Working through the 17 technology link-shares that accumulated on this tracker between 2021 and 2026 — the set an earlier audit tiered as "convert to Discussions / parking". Six are done. This issue records where to resume and how.
Where to resume
Next: #78 and #100, together. They are the same LMS thread six months apart, and verifying them separately would duplicate the work. Neither produces a technology-register entry — they are programme questions, not tool evaluations.
Done so far
| Issue |
Outcome |
| #52 Singularity containers |
Closed as answered — containers were adopted, in Docker/OCI form via #69, now QuantEcon/actions/containers |
| #53 asciinema |
Closed unowned. Created the technology register |
| #54 Plotly Dash |
Closed as declined — no Python process on GitHub Pages |
| #59 Mito |
Closed as declined — unreachable on every surface we publish through |
| #64 Google Workspace workshop notes |
Closed as superseded, nothing retained |
| #65 Hivebrite |
Closed as declined — we ran this experiment and retired it in March 2025 |
Remaining
Not technology options — handle as programme questions, no register entry: #78 and #100 (LMS), #114 (outreach link-share).
Technology options — each produces a register entry: #90 manim, #136 colour tools, #163 glyphx, #169 Humanitix, #230 marimo, #281 CODON, #294 OCR tooling, #295 AI swarms.
#163 carries a live unanswered question — @jstac asked directly whether we should switch to glyphx and it was never answered. An entry answers it either way.
Then #117, closing out the Discussions conversion experiment.
The method, and why it is not being batched
Each issue gets independent investigation passes followed by adversarial verification tasked with refuting the proposed disposition, then a spot-check of the load-bearing claims before anything is written.
Six for six, the close action has survived and the stated reason has been refuted. That is not a coincidence: the audit reasoned from what each issue said, and these are one-line link-shares, so its reasoning had nothing to check itself against. Verification has so far prevented publishing that Singularity is dead (it is not), that lockfiles replaced containers (no lockfile exists anywhere in the org), that our rendering model excludes interactivity (the style-guide page serves a live interactive plotly figure from GitHub Pages), that Mito charges to disable telemetry (a free opt-out exists), and that Hivebrite costs $799/month across three tiers that do not exist.
It has also surfaced live defects that had nothing to do with the technology under evaluation — see below. That is the main argument for continuing one at a time.
Where the outcomes go
Technology evaluations are recorded in a new register at docs/tech in QuantEcon/lectures (QuantEcon/lectures#6). Each entry records what the option is including current project health, which named files it would apply to, and the decision with a reason and a date. declined is a first-class terminal state — an option we rejected with the reason written down is more useful than one nobody recorded.
The register is deliberately not in a project-* repo: QEP-3 defines those as goal-scoped and private, so a standing reference would die with the project.
Discussions were considered as the destination and rejected on evidence — the three link-share discussion buckets hold one, zero and zero comments between them.
Filed from these six so far
A recurring pattern worth noting: the residue is consistently older than the issue being closed. lecture-datascience.myst#214 has been open 3.7 years; workspace.md, the lecture that most needed #53's idea, was created 23 months after #53 was filed; and #65's answer was a closure announcement written by its own author three years later.
Working through the 17 technology link-shares that accumulated on this tracker between 2021 and 2026 — the set an earlier audit tiered as "convert to Discussions / parking". Six are done. This issue records where to resume and how.
Where to resume
Next: #78 and #100, together. They are the same LMS thread six months apart, and verifying them separately would duplicate the work. Neither produces a technology-register entry — they are programme questions, not tool evaluations.
Done so far
QuantEcon/actions/containersRemaining
Not technology options — handle as programme questions, no register entry: #78 and #100 (LMS), #114 (outreach link-share).
Technology options — each produces a register entry: #90 manim, #136 colour tools, #163 glyphx, #169 Humanitix, #230 marimo, #281 CODON, #294 OCR tooling, #295 AI swarms.
#163 carries a live unanswered question — @jstac asked directly whether we should switch to glyphx and it was never answered. An entry answers it either way.
Then #117, closing out the Discussions conversion experiment.
The method, and why it is not being batched
Each issue gets independent investigation passes followed by adversarial verification tasked with refuting the proposed disposition, then a spot-check of the load-bearing claims before anything is written.
Six for six, the close action has survived and the stated reason has been refuted. That is not a coincidence: the audit reasoned from what each issue said, and these are one-line link-shares, so its reasoning had nothing to check itself against. Verification has so far prevented publishing that Singularity is dead (it is not), that lockfiles replaced containers (no lockfile exists anywhere in the org), that our rendering model excludes interactivity (the style-guide page serves a live interactive plotly figure from GitHub Pages), that Mito charges to disable telemetry (a free opt-out exists), and that Hivebrite costs $799/month across three tiers that do not exist.
It has also surfaced live defects that had nothing to do with the technology under evaluation — see below. That is the main argument for continuing one at a time.
Where the outcomes go
Technology evaluations are recorded in a new register at
docs/techin QuantEcon/lectures (QuantEcon/lectures#6). Each entry records what the option is including current project health, which named files it would apply to, and the decision with a reason and a date.declinedis a first-class terminal state — an option we rejected with the reason written down is more useful than one nobody recorded.The register is deliberately not in a
project-*repo: QEP-3 defines those as goal-scoped and private, so a standing reference would die with the project.Discussions were considered as the destination and rejected on evidence — the three link-share discussion buckets hold one, zero and zero comments between them.
Filed from these six so far
itables, closing a table defect open since 2022A recurring pattern worth noting: the residue is consistently older than the issue being closed.
lecture-datascience.myst#214has been open 3.7 years;workspace.md, the lecture that most needed #53's idea, was created 23 months after #53 was filed; and #65's answer was a closure announcement written by its own author three years later.