Past 48 hours change report - what changed and why

Living document. Window audited: 2026-06-10 10:14:50 UTC -> 2026-06-12 10:14:50 UTC. This report covers the committed work I made across the main donto substrate repository and the public donto-web site during that window. For the full reports index, see Reports.

The short version: the last 48 hours turned donto from a strong claim-store kernel into a broader, documented substrate prototype. The work landed in two places:

• donto: 42 commits, 211 files changed, 29,776 insertions, 1,014 deletions. This completed the execution-plan waves from W0-2 through H8, then published the execution/future reports in the docs app.
• donto-web: 30 commits, 110 files changed, 24,144 insertions, 396 deletions. This built the public donto.org documentation, benchmark boards, comparison pages, reports system, LoCoMo analysis reports, and the corrected /reports index/linking behavior.

TL;DR

The main reason for the work was to move donto from "we can store many claims" toward "we can explain, constrain, export, audit, compare, and govern memory under pressure."

1. Substrate execution work in donto

The execution plan work landed as a sequence of small commits, each with a matching migration, code path, test, or documentation update where appropriate. The pattern was deliberate: do the smallest durable change, prove it with a check, then move to the next layer.

W0: Make the base state auditable

Why this wave came first: if the system cannot explain where claims came from, how blobs were stored, and what live settings were active, later "intelligence" features become unverifiable.

W1: Add lenses, standing, release shape, and review paths

Reasoning: this is where donto starts to become a substrate rather than a store. Claims need views, standing, review pressure, and exports, otherwise downstream agents only get undifferentiated facts.

W2: Add feedback loops and loss accounting

Reasoning: loss reports and evidence suggestions are the accountability layer. They tell an agent not just what donto knows, but where donto is weak, lossy, or ready for another pass.

W3: Expand query, contradiction, interop, and TUI surfaces

Reasoning: this wave made donto more inspectable and interoperable. It also tightened the hard parts: identity cannot-links, predicate closure, paraconsistency, lens algebra, and API conformance.

H1-H8: Horizon features

Reasoning: the H-series is where the system starts to resemble AI-memory infrastructure: agent discovery, reputation, policy ordering, analogy, scientific calibration, and tournament-style hypothesis pressure.

Documentation publication in donto

2. Public site work in donto-web

The web work did two jobs: expose donto clearly to humans, and publish honest benchmark/report pages that keep the substrate claims grounded in measured behavior.

Site foundation, docs, and product surface

Reasoning: the site needed more than a landing page. It needed docs, comparison, benchmarks, reports, admin surfaces, and visual conventions so future changes have a coherent public home.

Extraction and LoCoMo benchmark reporting

Reasoning: these reports matter because they stop the system from overclaiming. donto can hold rich claims, but the benchmark showed the reader needs the right return shape, temporal tags, and better recall/reranking before the substrate wins cleanly.

Reports, benchmark interpretation, and the corrected /reports index

Reasoning: /reports must be an index, not a single report. The correction split the future substrate report onto /reports/future-substrate, kept /reports as the report directory, and then improved the future report formatting to match the other public notes.

3. Why these changes fit together

The work was not random feature accumulation. It follows one spine:

source bytes
  -> extracted claims with run provenance
  -> evidence links, valid time, policy, identity, standing
  -> contradiction-aware lenses and query surfaces
  -> loss reports, releases, and conformance checks
  -> public reports that say what worked, failed, and remains missing

The substrate side makes the memory more accountable. The public-site side makes the accountability visible.

The most important design decisions were:

4. What changed for users and agents

5. What the future should look like from here

The next phase should be less about adding isolated powers and more about making the powers boring, versioned, and externally trustworthy.

6. Caveats and boundaries

• This report covers committed work visible in local git history during the audited 48-hour window.
• It does not claim unrelated dirty local benchmark files are finished. At the time this report was written, there were uncommitted local changes under the benchmark run surface; those were intentionally not included in this report commit.
• Some public benchmark conclusions are explicitly provisional. The LoCoMo reports already say where the current cheap regime tops out and where reranking/reader strength remain blocked or unproven.
• The report index mistake was corrected: /reports is the index, and individual reports live under their own slugs.

7. Commit ledger

donto ledger

donto-web ledger

Bottom line

In the last 48 hours, donto gained a much stronger governed-memory spine: safety guards, provenance, object storage, standing, lenses, review/argument machinery, loss reporting, rule agendas, loop routes, query/interop expansion, conformance checks, agent discovery, reputation overlays, hypothesis tournaments, export closures, and policy ordering.

donto.org gained the public scaffolding needed to make that work inspectable: docs, comparison pages, benchmark boards, a reports index, detailed LoCoMo/Hyades reports, a corrected future substrate report route, and now this change report.

The future work is to make these capabilities stable enough that external agents and independent implementations can rely on them without needing private context from the current codebase.

See also: Future Substrate Report · The shape of donto's return · Single-shot vs agentic memory · LoCoMo Config C.