Skip to content

Spec 001 T009: the analysis service's answers, measured (replaces #113) - #114

Closed
adamjohnwright wants to merge 3 commits into
mainfrom
t009/review-2
Closed

adamjohnwright wants to merge 3 commits into
mainfrom
t009/review-2

Conversation

@adamjohnwright

Copy link
Copy Markdown
Contributor

Spec 001, task T009 (research R6). This replaces #113, which replaced #112; each was closed after its review. The first commit is #112's, the second holds #112's review fixes, and the third holds #113's.

Measured against a non-production instance of the analysis service on release 98, never production. Nothing in its storage was touched.

Results:

  • Token answers: with no stored result, a token dated within 7 whole days (or in the future) is 410, and an older one is 404. A malformed token, or Base64 of _ alone, gets the catch-all 500. A / gets a 400 page or reaches another route.

  • Expiry: the service decides it. How long a result is kept depends on the results folder's size and on deletions made outside the service's code, not on the token's age. pdf-export checks only the token's form, which must decode to {14 digits}_{digits}, and the --retention setting is gone.

  • Counts:

    • the identifiers found come from the found identifiers' CSV;
    • the total adds the not-found count;
    • the pathways hit are pathwaysFound with includeDisease=true.

    These match today's book under FR-006's defaults, except possibly for pathways without a p-value. The comparison (FR-012) records any such difference.

From #113's review:

  • High: the form rule accepted Xw==, which makes the service fail.
  • Medium: R6 misdescribed how today's book counts.
  • Medium: the cleanup was presented as the only way results are deleted.
  • Medium: spec 002 still said 410 means a release deleted the result. That is now corrected. Whether spec 002's report should check the token's form is left to its tasks.
  • Lows: the includeDisease echo, empty or non-ASCII tokens, the token reuse on resubmission, and the quickstart wording.

This PR also records #111's three low review findings under T008. It changes documents only.

🤖 Generated with Claude Code

adamjohnwright and others added 3 commits November 3, 2026 12:35
Against a non-production instance on release 98, with tokens the
service reads by its own rules and one small analysis; nothing in its
storage was touched.

- The service reads a stored result first and judges a token's date
  (7 days, a constant in analysis-core) only when there is none: 410
  within it, 404 after. A result kept longer is still answered, so
  pdf-export checks only the token's form, and the --retention setting
  is gone.
- A malformed token is the service's catch-all 500, in its JSON form.
- found/interactors with none is a 404, with or without interactors.
- Neither the summary nor result.json carries the found count; the
  found identifiers' CSV does, from the same set as today's book.

Also records #111's three low review findings under T008.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- pathwaysFound follows the analysis's includeDisease (229 with, 191
  without here); it is asked for with includeDisease=true, as today's
  book counts the unfiltered list.
- How long a result is kept is the service's own cleanup (by the
  results folder's size and last access), recorded as such; the 7 days
  only tell 410 from 404.
- The malformed-token rule is now the form the service issues: wrong
  padding (QQ=) and '/' (a 400 page, or another route) were missed.
- R6's older text, which contradicted the measurement, is corrected;
  the boundary (whole days, future-dated is 410), the time zone, the
  CSV's header and the help-desk placeholder are stated.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…'s counts

- Base64 of '_' alone passed the form rule and the service fails on it
  (500); the rule now also requires the decoded {14 digits}_{digits}.
- Today's book counts after its own species and resource filter, not
  the stored list: recorded, with the edge for a pathway without a
  p-value, left to the comparison (FR-012).
- Results are also removed outside the service's cleanup, access times
  depend on the mount, and a resubmission can bring a token back.
- Spec 002: "410 for one a release deleted" corrected; whether its
  report checks the token's form is left to its tasks.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@adamjohnwright

Copy link
Copy Markdown
Contributor Author

Closed after its review: the form rule needs a bound on the counter (a 7,000-digit one reaches Tomcat's HTML 400), and R6's account of today's book's counts (species, p-value) disagrees with the code and with spec 002. Replaced by a new PR.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant