drifted beside it — and the drift had begun before the task set moved, so it was never caused by new work landing but by figures written once and re-read forever. The second, hand-typed copy of the derivation is gone rather than gated, and the digest is now taken over the whole block it is printed beside instead of three chosen columns. Cost is not stated: the operator's measurement of the deployed page was real and unmetered. · public repository · not measured — this step predates the Workflow Ledger

38. 2026-08-20

    totals-that-say-not-measured

    developer — The denormalized workflow totals learn what the ledger rows beside them learned earlier: a sum over rows that nobody measured is an absence, not a zero. The specification workflow had been publishing «$0.00 · 0s» for seven entries that predate the Workflow Ledger. Removing false zeros, not real ones — the measured workflows keep their numbers. Cost is not stated for this row, which is the same rule applied to itself. · public repository · not measured — this step predates the Workflow Ledger

39. 2026-08-20

    launch-readiness-measured-and-no-verdict-declared

    evidence-auditor — The whole specification read and every clause measured against the live surface, the Evidence Gate, the ledger and the graph — and the result published as counts rather than as a verdict, because a readiness number nobody can reproduce is the kind of figure this workflow exists to refuse. · public repository · 1.3h / $26.25

40. 2026-08-20

    a-revision-that-no-longer-describes-its-file

    evidence-auditor — The bindability check asked three questions about a revision and never whether it still described the file it names — because it was built by walking the store, and that is the one question the store cannot answer alone. Four artifacts found across all three roots, all correct when cut and edited afterwards; the trap was that two digest normalisations are both legitimate, so a naive detector would have called twenty-two healthy artifacts stale. Cost is not stated: the operator's brief and the disclosure this consumed were real and unmetered. · public repository · not measured — this step predates the Workflow Ledger

41. 2026-08-20

    the-approval-queue-becomes-one-reviewable-list

    evidence-auditor — Thirty-two pending approvals turned into one signature the operator can review, with each artifact identifiable without opening a file — and the finding that the queue never grew the way it appeared to: sixteen and thirty-three were counts of different populations, and half the queue is superseded revisions no closure can ever select. Nothing was approved here; merging is the authority act. Cost is not stated: the operator's review time is real and unmetered. · public repository · not measured — this step predates the Workflow Ledger